Leopold Aschenbrenner — 2027 AGI, China/US super-intelligence race, & the return of history

Chatted with my friend Leopold Aschenbrenner https://www.forourposterity.com/ on the trillion dollar nationalized cluster, CCP espionage at AI labs, how unhobblings and scaling can lead to 2027 AGI, dangers of outsourcing clusters to Middle East, leaving OpenAI, and situational awareness. Watch on YouTube https://youtu.be/zdbVtZIn9IM. Listen on Apple Podcasts https://podcasts.apple.com/us/podcast/leopold-aschenbrenner-china-us-super-intelligence-race/id1516093381?i=1000657821539, Spotify https://open.spotify.com/episode/5NQFPblNw8ewxKolIDpiYN?si=6NaTHAugT2SxZrspW3lziw, or any other podcast platform. Read the full transcript here https://www.dwarkeshpatel.com/i/145136502/transcript. Follow me on Twitter https://twitter.com/dwarkesh_sp for updates on future episodes. Follow Leopold on Twitter https://x.com/leopoldasch. Timestamps (00:00:00) – The trillion-dollar cluster and unhobbling (00:20:31) – AI 2028: The return of history (00:40:26) – Espionage & American AI superiority (01:08:20) – Geopolitical implications of AI (01:31:23) – State-led vs. private-led AI (02:12:23) – Becoming Valedictorian of Columbia at 19 (02:30:35) – What happened at OpenAI (02:45:11) – Accelerating AI research progress (03:25:58) – Alignment (03:41:26) – On Germany, and understanding foreign perspectives (03:57:04) – Dwarkesh’s immigration story and path to the podcast (04:07:58) – Launching an AGI hedge fund (04:19:14) – Lessons from WWII (04:29:08) – Coda: Frederick the Great This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.dwarkesh.com https://www.dwarkesh.com?utm_medium=podcast&utm_campaign=CTA_1

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Dwarkesh Patel?

  Okay, today I'm chatting with my friend Leopold Ochenbrenner. He grew up in Germany, graduated valedictorian of Columbia when he was 19. And then he had a very interesting gap year, which we'll talk about. And then he was on the OpenAI Superalliament team, may it rest in peace. And now he, with some anchor investments from Patrick and John Collison and Daniel Gross and Nat Friedman, is launching an investment firm. So, Leopold, I know you're off to a slow start, but life is long and I wouldn't worry about it too much. You'll make up for it in due time. But thanks for coming on the podcast.

• 0:37Speaker 01

  Thank you. You know, I first discovered your podcast when your best episode had, you know, like a couple hundred views. And so it's just been, it's been amazing to follow your trajectory and it's a delight to be on.

• 0:47Dwarkesh Patel?

  Yeah, yeah. Well, I think in the show or in Trenton episode, I mentioned that a lot of the things I've learned about AI, I have learned from talking with them. And the third part of this triumvirate, probably the most significant in terms of the things that I've learned about AI, has been you. We'll all have this stuff on the record now.

• 1:04Speaker 01

  Great.

• 1:05Dwarkesh Patel?

  Okay, first thing I had to get on record. Tell me about the trillion dollar cluster. But by the way, I should mention, so the context of this podcast is today, there's, you're releasing a series called Situational Awareness. We're going to get into it. First question about that is tell me about the trillion dollar cluster.

• 1:20Speaker 01

  Yeah, so, you know, unlike basically most things that have come out of Silicon Valley recently, you know, AI is kind of this industrial process. You know, the next model doesn't just require, you know, some code, it's building a giant new cluster, you know, now it's building giant new power plants, you know, pretty soon it's going to be building giant new fabs. And, you know, since ChatGPT, this kind of extraordinary sort of techno-capital acceleration has been set into motion. I mean, basically, you know, exactly a year ago today, you know, Nvidia had their first kind of blockbuster earnings call, right? Where it like went out 25% after hours and everyone was like, oh my God, AI, it's a thing. You know, I mean, I think within a year, you know, Nvidia, Nvidia data center revenue has gone from like, you know, a few billion a quarter to like, you know, 20, 25 billion

• 2:03Speaker 01

  a quarter now and, you know, continuing to go up, like, you know, big tech capex is skyrocketing. And You know, it's funny because it's both there's this sort of this kind of crazy scramble going on. But in some sense, it's just the sort of continuation of straight lines on a graph. Right. There's this kind of like long run trend, basically almost a decade of sort of training compute of the sort of largest AI systems growing by about, you know, half an order of magnitude, you know, 0.5 booms a year.

• 2:27Speaker 01

  And you can just kind of play that forward, right? So, you know, GPT-4, you know, rumored or reported to have finished pre-training in 2022. You know, the, the sort of cluster size there was rumored to be about, you know, 25,000 H100s, you know, sorry, A100s on semi-analysis. You know, that's, that's roughly, you know, if you do the math on that, it's maybe like a $500 million cluster. You know, it's very roughly 10 megawatts.

• 2:50Speaker 01

  And, um, you know, just play that forward, half a boom a year, right? So then 2024, that's a, you know, that's a cluster that's, uh, you know, 100 megawatts. That's like 100,000 H100 equivalents. You know, that's, um, uh, you know, costs in the sort of billions, you know, play it forward, you know, two more years, 2026, that's a cluster, that's a gigawatt. You know, that's a, that's a, you know, sort of a large nuclear reactor size. It's like the power of the Hoover Dam. You know, that costs tens of billions of dollars. That's like a million H100 equivalents. You know, 2028, that's a cluster that's 10 gigawatts, right? That's, that's more power than kind of like most US states. That's, you know, like 10 million H100s equivalents, you know, costs hundreds of billions of dollars. And then 2030, um, trillion dollar cluster, uh, 100 gigawatts, over 20% of US electricity production. You know, 100 million H100 equivalents.

• 3:35Speaker 01

  And, um, and that's just the training cluster, right? That's like the one largest training cluster, you know? And then there's more inference GPUs as well, right? Most of, you know, once there's products, most of them are going to be inference GPUs. And so, you know, US power production has barely grown for like, you know, decades. And now we're really in for a ride.

• 3:52Dwarkesh Patel?

  So, I mean, I, when I had Zuck on the podcast, he was claiming not a plateau per se, but that AI progress would be bottlenecked by specifically this constraint on energy. And specifically like, oh, gigawatt data centers are going to build another three gorgeous dam or something. I know that there's companies, according to public reports, who are planning things on the scale of a gigawatt data center. 10 gigawatt data center, who's going to be able to build that? I mean, 100 gigawatt center, like a state, are you getting, are you going to pump that into one's physical data center? How is this going to be possible?

• 4:26Speaker 01

  Yeah.

• 4:26Dwarkesh Patel?

  What is Zuck missing?

• 4:27Speaker 01

  I mean, you know, I don't know. I think the 10 gigawatt, you know, like six months ago, you know, 10 gigawatts was the taco of town. I mean, I think, I feel like now, you know, people have moved on, you know, 10 gigawatts is happening. I mean, I don't know, there's the information report on OpenAI and Microsoft planning a $100 billion cluster. So you gotta, you know, if you do the math.

• 4:43Dwarkesh Patel?

  Is that the gigawatt or is that the 10 gigawatt?

• 4:44Speaker 01

  I mean, I don't know, but you know, if you try to like map out, you know, how expensive would the, would the 10 gigawatt cluster be, you know, that's maybe a couple hundred billion. So it's sort of on that scale. And they're planning it, they're working on it, you know? So, so the, you know, it's not just sort of my crazy take, I mean AMD, AMD, I think forecasted a $400 billion AI accelerator market by 27. You know, I think, I think it's, you know, and AI accelerators are only part of the expenditures. It's sort of, you know, I think sort of a trillion dollars of sort of like total AI investment by 2027 is sort of like, we're very much in track on. I think the trillion dollar cluster is going to take a bit more sort of acceleration. But, you know, we saw how much sort of chat GPT unleashed, right? And so like every generation, you know, the models are going to be kind of crazy and people, it's going to shift the Overton window. And then, and then, you know, obviously the revenue comes in, right?

• 5:28Speaker 01

  So these are forward looking investments. The question is, do they pay off? Right? And so if we sort of estimated the, you know, the GPT four cluster at around 500 million. By the way, that's, that's sort of a common mistake people make is they say, you know, people say like a hundred million dollars for GPT four, but that's just the rental price, right? They're like, ah, you rent the cluster for three months. But it's, you know, if you're building the Vegas cluster, you gotta like, you gotta build the whole cluster, you gotta pay for the whole cluster, you can't just rent it for three months.

• 5:50Dwarkesh Patel?

  But can't you?

• 5:50Speaker 01

  But then, I mean, really, you know, once, once you're trying to get into the sort of hundreds of billions, eventually you gotta get to like 100 billion a year revenue. I think this is where it gets really interesting for the big tech companies, right? Cause like their revenues are on order, you know, hundreds of billions, right? So it's like 10 billion, fine, you know, and it'll pay off the, you know, 2024 sized training cluster. But, you know, really when sort of big tech, it'll be gangbusters, is 100 billion a year. And so the question is sort of how feasible is 100 billion a year from AI revenue?

• 6:15Speaker 01

  You know, it's a lot more than right now, but I think, you know, if you sort of believe in the trajectory of the AI systems as I do, and which we'll probably talk about, it's not that crazy, right? So there's, I think there's like 300 million, you know, ish Microsoft Office subscribers, right? And so they have Copilot now, and I don't know what they're selling it for, but you know, suppose you sold some sort of AI add-on for a hundred bucks a month. And you sold that to, you know, a third of Microsoft Office subscribers subscribe to that. That'd be a hundred billion right there. You know, a hundred dollars a month is, you know,

• 6:41Dwarkesh Patel?

  That's a lot, yeah.

• 6:42Speaker 01

  It's a lot, it's a lot, but it's

• 6:43Dwarkesh Patel?

  For a third of Office subscribers?

• 6:44Speaker 01

  Yeah, but it's, but it's, you know, for the average college worker, it's like a few hours of productivity a month. And it's, you know, kind of like, you have to be expecting pretty lame AI progress to not hit like, you know, some few hours of productivity a month of, of, of, yeah.

• 6:55Dwarkesh Patel?

  Okay, sure, sure. Let's assume all this.

• 6:57Speaker 01

  Yeah.

• 6:57Dwarkesh Patel?

  What, what happens in the next few years in terms of, what is the one gigawatt training, the AI that's trained on the one gigawatt data center? What can it do, the one on the 10 gigawatt data center? Just map out the next few years of AI progress for me.

• 7:11Speaker 01

  Yeah, I think probably the sort of 10 gigawatt-ish range is sort of my best guess for when you get the sort of true AGI. I mean, yeah, I think it's sort of like one gigawatt data center. And again, I think actually computers are rated and we're going to talk about that, but we will talk about compute right now. So, you know, I think sort of 25, 26, we're going to get models that are, you know, basically smarter than most college graduates. I think sort of the practice, a lot of the economic usefulness, I think, really depends on sort of, you know, sort of un-hobbling. Basically, it's, you know, the models are kind of, you know, they're smart, but they're limited, right? There, you know, there's this chatbot, you know, and things like being able to use a computer, things like being able to do kind of like agentic long horizon tasks.

• 7:44Dwarkesh Patel?

  Yeah.

• 7:46Speaker 01

  And then I think by 27, 28, you know, if you extrapolate the trends and, and, you know, we'll talk about that more later, and I talked about in the series. I think we hit, you know, basically, you know, like as smart as the smartest experts. I think that un-hobbling trajectory kind of points to, you know, it looks much more like an agent than a chatbot. And much more, almost like basically a drop-in remote worker, right? So it's not like, I think basically, I mean, I think this is the sort of question on the economic returns. I think a lot of the, a lot of the intermediate AI systems could be really useful, but you know, it actually just takes a lot of schlep to integrate them, right? Like GPT-4, you know, whatever, 4.5, you know, probably there's a lot you could do with them in a business use case. But, you know, you really got to change your workflows to make them useful. And it's just like, there's a lot of, you know, it's a very Tyler Cowen-esque take. It just takes a long time to diffuse.

• 8:24Dwarkesh Patel?

  Yeah.

• 8:24Speaker 01

  It's like, you know, we're in SF and so we missed out or whatever.

• 8:29Speaker 01

  But I think in some sense, You know, the way a lot of these systems want to be integrated is you kind of get this sort of sonic boom where it's, you know, the sort of intermediate systems could have done it, but it would have taken schlap. And before you do the schlap to integrate them, you get much more powerful systems, much more powerful systems that are sort of unhobbled. And so they're this agent and there's drop and remote worker. And, you know, and then you're kind of interacting with them like a coworker, right? You know, you can take, do Zoom calls with them and you're slacking them and you're like, ah, can you do this project? And then they go off and they, you know, go away for a week and write a first draft and get feedback on them and, you know, run tests on their code. And then they come back and then you see and you tell them a little bit more things or, you know, and that'll be much easier to integrate. And so, you know, it might be that actually you

• 9:12Speaker 01

  need a bit of overkill to make the sort of transition easy and to really harvest the gains.

• 9:16Dwarkesh Patel?

  What do you mean by the overkill? Overkill on the model capabilities?

• 9:19Speaker 01

  Yeah, yeah. So basically the intermediate models could do it, but it would take a lot of schlap. And so then, you know, they're like, actually, it's just the drop in remote worker kind of AGI that can automate, you know, cognitive tasks that actually just ends up kind of like, you know, basically it's you're like, you know, the intermediate models would have made the software engineer more productive, but you know, will the software engineer adopt it? And then the, you know, 27 model is, well, you know, you just don't need the software engineer. You can literally interact with it like a software engineer and it'll do the work of a software engineer.

• 9:43Dwarkesh Patel?

  So the last episode I did was with John Schulman.

• 9:45Speaker 01

  Yeah.

• 9:46Dwarkesh Patel?

  And I was asking about basically this. And one of the questions I asked is, we have these models that have been coming out in the last year and none of them seem to have significantly surpassed GPT-4 and certainly not in the agentic way in which they are interacting with as a coworker. You know, they'll brag that they got a few extra points on MMLU or something. And even GPT-4O, it's cool that they can talk like Scarlett Johansson or something, but like, and honestly, I'm gonna use that. Oh, I guess not anymore.

• 10:15Speaker 01

  I gotta say, yeah.

• 10:16Dwarkesh Patel?

  Okay, but the whole coworker thing. So this is gonna be a wrong question, but you can address it in any order, but. The, it makes sense to me why they'd be good at answering questions. They have a bunch of data about how to complete Wikipedia text or whatever. Where is the equivalent training data that enables it to understand what to make, what's going on this Zoom call? How does this connect with what they were talking about in the Slack? What is the cohesive project that they're going after based on all this context that I have? Where is that training data coming from?

• 10:49Speaker 01

  Yeah, so I think a really key question for sort of AI progress in the next few years is sort of how hard is it to do sort of unlock the test time compute overhang? So, you know, right now, GPT-4 answers a question and, you know, it kind of can do a few hundred tokens of kind of chain of thought. And that's already a huge improvement, right? Sort of like, this is a big un-hobbling before, you know, answer a math question is just shotgun. And, you know, if you try to kind of like answer a math question by saying the first thing that came to mind, you know, you wouldn't be very good. So, you know, GPT-4 thinks for a few hundred tokens.

• 11:18Speaker 01

  And, you know, if I thought for a few hundred, you know, if I think at like a hundred tokens a minute and I thought for a few minutes.

• 11:22Dwarkesh Patel?

  You think a bunch more than a hundred tokens a minute.

• 11:24Speaker 01

  I don't know. If I thought for like a hundred tokens a minute, you know, it's like what GPT-4 does, maybe it's like, you know, it's equivalent to me thinking for three minutes or whatever, right?

• 11:33Speaker 01

  You know, suppose GPT-4 could think for millions of tokens, right? That's sort of plus four rooms, plus four orders of magnitude on test time compute, just like on one problem. It can't do it right now. It kind of gets stuck, right? It like writes some code, even if, you know, you can do a little bit of iterative debugging, but eventually it just kind of like, it can't, it kind of gets stuck in something. It can't correct its errors and so on. And, you know, in a sense, there's this big overhang, right? And like other areas of ML, you know, there's this great paper on AlphaGo, right? Where you can trade off train time and test time compute. And if you can use, you know, four rooms more test time compute, that's almost like, you know, a three and a half room bigger model. Just because, again, like you can, you know, if a hundred tokens a minute, a few million tokens, that's, that's a few months of sort of working time. There's a lot more you can do in a few months of working time than, and then right now. So the question is, how hard is it to unlock that? And I think the

• 12:19Speaker 01

  You know, the sort of short timelines AI world is if it's not that hard. And the reason that might not be that hard is that, you know, there's only really a few extra tokens you need to learn, right? You need to kind of learn error correction tokens, the tokens where you're like, ah, I think I made a mistake. Let me think about that again. You need to learn the kind of planning tokens. That's kind of like, I'm going to start by making a plan. Here's my plan of attack. And then I'm going to write a draft and I'm going to like, now I'm going to critique my draft. I'm going to think about it. And so it's not, it's not things that models can do right now, but you know, the question is how hard is that? And in some sense also, you know, there's sort of two paths to agents, right? You know, when Chulto was on your podcast, you know, he talked about kind of scaling leading to more nines of reliability. And so that's one path. I think the other path is a sort of like unhobbling path where you, it needs, it needs to learn

• 13:04Speaker 01

  this kind of like system two process. And if it can learn this sort of system two process, it can just use kind of millions of tokens and think for them and be cohesive and be coherent. You know, one analogy, so when you drive, here's an analogy, when you drive, right? Okay, you're driving and most of the time you're kind of on autopilot, right? You're just kind of driving and you're doing well. And then, but sometimes you hit like a weird construction zone or a weird intersection, you know, and then I sometimes like, you know, my, my passenger seat, my girlfriend, I'm kind of like, ah, be quiet for a moment. I need to like figure out what's going on, right? Right? And that's sort of like, you know, you go from autopilot to like the system two is jumping in and you're thinking about how to do it. And so the scale, scaling is improving that system one autopilot. And I think it's sort of, it's the brute force way to get to kind of agents who just improve that sort of system.

• 13:46Speaker 01

  But if you can get that system two working, then, you know, I think you could like quite quickly jump, you know, to sort of this like more identified, you know, test time compute overhang is unlocked.

• 13:56Dwarkesh Patel?

  What's the reason to think that this is an easy win in the sense that, oh, you just get the, there's like some loss function that easily enables you to train it to enable the system two thinking?

• 14:10Speaker 01

  Yeah.

• 14:11Dwarkesh Patel?

  There's not a lot of animals that have system two thinking, you know? It like took a long time for evolution to give us system two thinking.

• 14:16Speaker 01

  Yeah.

• 14:16Dwarkesh Patel?

  The pre-training it like, listen, I get it. You got like trillions of tokens of internet texts. I get that like, yeah, you like match that and you get all these, all this free training capabilities. What's the reason to think that this is an easy and hobbling?

• 14:29Speaker 01

  Yeah, so, okay, a bunch of things. So, first of all, pre-training is magical, right? And it gave us this huge advantage for models of general intelligence, because you could just predict the next token, but predicting the next token, I mean, it's sort of a common misconception. But what it does is it lets this model learn these incredibly rich representations, right? Like, these sort of representation learning properties are the magic of deep learning. You have these models, and instead of learning just kind of like, you know, whatever, statistical artifacts or whatever, it learns sort of these models of the world. You know, that's also why they can kind of like generalize, right? Because it learned the right representations. And so, you know, you pre-train these models, and you have this sort of like raw bundle of capabilities that's really useful, and sort of this almost unformed raw mass.

• 15:11Speaker 01

  And sort of the unhobbling we've done over sort of like GPT-2 to GPT-4 was you kind of took this sort of like raw mass, and then you like RL-HF'd it into a really good chatbot. And that was a huge win, right? Like, you're going, going, you know, in the original, I think, InstructGPT paper, you know, RL-HF versus non-RL-HF model, it's like 100x model size win on sort of human preference rating, you know, it started to be able to do like simple chain of thought and so on. But you still have this advantage of all these kind of like raw capabilities, and I think there's still like a huge amount that you're not doing with them. And by the way, I think this sort of this pre-training advantage is also sort of the difference to robotics, right? Where I think robotics, you know, I think people used to say it was a hardware problem, but I think the hardware stuff is getting solved. But the thing we have right now is you don't have this huge advantage of being able to bootstrap yourself with pre-training. You don't have all this sort of unsupervised learning you can do.

• 15:56Speaker 01

  You have to start right away with this sort of RL, self-play, and so on. All right, so now the question is, why, you know, why might some of this unhobbling and RL and so on work?

• 16:09Speaker 01

  And again, there's sort of this advantage of bootstrapping, right? So, you know, your Twitter bio is being pre-trained, right? But you're actually not being pre-trained anymore. You are not being pre-trained anymore. You were pre-trained in like grade school and high school. At some point, you transitioned to be able, being able to like learn by yourself, right? You weren't able to do that in elementary school. I don't know, middle school, probably, I mean, high school is maybe when it sort of started, you need some guidance. You know, college, you know, if you're smart, you can kind of teach yourself. And then sort of models are just starting to enter that regime, right? And so, it's sort of like, it's a little bit, it's probably a little bit more scaling. And then you got to figure out what goes on top. And it won't be trivial, right? So, a lot of

• 16:48Speaker 01

  A lot of deep learning is sort of like, you know, it sort of seems very obvious in retrospect. And there's sort of this, some obvious cluster of ideas, right? There's sort of some kind of like thing that seems a little dumb, but that just kind of works. But there's a lot of details you have to get right. So I'm not saying this, you know, we're going to get this, you know, next month or whatever. I think it's going to take a while to like really figure out the details.

• 17:04Dwarkesh Patel?

  A while for you is like half a year or something?

• 17:06Speaker 01

  Well, I don't know.

• 17:07Dwarkesh Patel?

  That makes a month, six months.

• 17:09Speaker 01

  Between six months and three years, you know? But I, but you know, I think it's possible. And I think there's, um, you know, I think, and this is, I think is also very related to the sort of issue of the data wall. But I mean, I think the, um, you know, one intuition on the sort of like learning, learning, learning by yourself, right? It's sort of pre-training is kind of the words are flying by.

• 17:28Dwarkesh Patel?

  Yeah.

• 17:28Speaker 01

  Right? You know, and, and, um, or it's like, you know, the teacher is lecturing to you and the models, you know, the words are flying by, you know, they're taking, they're just getting a little bit from it. But that's sort of not what you do when you learn from yourself, right? When you learn by yourself, you know, so you're reading a dense math textbook. You're not just kind of like skimming through it once, you know, you wouldn't learn that much from it. I mean, some word cells just skim through, you know, reread and reread the math textbook and then they memorize. And that's sort of, you know, like if you just repeated the data, then they memorize. What you do is you kind of like, you read a page, kind of think about it. You have some internal monologue going on. You have a conversation with a study buddy. You try a practice problem, you know, you fail a bunch of times. At some point it clicks and you're like, this made sense. Then you read a few more pages. And so we've kind of bootstrapped our way to being, being able to do that now with models or like just starting to be able to do that.

• 18:13Speaker 01

  And then the question is, you know, being able to like read it, think about it, you know, try problems. And the question is, can you, you know, all this sort of self-play synthetic data RL is kind of like making that thing work.

• 18:23Dwarkesh Patel?

  Yep.

• 18:26Speaker 01

  So basically translate, trade it translating like in context, like right, like right now there's like in context learning, right? Super sample efficient. There's that, you know, in the Gemini paper, right? It just like learns a language in context. And then you have pre-training, not at all sample efficient.

• 18:41Speaker 01

  But you know, what humans do is they kind of like, they do in context learning. You read a book, you think about it until eventually it clicks. But then you somehow distill that back into the weights. And in some sense, that's sort of like what RL is trying to do. And like when RL is super finicky, but when RL works, RL is kind of magical because. It's sort of the best possible data for the model. It's like when you try a practice problem and it, you know, and then you fail. And at some point you kind of figure it out in a way that makes sense to you. That's sort of like the best possible data for you because like the way you would have solved the problem. And that's sort of, that's what RL is. Rather than just, you know, you kind of reads how somebody else solved the problem and doesn't, you know, initially click.

• 19:17Dwarkesh Patel?

  Yeah. By the way, if that take sounds familiar, because it was like part of the question I asked John Scholman. That goes to illustrate the thing I said in the intro where like a bunch of the things I've learned about AI. It's just like, we do these dinners before the interviews and you show it to a couple and like, ah, what should I ask John Scholman? What should I ask Dario?

• 19:37Dwarkesh Patel?

  Okay, suppose this is the way things go and we get these unhobblings.

• 19:40Speaker 01

  Yeah. Well, and the scaling, right? So it's like you have this baseline, just enormous force of scaling, right? Where it's like GPT two to GPT four, you know, GP two could kind of like, it was amazing, right? Could string together plausible sentences, but you know, it could, it could barely do anything. It was kind of like preschooler. And then GPT four is, you know, it's writing code, it like, you know, can do hard math. It's, and so it's sort of like smart high school. And so this big jump and you know, and sort of the essay series I go through and kind of count the orders of magnitude of compute scale up of algorithmic progress. And so sort of scaling alone, you know, sort of by 27, 28 is going to do another kind of preschool to high school jump on top of GPT four. And so that'll already be just like at a per token level, just incredibly smart. That'll get you some more reliability. And then you add these unhobblings that make it look much less like a chatbot, more like this agent, like a drop in remote worker.

• 20:26Speaker 01

  And, you know, that's when things really get going.

• 20:29Dwarkesh Patel?

  Okay.

• 20:29Speaker 01

  Yeah.

• 20:31Dwarkesh Patel?

  I want to ask most questions about this.

• 20:32Speaker 01

  Well, yeah.

• 20:33Dwarkesh Patel?

  Yeah, yeah, yeah. Let's zoom out. Okay. So suppose you're right about this.

• 20:36Speaker 01

  Yeah.

• 20:37Dwarkesh Patel?

  And I guess you, this is because of the 2027 cluster, you've got 10 gigawatt, 2027, 10 gigawatts, something.

• 20:45Speaker 01

  28 is the 10 gigawatt.

• 20:46Dwarkesh Patel?

  Okay.

• 20:46Speaker 01

  I mean, you'll be pulled for it.

• 20:48Dwarkesh Patel?

  Okay, sure. Something.

• 20:49Speaker 01

  Yeah.

• 20:49Dwarkesh Patel?

  And so I guess that's like 5.5 level by 2027, like whatever that's called, right? What does the world look like at that point? You have these remote workers who can replace people. What is the reaction to that in terms of the economy, politics, geopolitics?

• 21:06Speaker 01

  Yeah, so. You know, I think 2023 was kind of a really interesting year to experience as somebody who was like, you know, really following the AI stuff where, you know, before that.

• 21:15Dwarkesh Patel?

  What were you doing in 2023?

• 21:17Speaker 01

  I mean, OpenAI.

• 21:18Dwarkesh Patel?

  Okay.

• 21:19Speaker 01

  Yeah, yeah, yeah. And, and, and.

• 21:23Speaker 01

  You know, it kind of went, you know, I mean, I was, I was been thinking about this and, you know, like talking to a lot of people, you know, in the years before, and it was this kind of weird thing, you know, you almost didn't want to talk about AI or AGI, and it was kind of a dirty word, right? And then 2023, you know, people saw ChatGPT for the first time, and they saw GPT-4, and it just like exploded, right? It triggered this kind of like, you know, you know, huge sort of capital expenditures from all these firms and, and, and, you know, the explosion in revenue from Nvidia and so on. And, um, you know, things have been quiet since then, but, you know, the next thing has been in the oven. And I sort of expect sort of every generation, these kind of like G-forces to intensify, right? It's like, people see the models, um, there's like, you know, people haven't counted them, so they're gonna be surprised, and it'll be kind of crazy. And then, you know, revenue is gonna accelerate, you know, suppose you do hit the 10 billion, you know, end of this year.

• 22:07Speaker 01

  Suppose it like just continues on this sort of doubling trajectory of, you know, like every six months of revenue doubling. You know, it's like, you're not actually that far from 100 billion, you know, maybe that's like 26. And so, you know, at some point, you know, like, you know, sort of what happened to Nvidia is gonna happen to Big Tech, you know, like their stocks, they're, you know, that's gonna explode.

• 22:25Speaker 01

  And I mean, I think a lot more people are gonna feel it, right? I mean, I think the, um, I think 2023 was the sort of moment for me where it went from kind of AGI as a sort of theoretical abstract thing, and you'd make the models to like, I see it, I feel it. And like, I see the path, I see where it's going. I like, I think I can see the cluster where it's trained on, like the rough combination of algorithms, the people, like how it's happening. And I think, you know, most of the world is not, you know, most of the people feel it are like right here, you know, right? But, but, you know, I think a lot more of the world is gonna start feeling it.

• 22:56Speaker 01

  And I think that's gonna start being kind of intense.

• 22:58Dwarkesh Patel?

  Okay, so, right now, who feels it? You can, you go on Twitter, and there's these GPT rapper companies, like, whoa, GPT-4 is gonna change our business.

• 23:06Speaker 01

  I'm so, so bearish on the rapper companies, right? Because like, they're, they're the ones that are gonna be, like, the rapper companies are betting on stagnation, right? Rapper companies are betting, like, you have these intermediate models, and it takes so much luck to integrate them. And I'm kind of like, I'm really bearish, because I'm like, we're just gonna sonic boom you, you know, and we're gonna get the unhobbling, we're gonna get the drop-in remote worker, and then, you know, your stuff is not gonna matter.

• 23:23Dwarkesh Patel?

  Okay, sure, sure, so that, that's done. Now, um, who, so SF is paying attention now, or this crowd here is paying attention. Who is gonna be paying attention in 2026, 2027? And presumably, this is, these are years in which the hundreds of billions of capex is being spent on AI.

• 23:40Speaker 01

  I mean, I think the, uh, the national security state is gonna be starting to pay a lot of attention.

• 23:46Speaker 01

  And I, you know, I hope we get to talk about that.

• 23:47Dwarkesh Patel?

  Okay, let's talk about it now.

• 23:48Speaker 01

  We're gonna talk about it now.

• 23:50Dwarkesh Patel?

  Like, what is this sort of political reaction immediately?

• 23:52Speaker 01

  Yeah.

• 23:52Dwarkesh Patel?

  And even like internationally, like what people see like right now, I don't know if like Xi Jinping like reads the news and sees like,

• 23:58Speaker 01

  Yeah, I don't know.

• 23:58Dwarkesh Patel?

  Well, GBD4, oh my God, like MML you score on that. What are you doing about this comrade?

• 24:03Speaker 01

  Yeah.

• 24:05Dwarkesh Patel?

  So what happens when the like, what, where the GBD, he's like sees a remote replacement and it has a hundred billion dollars in revenue? There's a lot of businesses that have a hundred billion dollars in revenue and people don't like, aren't staying up all night talking about it.

• 24:17Speaker 01

  The question, I think the question is, when, when does the CCP and when does the sort of American national security establishment realize that superintelligence is going to be absolutely decisive for national power, right? And this is where, you know, the sort of intelligence explosion stuff comes in, which, you know, we should also talk about later. You know, it's sort of like, you know, you have AGI, you have this sort of drop in remote worker that can replace, you know, you or me. At least that's sort of remote jobs, you know, kind of jobs. And then, you know,

• 24:43Speaker 01

  I think fairly quickly, you know, I mean, by default, you know, you turn the crank, you know, one or two more times, you know, and then you get a thing that's smarter than humans. But I think even, even more than just turning the cramp a few more time, crank a few more times. You know, I think one of the first jobs to be automated is going to be that of sort of an AI researcher engineer. And if you can automate AI research, you know, I think things can start going very fast. You know, right now, there's already this trend of, you know, half an order of magnitude a year of algorithmic progress. You know, suppose, you know, at that point, you're, you're going to have GPU fleets in the tens of millions for inference, you know, or more. And, um, you're going to be able to run like a hundred million human, human equivalents of these sort of automated AI researchers. And if you can do that, you know, you can maybe do, you know, a decade's worth of sort of ML research progress in a year. You know, you get this some sort of 10x speed up.

• 25:29Speaker 01

  And if you can do that, I think you can make the jump to kind of like AI that is vastly smarter than humans, you know, within a year, a couple of years. And then, you know, that broadens, right? So you have this, you have this sort of initial acceleration of AI research that broadens to like, you apply R and D to a bunch of other fields of technology. And the sort of like extremes, you know, at this point, you have like a billion just super intelligent researchers, engineers, technicians, everything. You're superbly competent at all the things, you know, they're going to figure out robotics. Or we talked about it being a software problem. Well, you know, you have, you have a billion of super smart, smarter than the smartest human researchers, AI researchers on your cluster, you know, at some point during the intelligence explosion. They're going to be able to figure out robotics, you know, and then again, that expands and.

• 26:12Speaker 01

  You know, I think if you play this picture forward, I think it is fairly unlike any other technology in that it will, I think, you know, a couple of years of lead could be utterly decisive in, say, like, military competition, right? You know, if you look at like Gulf War I, right? Gulf War I, you know, like, you know, Western coalition forces, you know, they had, you know, like, a hundred to one kill ratio, right? And that was like, they had better sensors on their tanks, you know, and they had, they had better, you know, more precision, precision missiles, right, like GPS, and they had, you know, stealth, and they had sort of a few, you know, maybe twenty, thirty years of technological lead, right? And they, you know, just completely crushed them. Superintelligence applied to sort of broad fields of R and D, and then, you know, the sort of industrial explosion

• 26:55Speaker 01

  as well, you have the robots, you're just making lots of material, you know, I think that could compress, I mean, basically compress kind of like a century worth of technological progress in less than a decade. And that means that, you know, a couple of years could mean a sort of Gulf War I style, like, you know, advantage in, in military affairs. And, you know, including like, you know, a decisive advantage that even like preempts nukes, right? Suppose like, you know, how do you find the stealth in nuclear submarines? Like, right now, that's a problem of like, you have sensors, you have the software to like, attack where they are, you know, you can do that, you can find them, you have kind of like millions or billions of like mosquito-like, you know, sized drones, and that, you know, they take out the nuclear submarines, they take out the missile launchers, they take out the other nukes. And anyway, so I think enormously destabilizing, enormously

• 27:37Speaker 01

  important for national power. And at some point, I think people are going to realize that. Not yet, but they will. And when they will, I think there will be sort of, you know, I don't think it'll just be the sort of AI researchers in charge. And, you know, I think on the, you know, the CCP is gonna, you know, have sort of an all-out effort to like infiltrate American AI labs, right? You know, like billions of dollars, thousands of people, you know, full force of the sort of, you know, Ministry of State Security. CCP is gonna try to, you know, like, outbuild us, right? Like, they, you know, their, you know, power in China, you know, like, the electric grid, you know, they added a US is, you know, a complete, like, they added as much power in the last decade as like sort of entire US electric grid. So like, the hundred gigawatt cluster, you know, at least the hundred gigawatts is gonna be a lot easier for them to get. And so I think

• 28:20Speaker 01

  sort of, you know, by this point, I think it's gonna be like a extremely intense sort of international competition.

• 28:25Dwarkesh Patel?

  Hmm. Okay, so in this picture...

• 28:30Dwarkesh Patel?

  One thing I'm uncertain about is whether it's more like what you say where it's more of an implosion of you have developed an AGI and then you make it into an AI researcher. And for a while, a year or something, you're only using this ability to make hundreds of millions of other AI researchers. And then like the thing that comes out of this really frenetic process is a super intelligence. And then that goes out in the world and is developing robotics and helping you take over other countries.

• 29:00Speaker 01

  I think it's a little bit more, you know, it's a little bit more kind of like, you know, it's not like, you know, on and off. It's a little bit more gradual, but it's sort of like, it's an explosion that starts narrowly. It can do cognitive jobs, you know, the highest ROI used for cognitive jobs is make the AI better, like solve robotics, you know, and as, as, as you solve robotics, now you can do R and D and, you know, like biology and other technology. Um, you know, initially you start with the factory workers, you know, they're wearing the glasses and the AirPods, you know, and the AI is instructing them, right? Because, you know, you kind of make any worker into a skilled technician. And then you have the robots come in. And anyway, so it sort of expands, this process expands.

• 29:30Dwarkesh Patel?

  Metas Rayvans are a compliment to their llama.

• 29:32Speaker 01

  Well, you know, it's like, whatever, like, you know, the fabs in the US are constrained to skilled workers, right? You have, you have, even if you don't have robots, you have the cognitive super intelligence and, you know, it can kind of make them all into skilled workers immediately. But that's, you know, it's a very brief period, you know, robots will come soon.

• 29:44Dwarkesh Patel?

  Sure. Okay. Okay. So suppose this is actually how the tech progresses in the United States, maybe because these companies are already experiencing hundreds of billions of dollars of AI revenue.

• 29:54Speaker 01

  At this point, you know, companies are borrowing, you know, hundreds of billions of more in the corporate debt markets, you know.

• 29:58Dwarkesh Patel?

  But why is a CCP bureaucrat, some 60 year old guy, he looks at this and he's like, oh, it's like Copilot has gotten better now. Um, why are they now?

• 30:06Speaker 01

  I mean, this is much more than Copilot has gotten better now. I mean,

• 30:09Dwarkesh Patel?

  But to them, like, yeah, so they're, because to shift the production of an entire country to dislocate energy that is otherwise being used for consumer goods or something, and to make it that all feed into the data centers, what, will that be obvious? Because like, part of this whole story is you realize the super intelligence is coming soon, right? And I guess you realize it, maybe I realize it. I'm not sure how much I realize it. But will the, will the national security apparatus in the United States and will the CCP realize it?

• 30:41Speaker 01

  Yeah. I mean, look, I think in some sense, this is a really key question. Um, I think we have sort of a few more years of mid-game, basically, and where you have a few more 2023s and that just starts updating more and more people. And, you know, I think, you know, the trend lines, you know, will become clear.

• 30:57Speaker 01

  You know, I think, I think you will see some amount of the sort of COVID dynamic, right? You know, like COVID was like, you know, February, February of 2020, you know, it's like, honestly feels a lot like today, you know, where it's like, you know, it feels like this utterly crazy thing is hap- is about, you know, is impending, is coming. You kind of see the exponential and yet most of the world just doesn't realize, right? The mayor of New York is like, go out to the shows and this is just, you know, like Asian racism or whatever, you know, and, and, but, you know, at some point, the exponential like, you know, some point, people saw it. And then, you know, like just kind of crazy radical reactions came.

• 31:32Dwarkesh Patel?

  Right. Okay. So by the way, what were you doing during COVID? Or when, like February? Okay. Like freshman, sophomore, what?

• 31:39Speaker 01

  Junior.

• 31:40Dwarkesh Patel?

  Hmm. But still like, what were you like 17 year old junior or something?

• 31:45Dwarkesh Patel?

  And, and then you like, did you short the market or something?

• 31:48Speaker 01

  Yeah, yeah, yeah, yeah.

• 31:49Dwarkesh Patel?

  Did you, did you, did you sell at the right time?

• 31:51Speaker 01

  Yeah.

• 31:52Dwarkesh Patel?

  Okay. Yeah. So there will be like a March 2020 moment that the thing that was COVID but here. Now, then you can like make the analogy that you make in a series that this will then cause the reaction of like, we got to do the Menhan project for America here. I wonder what the politics of this will be like because the difference here is it's not just like, we need the bomb to beat the Nazis. It's, we're building this thing that's making all our entry prices rise a bunch and it's automating a bunch of our jobs and the climate change stuff, like people are going to be like, oh my God, it's making climate change worse and it's helping big tech. Like politically, this doesn't seem like a dynamic where the national security apparatus or the president is like, we have

• 32:36Dwarkesh Patel?

  to step on the gas here and like make sure America wins.

• 32:40Speaker 01

  Yeah. I mean, again, I think a lot of this really depends on sort of how much people are feeling it, how much people are seeing it.

• 32:49Speaker 01

  You know, I think there's a thing where, you know, kind of basically our generation, right, we're kind of so used to kind of, you know, basically peace and like, you know, the world, you know, American hegemony and nothing matters. But, you know, the sort of like extremely intense and these extraordinary things happening in the world and like intense international competition is like very much the historical norm. Like, in some sense, it's like, you know, sort of, there's a sort of 20-year, very unique period, but like, you know, the history of the world is like, you know, like in World War II, right, it was like 50% of GDP went to, you know, like, you know, war-pride and production. The US borrowed over 60% of GDP, you know, and, you know, I think Germany and Japan over 100%. World War I, you know, UK, Japan, sorry, UK, France, Germany, all borrowed over 100%

• 33:34Speaker 01

  of GDP. And, you know, I think the sort of, much more was on the line, right? Like, you know, and, you know, people talk about World War I being so destructive and, you know, like 20 million Soviet soldiers dying and like 20% of Poland, but, you know, that was just a sort of like, that happened all the time, right? You know, like, seven years' war, you know, like, whatever, 20, 30% of Prussia died, you know, like, 30 years' war, you know, like, I think, like, you know, up to 50% of like, large swath of Germany died. And, you know, I think the question is, will these sort of like, will people see that the stakes here are really, really high and that basically is sort of like, history is actually back?

• 34:17Speaker 01

  And I think, you know, I think the American national security state thinks very seriously about stuff like this. They think very seriously about competition with China. I think China very much thinks of itself on this as our historical mission, rejuvenation of the Chinese nation, a lot about national power, I think a lot about like, the world order. And then, you know, I think there's a real question on timing, right? Like, do they, do they start taking this seriously, right? Like, when the intelligence explosion was already happening, like, quite late? Or do they start taking this seriously like, two years earlier? And that matters a lot for how things play out. But at some point, they will. And at some point, they will realize that this will be sort of utterly decisive for...

• 34:53Speaker 01

  You know, not just kind of like some proxy war somewhere, but you know, like whether liberal democracy can continue to thrive, whether you know, whether the CCP will continue existing. And I think that will activate sort of forces that we haven't seen in a long time.

• 35:05Dwarkesh Patel?

  Hmm. The great conflict, the great power conflict thing definitely seems compelling. I think just all kinds of different things seem much more likely when you think from a historical perspective, when you zoom out beyond the liberal democracy that we've been living in, had the pleasure to live in America, let's say 80 years, including dictatorships, including all, obviously war, famine, whatever. I was reading the Gulag Archipelago and one of the chapters begins with Sojanitsyn saying, if you would have told Russian citizens under the czars that because of all these new technologies, we wouldn't see some great Russian revival or becomes a great power and the citizens are made wealthy, but instead what you would see is tens of millions of Soviet citizens

• 35:48Dwarkesh Patel?

  tortured by millions of beasts in the worst possible ways. And that this is what would be the result of the 20th century, they wouldn't have believed you, they'd have called you a slanderer.

• 35:58Speaker 01

  Yeah, and you know, the, you know, the possibilities for dictatorship with superintelligence are sort of even crazier, right? I think, you know, imagine you have a perfectly loyal military and security force, right? That's it. No more, no more rebellions, right? No more popular uprisings, you know, perfectly loyal, you know, you have, you know, perfect lie detection, you know, you have surveillance of everybody, you know, you can perfectly figure out who's the dissenter, weed them out, you know, no Gorbachev would have ever risen to power, who had some doubts about the system, you know, no military coup would have ever happened. And I think you, I mean, you know, I think there's a real way in which, you know, part of why things have worked out is that, you know, ideas can evolve and you know, there's sort of like some, some sense in which sort of time heals a lot of wounds and time, you

• 36:43Speaker 01

  know, and solves, solves, you know, a lot of debates and a lot of people had really strong convictions, but you know, a lot of those have been overturned by time because there's been this continued pluralism and evolution. I think there's a way in which kind of like, you know, if you take a CCP like approach to kind of like truth, truth is what the party says, and you supercharge that with superintelligence, I think there's a way in which that could just be like locked in and enshrined for, you know, a long time. And I think the possibilities are pretty terrifying.

• 37:07Speaker 01

  You know, your point about, you know, history and sort of like living in America for the past eight years, you know, I think this is one of the things I sort of took away from growing up in Germany is a lot of this stuff feels more visceral, right? Like, you know, my, my mother grew up in the former East, my father in the former West, they like met shortly after the wall fell, right? Like, the end of the Cold War was this sort of extremely pivotal moment for me, because it's, you know, it's the reason I exist, right? And then, you know, growing up in Berlin and, you know, the former wall, you know, my great grandmother, who is, you know, still alive, is very important in my life. You know, she was born in 34, you know, grew up, you know, during the Nazi era, during, you know, all that, you know, then World War II, you know, like saw the firebombing of Dresden from this sort of, you know, country cottage or whatever, where, you know, they as kids were, you know, then, and then, you know, then spends most of her life in

• 37:52Speaker 01

  sort of the East German Communist dictatorship. You know, she'd tell me about, you know, in like 54, and there's like the popular uprising, you know, and Soviet tanks came in, you know, her, her husband was telling her to get home really quickly, you know, get off, off the streets, you know, had a, had a son who, who tried to, you know, ride a motorcycle across, across the Iron Curtain, and then was put in a Stasi prison for a while. Um, you know, and then finally, you know, when she's almost 60, you know, it's the first time she lives in, you know, a free country, um, and, and a wealthy country. And, um, you know, when I was a kid, she was, um, she, the thing she always really didn't want me to do was like get involved in politics, because like joining a political party was just, you know, it was a very bad connotations for her. Um, anyway, and

• 38:37Speaker 01

  she, and she sort of raised me when I was young, you know, and so, um, it, you know, it doesn't feel that long ago, it feels very close.

• 38:43Dwarkesh Patel?

  Yeah. So I, I wonder when we're talking today about the CCP, listen, the people in China who will be doing the, their version of the project will be AI researchers who are somewhat Westernized, who interact with, either got educated in the West or have colleagues in the West. Are they going to sign up for the CCP project that's going to hand over control to Xi Jinping? What, what, what's your sense on, I mean, you're just like, fundamentally, they're just people, right? Like, can't you like convince them about the dangers of superintelligence or something?

• 39:18Speaker 01

  Will they be in charge, though? I mean, in some sense, this is also the case, you know, in the US or whatever.

• 39:25Dwarkesh Patel?

  Right.

• 39:25Speaker 01

  This is sort of like rapidly depreciating influence of the lab employees. Like, right now, the sort of AI lab employees have so much power, right? Over this, you know, like you saw this-

• 39:32Dwarkesh Patel?

  But they're gonna get automated and then-

• 39:33Speaker 01

  Well, yeah. We saw this in November event, so much power, right? But both, I mean, both are gonna get automated and they're gonna lose all their power. And it'll just be, you know, kind of like a few people in charge with their sort of armies of automated AI. But also, you know, it's sort of like the politicians and the generals and the sort of national security state. You know, a lot of, you know, I mean, there's sort of, this is the sort of, some of these classic scenes from, you know, the Oppenheimer movies. You know, the scientists built it and then it was kind of, you know, and the bomb was shipped away and it was out of their hands. You know, I actually, yeah, I think, I actually think it's good for like lab employees to be aware of this. It's like, you have a lot of power now, but, you know, maybe not for that long. And, you know, use it wisely.

• 40:09Speaker 01

  Yeah, I do, I do think they would benefit from some more, you know, organs of representative democracy.

• 40:13Dwarkesh Patel?

  What do you mean by that?

• 40:13Speaker 01

  Oh, I mean, I, you know, in the sort of, in the OpenAI board events, you know, employee power was exercised in a very sort of direct democracy way.

• 40:19Dwarkesh Patel?

  Right.

• 40:19Speaker 01

  And I feel like that's how some of had that one about, you know, I think it really highlighted the benefits of representative democracy and having some deliberate organs.

• 40:26Dwarkesh Patel?

  Interesting.

• 40:26Speaker 01

  Yeah.

• 40:27Dwarkesh Patel?

  Well, let's go back to the 100 billion revenue, whatever, and so these companies-

• 40:31Speaker 01

  Million dollar cluster.

• 40:32Dwarkesh Patel?

  Yeah, the companies are deploying, trying to build clusters that are this big.

• 40:36Speaker 01

  Yeah.

• 40:37Dwarkesh Patel?

  Where are they building it? Because if, if you say it's the amount of energy that would be required for a small or medium-sized US state, is it then Colorado gets no power and it's happening in the United States or is it happening somewhere else?

• 40:47Speaker 01

  Oh, I mean, I think that, I mean, in some sense, this is the thing that I always find funny is, you know, you talk about Colorado gets no power. You know, the easy way to get the power would be like, you know, displace less economically useful stuff. You know, it's like, whatever, buy up the aluminum smelting plant and, you know, that has a gigawatt and, you know, we're gonna replace it with, with the data center because that's important. I mean, that's not actually happening because a lot of these power contracts are really sort of long-term locked in. You know, because obviously people don't like things like this. And so it's sort of, it seems like in practice, what it's, what it's requiring, at least right now, is building new power. The, that might change. And I think that that's when things get really interesting, when it's like, no, we're just dedicating all of the power to the AGI. Anyway, so right now it's building new power, 10 gigawatt, I think quite doable. You know, it's like a few percent of like US natural gas production.

• 41:28Speaker 01

  You know, I mean, when you have the 10 gigawatt chain and cluster, you have a lot more inference. So that starts getting more, you know, I think 100 gigawatt, that starts getting pretty wild. You know, that's, you know, again, it's like over 20% of US electricity production. I think it's pretty doable, especially if you're willing to go for like natural gas. I do, I do think, I do think it is incredibly important, incredibly important that these clusters are in the United States.

• 41:49Dwarkesh Patel?

  And why does it matter it's in the US?

• 41:53Speaker 01

  I mean, look, I think there's some people who are, you know, trying to build clusters elsewhere. And, you know, there's like a lot of free flowing Middle Eastern money that's trying to build clusters elsewhere.

• 42:05Speaker 01

  I think this comes back to the sort of like national security question we talked about earlier. Like, would you, I mean, would you do the Manhattan Project in the UAE, right? And I think, I think basically like putting, putting the clusters, you know, I think you can put them in the US. You can put them in sort of like ally democracies. But I think once you put them in kind of like dictatorships, authoritarian dictatorships, you kind of create this irreversible security risk, right? So, I mean, one cluster is there, much easier for them to exfiltrate the weights. You know, they can like literally steal the AGI, the super intelligence. It's like they got a copy of the, you know, of the, of the atomic bomb, you know, and they just got a direct replica of that. And it makes it much easier to them. I mean, we're tied to China, you can ship that to China. So that's a huge risk. Another thing is they can just seize the compute, right? Like maybe right now they just think of this, I mean, in general, I think people, you know, I think the issue here is people are thinking of this as a, you know, chat GPT, big tech product clusters.

• 42:48Speaker 01

  But I think the clusters being planned now, you know, three to five years out, like it will be the like AGI, super intelligence clusters. And so anyway, so like when things get hot, you know, they might just seize the compute. And I don't know, supposedly put like, you know, 25% of the compute capacity in these sort of Middle Eastern dictatorships. Well, they seize that and now it's sort of a ratio of compute of three to one. And, you know, we still have some more, but even like, even, even only, only 25% of compute there. Like, I think it starts getting pretty hairy, you know, I think three to one is like not that great of a ratio. You can do a lot with that amount of compute. And then look, even, even if they don't actually do this, right? Even they don't actually seize the compute, even they actually don't steal the weights. There's just a lot of implicit leverage you get, right? They get, they get the seat at the AGI table.

• 43:28Speaker 01

  And, um, you know, I don't know why we're giving authoritarian dictatorships the seat at the AGI table.

• 43:36Dwarkesh Patel?

  Okay, so there's gonna be a lot of compute in the Middle East if these deals go through. First of all, who's who is it just like every single big tech company is just trying to figure out?

• 43:44Speaker 01

  Not everyone, some.

• 43:45Dwarkesh Patel?

  Okay, okay. Well, I guess there's reports, I think, Microsoft or?

• 43:48Speaker 01

  Yeah, yeah, yeah, yeah, yeah.

• 43:50Dwarkesh Patel?

  Which we'll get into.

• 43:51Speaker 01

  Yeah.

• 43:51Dwarkesh Patel?

  So they UAE gets a bunch of compute because we're building the clusters there. And why? So let's say they have 25% of the why does a compute ratio matter? Is it if it's about them being able to kick off the intelligence explosion? Isn't it just some threshold where you have 100 million AI researchers or you don't?

• 44:12Speaker 01

  I mean, you can do a lot with, you know, 33 million extremely smart scientists. And, you know, and again, a lot of the stuff, you know, so first of all, it's like, you know, that might be enough to build the crazy bioweapons, right? And then you're in a situation where like now, wow, we've just like they stole the weights. They seized the compute. Now they can make, you know, they can build these crazy new WMDs that, you know, will be possible super intelligence. And then you've just kind of like proliferated the stuff and, you know, it'll be really powerful. And also, I mean, I think, you know, three to three acts on compute isn't actually that much. And so the, you know, the, you know, I think a thing I worry a lot about is,

• 44:49Speaker 01

  I think everything, I think the riskiest situation is if we're in some sort of like really tight neck, feverish international struggle, right? If we're like really close with the CCP and we're like months apart. I think the situation we want to be in, we could be in if we played our cards right, is a little bit more like, you know, the US, you know, building the atomic bomb versus the German project. Way behind, you know, years behind. And if we have that, I think we just have so much more wiggle room, like to get safety right. We're going to be building like, you know, there's going to be these crazy new WMDs, you know, things that completely undermine, you know, nuclear deterrence, you know, intense competition. And that's so much easier to deal with if, you know, you're like, you know, it's not just, you know, you don't have somebody right on your tails. You gotta go, go, go. You gotta go at maximum speed. You have no wiggle room. You're worried that at any time they can overtake you.

• 45:34Speaker 01

  I mean, they can also just try to outbuild you, right? Like they can might, they might literally win, like China might literally win. If they can steal the weights because they can, they can outbuild you. And they maybe have less caution, both, you know, good and bad caution, you know, kind of like whatever unreasonable regulations we have. Or you're just in this really tight race. And I think is that sort of like, if you're in this really tight race, this sort of feverish struggle, I think that's when sort of there's the greatest peril of self destruction.

• 45:58Dwarkesh Patel?

  So then presumably the companies that are trying to build clusters in the Middle East realize this. What is the, is it just that it's impossible to do this in America? And if you want American companies to do this at all, then you do it in the Middle East or not at all? And then you just like have China build the three gorgeous dam cluster?

• 46:11Speaker 01

  I mean, there's a few reasons. One of them is just like, people aren't thinking about this as the AGI super intelligence cluster. They're just like, ah, you know, like cool clusters for my, you know, for my chat GPT.

• 46:19Dwarkesh Patel?

  So they're building, and the plans right now are clusters which are ones that are like, because if you're doing once or inference, presumably you could like spread them out across the country or something. But the ones they're building, they realize we're gonna do one training run in this thing we're building.

• 46:32Speaker 01

  I just think it's harder to distinguish between inference and training compute. And so people can claim it's training compute, but I think they might realize that actually, you know, this is gonna be useful for, sorry, they might say it's inference compute, and actually it's useful for training compute too.

• 46:44Dwarkesh Patel?

  Because of synthetic data and things like that?

• 46:46Speaker 01

  Yeah, the future of training, you know, like RL looks a lot like inference, for example, right? Or, or you just kind of like end up connecting them, you know, in time, you know, it's like a lot of raw material, you know, it's like, you know, it's, it's placing your uranium refinement facilities there.

• 46:57Dwarkesh Patel?

  Sure.

• 46:57Speaker 01

  Anyway, so a few reasons, right? One is just like, they don't think about this as the AGI cluster. Another is just like easy money from the Middle East, right? Another one is like, you know, people saying, some people think that, you know, you can't do it in the US. And, you know, I think we actually face this sort of real system competition here, because again, some people think it's only autocracies that can do this, that can kind of like top down, mobilize the sort of industrial capacity of the power, you know, get the stuff done fast. And again, this is the sort of thing, you know, we haven't faced in a while. But, you know, during the Cold War, like, we really, there was this sort of intense system competition, right? Like East-West Germany was this, right? Like, West Germany, kind of like liberal democratic capitalism versus kind of, you know, communist state plan. And, you know, now it's obvious that the sort of, you know, the free world

• 47:41Speaker 01

  would win. But, you know, even, even as late as like 61, you know, Paul Samuelson was predicting that the Soviet Union would, would outgrow the United States because they were able to sort of mobilize industry better. And so, yeah, there's some people who, you know, they shippost about loving America by day, but then in private, they're betting against America. They're betting against the liberal order. And I think, I basically just think it's a bad bet. And the reason I think it's a bad bet is I think this stuff is just really possible in the US. And so to make it possible in the US, there's some amount that we have to get our act together, right? So I think there's basically two paths to doing it in the US. One is you've just got to be willing to do natural gas. And there's ample natural gas, right? You put your cluster in West Texas, you put it in, you know, southwest Pennsylvania by the, you know, Marcella shale. 100 gigawatt cluster, super easy. 100 gigawatt cluster, also pretty doable.

• 48:24Speaker 01

  You know, I think, you know, natural gas production in the United States is, you know, almost doubled in a decade. If you do that, you know, one more time over the next, you know, seven years or whatever, you know, you could power multiple trillion dollar data centers. But the issue there is, you know, a lot of people have sort of these, made these climate commitments. And not just government, it's actually the private companies themselves, right? The Microsoft, the Amazons, and so on. They have these climate commitments, so they won't do natural gas. And, you know, I admire the climate commitments, but I think at some point, you know, the national interest and national security kind of is more important. The other path is like, you know, you can do this sort of green energy mega projects, right? You do the solar and the batteries and the, you know, the SMRs and, and geothermal. But if we want to do that, there needs to be sort of a sort of broad deregulatory push,

• 49:07Speaker 01

  right? So like, you can't have permitting take a decade, right? So you got to reform FERC. You got to like have, you know, blanket NICA, NEPA exemptions for this stuff. You know, there's like inane state level regulations, you know, that are like, yeah, you could build, you know, you can build the solar panels and batteries next to your data center, but it'll still take years because, you know, you actually have to hook it up to the state electrical grid. You know, and you have to like use governmental powers to create rights of way to kind of like, you know, have multiple clusters and connect them, you know, and have thick cables, basically. And so, look, I mean, ideally, we do both, right? Ideally, we do natural gas and the broad deregulatory agenda. I think we have to do at least one. And then I think this possible stuff is, is just possible in the United States.

• 49:44Dwarkesh Patel?

  Yeah. I think a good analogy for this, by the way, before the conversation, I was reading, there's a good book about World War II industrial mobilization in the United States called Freedom's Forge.

• 49:54Speaker 01

  Yeah.

• 49:56Dwarkesh Patel?

  I guess when we think back on that period, especially if you're from, if you read like the Patrick Hoss and Fast and the Progress Eddie stuff, it's like the hat seat capacity back then and people just got shit done, but now it's a clusterfuck.

• 50:09Speaker 01

  Wasn't it all the case?

• 50:10Dwarkesh Patel?

  No, so it was really interesting. So you have people who are from the Detroit auto industry side, like Knutson, who are running mobilization for the United States and they were extremely competent.

• 50:21Speaker 01

  Yeah.

• 50:22Dwarkesh Patel?

  But then at the same time, you had labor organization agitation, which is actually very analogous to the climate pledges and climate change concern we have today.

• 50:30Speaker 01

  Yeah.

• 50:31Dwarkesh Patel?

  Where they would have these strikes while, literally into 1941, that would cost millions of man hours worth of time when we're trying to make tens of millions, sorry, tens of thousands of planes a month or something. And they would just debilitate factories before, you know, trivial like pennies on the dollar kind of concessions from capital. And it was concerns that, oh, the auto companies are trying to use the pretext of a potential war to actually prevent paying labor that money deserves. And so what climate changes today, like you'd think, ah, fuck, America's fucked. Like, we're not going to be able to build this shit. Like, if you look at NEPA or something. But I didn't realize how debilitating labor was.

• 51:16Speaker 01

  Well, it wasn't just that, right? It was just, you know, before, at the, you know, it's sort of like 39 or whatever, the American military was in total shambles, right? You read about it and it reads a little bit like, you know, the German military today, right? It's like, you know, military expenditures, I think, were less than 2% of GDP. You know, all the European countries had gone, even in peacetime, you know, like above 10% of GDP, sort of this like rapid mobilization. There was nothing, you know, like we're making kind of like no planes, there's no military contracts, everything had been starved during the Great Depression. But there was this latent capacity. And, you know, at some point, the United States got their act together. I mean, the thing I'll say is, I think, you know, the supply is sort of the other way around, too, to basically to China, right? And I think sometimes people are, you know, they kind of count them out a little bit and they're like the export controls and so on. And, you know, they're able to make seven nanometer chips now. I think there's a question of like, how many could they make?

• 52:01Speaker 01

  But, you know, I think there's at least a possibility that they're going to be able to mature that ability and make a lot of seven nanometer chips. And there's a lot of latent industrial capacity in China and they are able to like, you know, build a lot of power fast. And maybe that isn't activated for AI yet, but at some point, you know, the same way the United States and like, you know, a lot of people in the US and the United States government is going to wake up, you know, at some point the CCP is going to wake up.

• 52:21Dwarkesh Patel?

  Yep.

• 52:22Speaker 01

  Okay.

• 52:23Dwarkesh Patel?

  Going back to the question of presumably companies, are they blind to the fact that there's going to be some sort of, well, okay, so they realize that there's going, they realize scaling is a thing, right? Obviously, their whole plans are contingent on scaling. And so they understand that we're going to be in 2028 building a 10 gigawatt data centers. And at this point, the people who can keep up are big tech, just potentially at like the edge of their capabilities, then sovereign wealth fund funded things, and also big major countries like America, China, whatever. So what's their plan? If you look at like these AI labs, what's their plan given this landscape? Do they not want the leverage of having, being in the United States?

• 53:07Speaker 01

  I mean, I think, I don't know, I think, I mean, one thing the Middle East does offer is capital, but it's like, America has plenty of capital, right? It's like, you know, we have trillion dollar companies, like what are these Middle Eastern states? They're kind of like trillion dollar oil companies. We have trillion dollar companies and we have very deep financial markets. And it's like, you know, Microsoft could issue hundreds of billions of dollars of bonds and they can pay for these clusters. I mean, look, I think another argument being made, and I think it's worth taking seriously, is an argument that, look, if we don't work with the UAE or with these Middle Eastern countries, they're just going to go to China, right? And so, you know, they're going to build data centers, they're going to pour money into AI regardless. And if we don't work with them, you know, they'll just support China. And, look, I mean, I think, I think there's

• 53:50Speaker 01

  some merit to the argument and in the sense that I think we should be doing basically benefit sharing with them, right? I think we should talk about this later, but I think basically sort of on the road to AGI, there should be kind of like two tiers of coalitions. It should be this sort of narrow coalition of democracies, that's sort of the coalition that's developing AGI. And then there should be a broader coalition where we kind of go to other countries, including dictatorships, and we're willing to offer them, you know, we're willing to offer them some of the benefits of AI, some of the sharing. So it's like, look, if the UAE wants to use AI products, if they want to run meta recommendation engines, if they want to run, you know, like last generation models, that's fine. I think by default, they just like wouldn't have had this seat at the AGI table, right? And so it's like, yeah, they have some money, but a lot of people have money. And, you know, the only reason they're getting this sort of coursey at the AGI table, the only reason

• 54:34Speaker 01

  we're giving, these dictators will have this enormous amount of leverage over this extremely national security relevant technology is because we're, you know, we're kind of getting them excited and offering it to them.

• 54:48Speaker 01

  You know, I think the other, yeah.

• 54:49Dwarkesh Patel?

  Who like who specifically is doing this? Like just the companies who are going there to fundraise are like this is the AGI is happening and you can fund it or you can't.

• 54:56Speaker 01

  It's been reported that, you know, there's been reported that, you know, Sam is trying to raise, you know, seven trillion or whatever for a chip project. And, you know, it's unclear how many of the clusters will be there and so on, but it's, you know, definitely, definitely stuff is happening. I mean, look, I think another reason I'm a little bit at least suspicious of this argument of like, look, if the US doesn't work with them, they'll go to China is, um, you know, I've heard, heard from multiple people, um, and this wasn't, you know, from my time at OpenAI and I haven't seen the memo, but I have heard from multiple people that, um, you know, at some point several years ago, OpenAI leadership had sort of laid out a plan to fund and sell AGI by starting a bidding war between the governments of, you know, the United States, China and Russia. And so, you know, it's kind of surprising to me that they're willing to sell AGI to the Chinese and Russian governments.

• 55:37Speaker 01

  But also there's something that sort of feels a bit eerily familiar about kind of starting this bidding war and then kind of like playing them off each other. And while, you know, if you don't do this, China will do it. So, anyway.

• 55:47Dwarkesh Patel?

  Interesting. Okay. So that's pretty fucked up, but given that that's, okay. So suppose that you're right about, we ended up in this place because we got, the way one of our friends put it is that the Middle East has like no other place in the world, billions of dollars or trillions of dollars up for persuasion and.

• 56:07Speaker 01

  True.

• 56:08Dwarkesh Patel?

  And what would you have at the form of Sam Alvin?

• 56:10Speaker 01

  Less accountability than like, you know, the Microsoft board. It's only, it's only the dictator, right?

• 56:13Dwarkesh Patel?

  Yeah. Okay. But, but so let's say you're right that you shouldn't have gotten them excited about AGI in the first place. But now we're in a place where they are excited about AGI.

• 56:21Speaker 01

  Yeah.

• 56:22Dwarkesh Patel?

  And they're like, fuck, we want us to have GPT-5. We're going to be off building superintelligence. This Adam's for peace thing doesn't work for us. Um, and if you're in this place,

• 56:33Dwarkesh Patel?

  are, don't they already have the leverage? Aren't you like, and you might as well just.

• 56:35Speaker 01

  I don't think, I think the UAE on its own is not competitive, right? It's like, I mean, they're, they're already export controlled. Like, you know, we're not, you know, there's like, you're not actually supposed to ship Nvidia chips over there, right? You know, it's not like they have any of the leading AI labs, you know, it's like, they have money, but you know, it's actually hard to just translate money into like.

• 56:49Dwarkesh Patel?

  But the other things you've been saying about laying out your vision is very much, there's this almost industrial process of you put in the compute and then you put in the algorithms.

• 56:58Speaker 01

  Yes.

• 56:58Dwarkesh Patel?

  You add that up and you get AGI on the other end.

• 57:02Speaker 01

  Yes.

• 57:02Dwarkesh Patel?

  If it's something more like that, then the case for somebody being able to catch up rapidly seems more compelling than if it's some bespoke.

• 57:09Speaker 01

  Well, well, if they can steal the algorithms and if they can steal the weight. And that's really, that's really where sort of, I mean, we should talk about this. This is really important. And I think. Yeah.

• 57:19Dwarkesh Patel?

  So like right now, how easy would it be for for an actor to steal the things that are like, not the things that are released about Scarlett Johansson's voice, but the the RL things you're talking about, the Un-Hobblings?

• 57:31Speaker 01

  I mean, I mean, all extremely easy, right? You know, I, you know, DeepMind even like, you know, they they don't make a claim that it's hard, right? DeepMind put out there like whatever frontier safety, something, and they like lay out security levels and they like, you know, security level zero to four and four is the one resistant to state actors. And they say we're at level zero, right? And then, you know, I mean, just recently, there was like an indictment of a guy who just like stole the code, a bunch of like really important AI code and went to China with it. And, you know, all he had to do to steal the code was that, you know, copy the code and put it into Apple Notes and then exported it as PDF. And that got past their monitoring, right? And Google is the best security of any of the AI labs, probably because they have the, you know, the Google infrastructure. I mean, I think, I don't know, roughly, I would think of this as like, you know, security of a startup, right? And like, what does security of a startup look

• 58:13Speaker 01

  like, right? You know, it's not that good. It's it's easy to steal.

• 58:17Dwarkesh Patel?

  So even if that's the case.

• 58:18Speaker 01

  Yeah.

• 58:19Dwarkesh Patel?

  A lot of your posts is making the argument that, oh, you know, why are we going to get the intelligence explosion? Because if we have somebody with the intuition of an Alec Radford to be able to come up with all these ideas.

• 58:30Speaker 01

  Yeah.

• 58:30Dwarkesh Patel?

  That intuition is extremely valuable and you scale that up. But if it's a matter of these, if it's just in the code, that like, if it's just the intuition, then that's not going to be just in the code, right? And also because of export controls, these countries are going to have slightly different hardware, you're going to have to make different trade offs and probably rewrite things to be able to be compatible with that. Including all these things, is it just a matter of getting the right pen drive and you plug it into the gigawatt data center next to the Three Gorges Dam and then you're off to the races?

• 59:04Speaker 01

  I mean, look, there's a few different things, right? So one one threat model is just stealing the weights themselves. And the weights one is sort of particularly insane, right? Because they can just like steal the literal like end product, right? Just like make a replica of the atomic bomb and then they're just like ready to go. And I think that one just is, you know, extremely important around the time we have AGI and superintelligence, right? Because it's, you know, China can build a big cluster. By default, we'd have a big lead, right? Because we have the better scientists, but we make the superintelligence, they just steal it, they're off to the races. Weights are a little bit less important right now, because who cares if they steal the GPT-4 weights, like whatever. And so, we still have to get started on weight security now, because, look, if we think AGI 27, this stuff is going to take a while. And it's not just going to be like, oh, we do some access control.

• 59:49Speaker 01

  If you actually want to be resistant to Chinese espionage, it needs to be much more intense. The thing, though, that I think people aren't paying enough attention to is the secrets, as you say. And I think the compute stuff is sexy, we talk about it, but I think people underrate the secrets because they're, I think, half in order of magnitude a year, just by default, sort of algorithmic progress, that's huge. If we have a few-year lead by default, that's 10, 30x, 100x bigger cluster if we protected them. And then there's this additional layer of the data wall, right? And so, we have to get through the data wall. That means we actually have to figure out some sort of basic new paradigm, sort of the alpha go step two, right? Alpha go step one is learns from human imitation. Alpha go step two is the self-play RL.

• 1:00:32Speaker 01

  And everyone's working on that right now, and maybe we're going to crack it. And if China can't steal that, then they're stuck. If they can't steal it, they're off to the races.

• 1:00:44Dwarkesh Patel?

  But whatever that thing is, is it like, literally, I can write down on the back of a napkin? Because if it's that easy, then why is it that hard for them to figure it out? And if it's more about the intuitions, then don't you just have to hire Alec Bradford? What are you copying now?

• 1:00:56Speaker 01

  Well, I think there's a few layers to this, right? So I think at the top is kind of like, sort of the fundamental approach, right? And sort of like, I don't know, on pre-training, it might be unsupervised learning, next token protection, train on the entire internet. You actually get a lot of juice out of that already. That one's very quick to communicate. Then there's a lot of details that matter. And you were talking about this earlier, right? It's like, probably the way that thing people are going to figure out is going to be somewhat obvious, or there's going to be some kind of like, clear, not that complicated thing that'll work. But there's going to be a lot of details to getting that right.

• 1:01:28Dwarkesh Patel?

  But if that's true, then again, why are we even, why do we think that getting state-level security in these startups will prevent China from catching? If it's just like, oh, we know some sort of self-play RL we require to get past a data wall. And if it's as easy as you say in some fundamental sense.

• 1:01:43Speaker 01

  I mean, again.

• 1:01:44Dwarkesh Patel?

  But it's gonna be solved by 2027, you say, like, right? It's like, not that hard.

• 1:01:47Speaker 01

  I just think, you know, the US and the sort of, I mean, all the leading AI labs in the United States, and they have this huge lead. I mean, by default, you know, China actually has some good LLMs. Why do they have good LLMs? They're just using the sort of open source code, right? You know, Llama or whatever. And so the, um, the, um, I think people really underrate the sort of, both the sort of divergence on algorithmic progress and the lead the US would have by default. Because by, you know, all this stuff was published until recently, right? Like, Chinchilla scaling laws were published, you know, there's a bunch of MOE papers. There's, you know, transformers and, you know, all that stuff was published. And so that's why open source is good. That's why China can make some good models. That stuff is now, I mean, at least they're not publishing it anymore. And, you know, if we actually kept it secret, it would be this huge edge. To your point about sort of like some tacit knowledge now like Bradford, you know, there's, there's another layer at the bottom that is something about like, you know,

• 1:02:30Speaker 01

  large-scale engineering work to make these big training runs work. I think that is a little bit more tacit knowledge. So I think that, but I think China will be able to figure that out. That's like sort of engineering schlap. They're gonna figure out.

• 1:02:39Dwarkesh Patel?

  But why can't they figure that out, but not how to get the RL thing working?

• 1:02:44Speaker 01

  I mean, look, I don't know, uh, Germany during World War II, you know, they went down the wrong path. They did heavy water and that was wrong. And, and there's actually, there's an amazing anecdote in, in the making of the atomic bomb on this, right? So, so secrecy is actually one of the most contentious issues, you know, early on as well. And, you know, part of it was sort of, you know, Zilard or whatever really thought, you know, this sort of nuclear chain reaction was possible. And so an atomic bomb was possible and he went around and he was like, this is gonna be of enormous strategic importance, military importance. And a lot of people didn't believe it, or they were kind of like, well, maybe this is possible, but you know, I'm gonna act as it's not possible. And, you know, science should be open and all these things. And, um, anyway, and so these early days, um, so there had been some sort of incorrect measurements made on, on graphite as a moderator.

• 1:03:28Speaker 01

  And that Germany had, and so they thought, you know, graphite was not gonna work. We have to do heavy water. But then Fermi made some new measurements on graphite and they indicated that graphite would work. You know, this is really important. And then, you know, Zilard kind of assaulted Fermi with the kind of another secrecy appeal. And Fermi was just kind of, he was pissed off, you know, had a temper tantrum, you know, he was like, he thought it was absurd, you know, like, come on, this is crazy. But, you know, you know, Zilard persisted. I think they roped in another guy, Pegram, and then Fermi didn't publish it. And that was just in time because Fermi not publishing it meant that the Nazis didn't figure out graphite would work. They went down this path of heavy water and that was the wrong path. That was why this, you know, this is a key reason why this sort of German

• 1:04:11Speaker 01

  project didn't work out. They were kind of way behind. And, you know, I think we face a similar situation on are we, are we just going to instantly leak the sort of how do we get past the data wall? What's the next paradigm? Or are we not?

• 1:04:23Dwarkesh Patel?

  So, and the reason this would matter is if there's like being one year ahead would be a huge advantage in the world where it's like you deploy AI over time and they're just like, ah, they're going to catch up anyway. I mean, I interviewed Richard Rhodes, the guy who wrote making an atomic bomb.

• 1:04:39Speaker 01

  Yeah.

• 1:04:40Dwarkesh Patel?

  And one of the anecdotes he had was when, so they realized America had the bomb, obviously, we dropped it in Japan.

• 1:04:46Speaker 01

  Yeah.

• 1:04:47Dwarkesh Patel?

  And Beria goes, the guy who ran the NKVD.

• 1:04:50Speaker 01

  Yeah.

• 1:04:51Dwarkesh Patel?

  And which is a famously ruthless guy, just evil. And he goes to, I forgot the name, but the guy, the Soviet scientist who was running their version of the Manhattan Project who says, comrade, you will get us the American bomb.

• 1:05:03Speaker 01

  Yeah.

• 1:05:03Dwarkesh Patel?

  And the guy says, well, listen, their implosion device actually is not optimal. We should make it a different way. And Beria says, no, you will get us the American bomb or your family will be camp dust. But the, the thing that's relevant about that anecdote is actually the Soviets would have had a better bomb if they hadn't copied the American design, at least initially. And which suggests that often in history, this is something that's not just for the Manhattan Project, but there's this pattern of parallel invention where because the tech tree implies that the certain thing is next, in this case, a self-play or whatever, then people are just like working on that and like people are going to figure out around the same time. There's not, there's not going to be that much gap in who gets

• 1:05:45Dwarkesh Patel?

  it first. Was it like famously the bunch of people were invented something like the light bulb around the same time and so forth?

• 1:05:50Speaker 01

  Yeah.

• 1:05:51Dwarkesh Patel?

  So, but is it just that like, yeah, that might be true, but it'll, the one year or the six months or whatever.

• 1:05:55Speaker 01

  Two years makes all the difference.

• 1:05:57Dwarkesh Patel?

  I don't know if it'll be two years though, like, right?

• 1:05:58Speaker 01

  I mean, I actually, I mean, I actually think if we lock down the labs, we have, we have much better scientists, we're way ahead, it would be two years. But even, I think even, I think, I think whether you, I think, yeah, I think even six months, a year would make a huge difference. And this gets back to the sort of intelligence. Like a year might be the difference between, you know, a system that's sort of like human level and a system that is like vastly superhuman, right? It might be like five, five ooms, you know, I mean, even even on the current pace, right? We went from, you know, I think, you know, on the math benchmark recently, right? Like, you know, three years ago on the math benchmark, we, um, you know, that was, you know, this is sort of really difficult high school competition math problems, you know, we were at, you know, a few percent couldn't solve anything. Now it's solved. Um, and that was sort of at the normal pace of AI progress. You didn't have sort of a billion super intelligent resources, researchers. So like a year is a huge

• 1:06:42Speaker 01

  difference. And then particularly after superintelligence, right? Once this is applied to sort of lots of elements of R and D, once you get the sort of like industrial explosion with the robots and so on. Um, you know, I think a year, you know, a couple of years might be kind of like decades worth of technological progress. And might, you know, again, it's like Gulf War one, right? 20, 30 years of technological lead, totally decisive. You know, I, I think it really matters. The other reason it really matters is, um, you know, suppose, suppose they steal the weight, suppose they steal the algorithms and, you know, they're close on our tails. Suppose we still pull out ahead, right? We just kind of, we, we're a little bit faster, you know, we're three months ahead. I think the sort of like world in which we're really neck and neck, you know, we only have a three month lead are incredibly dangerous, right? And we're in this fever struggle where

• 1:07:23Speaker 01

  like if they get ahead, they get to dominate, you know, sort of, maybe they get a decisive advantage. They're building clusters like crazy. They're, they're willing to throw all caution to the wind. We have to keep up. There's some crazy new WMDs popping up. And then we're going to be in the situation where it's like, you know, crazy new military technology, crazy new WMDs, you know, like deterrence, mutually disturbed instruction, like keeps changing, you know, every few weeks. And it's like, no, completely unstable volatile situation. That is incredibly dangerous. So it's, I think, I think, you know, both, both from just the technologies are dangerous from the alignment point of view, you know, I think it might be really important during the intelligence explosion to have the sort of six month, uh, you know, wiggle room to be like, look, we're going to like dedicate more compute to alignment during this period because we have to get it right. We're feeling uneasy about how it's going. And, um, so I think in some sense of

• 1:08:08Speaker 01

  like, one of the most important inputs to whether we will kind of destroy ourselves or whether we will get through this just incredibly crazy period is whether we have that buffer.

• 1:08:19Dwarkesh Patel?

  Why? So before we go further object level in this, I think it's very much worth noting that almost nobody, at least nobody I talked to, thinks about the geopolitical implications of AI. And I think I have some object level disagreements I will get into, but, or at least things I want to iron out. I may not disagree in the end. But the basic premise that obviously, if you keep scaling and obviously if people realize that this is where intelligence is headed, it's not just going to be like the same old world where like what model are we deploying tomorrow? And what is the latest? Like, people on Twitter are like, oh, the GPT-40 is going to shake your expectations or whatever. You know, COVID was really interesting

• 1:09:04Dwarkesh Patel?

  because before a year or something, when March 2020 hit, we, it became clear to the world, like, presidents, CEOs, media, average person, there's other things happening in the world right now, but the main thing we as a world are dealing with right now is COVID.

• 1:09:21Speaker 01

  Soon on AGI.

• 1:09:22Dwarkesh Patel?

  Yeah.

• 1:09:24Dwarkesh Patel?

  Okay. And then so-

• 1:09:25Speaker 01

  This is the quiet period. You know, if you want to go on vacation, you know, you want to like, you want to, yeah, you want to have, you know, maybe like now is the last time you can have some kids. You know, my girlfriend sometimes complains, you know, that I, you know, when I'm like, you know, off doing work or whatever. And she's like, I'm not spending time with her. She's like, you know, she threatens to replace me with like, you know, GPT-6 or whatever. And I'm like, GPT-6 will also be too busy doing AI research.

• 1:09:51Dwarkesh Patel?

  Okay, anyway, so let's answer the question of why, why aren't other people talking national security?

• 1:09:56Speaker 01

  I made this mistake with COVID, right? So I, you know, February of 2020, and I, you know, I thought just it was going to sweep the world and all the hospitals would collapse and it would be crazy. And then, and then, you know, and then it'd be over. And a lot of people thought this kind of the beginning of COVID. They shut down their offices a month or whatever. I think the thing I just really didn't price in was the societal reaction, right? And, and within weeks, you know, Congress spent over 10% of GDP on like COVID measures, right? The entire country was shut down. It was crazy.

• 1:10:25Speaker 01

  And so, I don't know, I didn't price it in with COVID sufficiently.

• 1:10:29Speaker 01

  I don't know, why do people underrate it? I mean, I think there's a sort of way in which being kind of in the trenches actually kind of, I think, gives you a less clear picture of the trend lines. You actually have to zoom out that much, only like a few years, right? But you're in the trenches, you're like trying to get the next model to work, you know, there's always something that's hard. You know, for example, you might underrate algorithmic progress because you're like, ah, things are hard right now, or you know, data wall or whatever. But you know, you zoom out just a few years and you actually try to like count up how much algorithmic progress made in the last few years and it's enormous. But I also just don't think people think about this stuff. Like, I think smart people really underrate espionage, right? And, you know, I think part of the security issue is I think people don't realize like how intense state-level espionage can be, right?

• 1:11:12Speaker 01

  Like, you know, this Israeli company had software that could just zero-click hack any iPhone, right? They just put in your number and then it's just like straight download of everything, right? Like, the United States infiltrated and air-gapped atomic weapons program, right? Wild, you know, like- Are you talking about SARS, then? Yeah, yeah, yeah. You know, the, you know, intelligence agencies have just stockpiles of zero days, you know, when things get really hot, you know, I don't know, maybe they'll send special forces, right, to like, you know, get, go to the data center or something that's, you know, or, you know, I mean, China does this, they threaten people's families, right? And they're like, look, if you don't cooperate, if you don't give us the intel. There's a good book, you know, along the lines of the Gulag or the Belov, you know, the Inside the Aquarium, which is by a Soviet GRU defector.

• 1:11:57Speaker 01

  GRU was like military intelligence. Ilya recommended this book to me. And, you know, I think reading that, I was just kind of like shocked that I have intense sort of state-level espionage. The whole book was about like, they go to these European countries and they try to like get all the technology and recruit all these people to get the technology. But, I mean, yeah, maybe one anecdote, you know, so when, so this spy, you know, this eventual defector, you know, so he's being trained, he goes to the kind of GRU spy academy. And so then to graduate from the spy academy, sort of before you're sent abroad, you kind of had to pass a test to show that you can do this. And the test was, you know, you had to, in Moscow, recruit a Soviet scientist and recruit them to give you information, sort of like you would do in the foreign country.

• 1:12:40Speaker 01

  But of course, for whomever you recruited, the penalty for giving away sort of secret information was death. And so to graduate from the Soviet spy, this GRU spy academy, you had to condemn a countryman to death. Mm-hmm. States do this stuff.

• 1:12:56Dwarkesh Patel?

  I started reading the book on we have a side in the series.

• 1:12:59Speaker 01

  Yeah.

• 1:13:00Dwarkesh Patel?

  And I was actually wondering, the fact that you use this anecdote.

• 1:13:03Speaker 01

  Yeah.

• 1:13:04Dwarkesh Patel?

  And then you're like, and a book recommended by Ilya, is this some sort of, is this some sort of Easter egg?

• 1:13:13Dwarkesh Patel?

  We'll leave that for our next exercise with the reader.

• 1:13:16Dwarkesh Patel?

  Okay, so.

• 1:13:17Speaker 01

  The beatings will continue until the morale improves.

• 1:13:22Dwarkesh Patel?

  So suppose that we live in the world in which the secrets are locked down. But China still realizes that this progress is happening in America.

• 1:13:34Dwarkesh Patel?

  In that world, especially if they realize, and I guess it's a very interesting open question.

• 1:13:39Speaker 01

  Probably won't be locked down.

• 1:13:40Dwarkesh Patel?

  Okay, but suppose.

• 1:13:41Speaker 01

  We're probably gonna live in the bad world.

• 1:13:42Dwarkesh Patel?

  Yeah.

• 1:13:43Speaker 01

  It's gonna be really bad.

• 1:13:44Dwarkesh Patel?

  Hmm. Why are you so confident that they won't be locked down?

• 1:13:47Speaker 01

  I mean, I'm not confident that it won't be locked down, but I think it's just, it's not happening.

• 1:13:51Dwarkesh Patel?

  Hmm. And but, and so tomorrow, the lab leaders get the message. How hard, like, what do they have to do? They get the more security guards, they like air gap the, what do they do?

• 1:14:02Speaker 01

  So again, I think, I think basically it's, you know, I think people, there's kind of like two, two reactions there, which is like, it's, you know, we're already secure, you're not. And there's, you know, fatalism, it's impossible. And I think the thing you need to do is you kind of got to stay ahead of the curve of basically how EGI pills the CCP.

• 1:14:19Dwarkesh Patel?

  Yeah.

• 1:14:19Speaker 01

  Right? So like, right now, you've got to be resistant to kind of like normal economic espionage.

• 1:14:25Speaker 01

  They're not, right? I mean, I probably wouldn't be talking about the stuff that the labs were, right? Because I wouldn't want to wake them up more, the CCP, but they're not, you know, this is like, this stuff is like really trivial for them to do right now. I mean, it's also, anyway, so they're not resistant to that. I think it would be possible for a private company to be resistant to it, right? So, you know, both of us have, you know, friends in the kind of like quantitative trading world, right? And, you know, I think actually those secrets are shaped kind of similarly where it's like, you know. You know, they've said, you know, yeah, if I got on a call for an hour with somebody from a competitor firm, I could, most of our alpha would be gone. And that's sort of like, that's the like list of details of like really how to, how to make.

• 1:14:59Dwarkesh Patel?

  You're gonna have to worry about that pretty soon. You're gonna have to worry about that pretty soon.

• 1:15:03Speaker 01

  Yeah, yeah, yeah. Well, anyway, and so, so all the alpha could be gone, but in fact, their alpha persists, right? And, you know, often, often for many years and decades. And so this doesn't seem to happen. And so I think there's like, you know, I think there's a lot you could go if you went from kind of current startup security. You know, you just got to look through the window and you can look at the slides. You know, it's, it's a kind of like, you know, you know, good private sector security hedge funds, you know, the way Google treats, you know, customer data or whatever.

• 1:15:27Speaker 01

  That'd be good right now. The issue is, you know, basically the CCP will also get more AGI built. And

• 1:15:35Speaker 01

  at some point, we're going to face kind of the full force of, you know, the Ministry of State Security. And again, you're talking about smart people underwriting espionage and the sort of insane capabilities of states. I mean, this stuff is wild, right? You know, they can get like, you know, there's papers about, you know, you can find out the location of like where you are on a video game map just from sounds, right? Like states can do a lot with like electromagnetic emanations, you know, like, you know, at some point, like you got to be working from a SCAP. Like your cluster needs to be air gapped and basically be a military base. It's like, you know, you need to have, you know, intense kind of security clearance procedures for employees. You know, they have to be like, you know, all their shit is monitored, you know, they're, you know, they basically have security guards, you know, it's, you know, you, you can't use any kind of like, you know, other dependencies. It's all got to be like intensely vetted and you know, all your hardware has to be intensely vetted.

• 1:16:19Speaker 01

  And

• 1:16:21Speaker 01

  you know, I think basically if, if they actually really face the full force of state level espionage, I don't really think this is a thing private companies can do. Both, I mean empirically, right? Like, you know, Microsoft recently had executives emails hacked by Russian hackers and, you know, government emails they posted hacked by government actors. But also, um, you know, it's basically, there's just a lot of stuff that only kind of, you know, the people behind the security countries know and only they deal with. And so, you know, I think to actually kind of resist the sort of full force of espionage, you're going to need the government. Anyway, so I think basically we could, we could do it by always being ahead of the curve. I think we're just going to always be behind the curve. And I think, you know, maybe unless we get the sort of government project.

• 1:16:57Dwarkesh Patel?

  Okay, so going back to the naive perspective of we're very much coming at this from there's going to be a race and the CCP, we must win. Um, and listen, I understand like bad people are in charge of the Chinese government, like the CCP and everything. But just stepping back in a sort of galactic perspective, humanity is developing AGI. And do we want to come at this from the perspective of we need to beat China to this? Our, our super intelligent Jupiter brain descendants won't know who China like China will be something like distant memory that they have America too. So shouldn't it be a more the initial approach, just come to them like, listen, this is super intelligence. This is something like we come from a cooperative.

• 1:17:41Dwarkesh Patel?

  Uh, perspective, why, why immediately sort of rush into it from a hawkish competitive perspective?

• 1:17:46Speaker 01

  I mean, look, I mean, one thing I want to say is like, a lot of the stuff I talk about in the series is, you know, is sort of primarily, you know, descriptive. Right. And so I think that on the China stuff, it's like, you know, yeah, in some ideal world, you know, we, we, you know, it's just all, you know, merry go round and cooperation. But again, it's sort of, I think, I think people wake up to AGI. I think the, the issue particular on sort of like, can we make a deal? Can we make an international treaty? I think it really relates to sort of what is the stability of sort of international arms control agreements, right? And so we did very successful arms control on nuclear weapons in the eighties, right? And the reason it was successful is because the sort of new equilibrium was stable, right? So you take go down from, you know, whatever 60,000 nukes to 10,000 nukes.

• 1:18:26Speaker 01

  You know, when you have 10,000 nukes, um, you know, basically breakout, breakout doesn't matter that much, right? Suppose the other guy now tried to make 20,000 nukes. Well, it's like, who cares, right? You know, like it's still mutually assured destruction. Suppose rogue state kind of went from zero nukes to one nukes. It's like, who cares? We still have way more nukes than you. I mean, it's still not ideal for destabilization, but it, you know, it'd be very different if the arms control agreement had been zero nukes. Right. Cause if had been zero nukes, then it's just like one rogue state makes one nuke. The whole thing is destabilized, breakout is very easy. You know, your, your adversary state starts making nukes. And so basically when, when you're going to sort of like very low levels of arms, or when you're going to kind of, and you're in a sort of very dynamic technological situation.

• 1:19:05Speaker 01

  Arms control is really tough because breakout is easy. You know, there's, there's, I mean, there's some other sort of stories about this in sort of like 1920s, 1930s. You know, it's like, you know, all the European states had done disarmament and Germany was, was kind of did this like crash program to build the Luftwaffe. And that was able to like massively destabilize things because not that, you know, they were the first, they were able to like pretty easily build kind of a modern, you know, air force. Cause the others didn't really have one and that, you know, that really destabilized things. And so I think the issue with AGI and superintelligence is the explosiveness of it. Right? So if you have an intelligence explosion, if you're able to go from kind of AGI to superintelligence, if that superintelligence is decisive, like either, you know, like a year after, because you've developed some crazy new WMD. Or because you have some like, you know, super hacking ability that lets you, you kind of, um, you know, completely deactivate the sort of enemy arsenal.

• 1:19:52Speaker 01

  That means like suppose, suppose you're trying to like put in a break, you know, like we both, we're both going to like cooperate and we're going to go slower, you know, on the cusp of AGI or whatever. There is going to be such an enormous incentive to kind of race ahead, to break out. And we're just going to do the intelligence explosion. If we can get three months ahead, we win. I think that makes it basically, I think any sort of arms control agreement that comes as a situation where it's close, very unstable.

• 1:20:15Dwarkesh Patel?

  That's really interesting. This is very analogous to kind of a debate I had with Rose on the podcast where he argued for nuclear disarmament. But if some country tries to break out and starts developing nuclear weapons, the six months or whatever that you would get is enough to get international consensus and invade the country and prevent them from getting nukes. And I thought that was sort of, that's not a stable equilibrium.

• 1:20:40Speaker 01

  It just seems really tough. Yeah.

• 1:20:42Dwarkesh Patel?

  But so on this, right? So like, maybe it's a bit easier because you have AGI. And so like you can monitor the other person's cluster or something like data centers, you can see them from space. Actually, you can see the energy draw they're getting. There's a lot of things, as you were saying, we give you, there's a lot of ways to get information from an environment if you're really dedicated. And also because unlike a nuke, the data centers are nukes, you have obviously the submarines, planes, you have bunkers, mountains, whatever. You have them so in different places, a data center that your 100 gigawatt data center, we can blow that shit up if you're like, we're concerned. Right? Like just some cruise missile or something.

• 1:21:17Speaker 01

  Yeah.

• 1:21:17Dwarkesh Patel?

  It's like very vulnerable to sabotage.

• 1:21:19Speaker 01

  That gets to the sort of, I mean, that gets to the sort of insane vulnerability of this period post-superintelligence, right? Because basically, I think, so you have the intelligence explosion, you have these like vastly superhuman things on your cluster. But you're like, you haven't done the industrial explosion yet. You don't have your robots yet. You haven't kind of, you haven't covered the desert in like robot factories yet. And that is the sort of crazy moment where, you know, say, say the United States is ahead, the CCP is somewhat behind. There's actually an enormous incentive for, for a strike, right? Because if they can take out your data center, they, you know, they know you're about to have just this commanding decisive lead. They know if we can just take out this data center, you know, then we can stop it. And, you know, they might get desperate. And, um, you know, I, so I think basically we're going to get into a position. It's, it's actually, I think it's going to be pretty hard to defend early on.

• 1:22:00Speaker 01

  I think we're basically going to be in a position where protecting data centers with like the threat of nuclear retaliation. It's like, maybe it sounds kind of crazy though, you know.

• 1:22:07Dwarkesh Patel?

  Is this the inverse of the eleitor? We gotta. The data centers.

• 1:22:12Speaker 01

  Nuclear deterrence for data centers. I mean, this is, you know, Berlin, you know, in the like late 50s, early 60s.

• 1:22:17Dwarkesh Patel?

  Yeah.

• 1:22:17Speaker 01

  Both Eisenhower and Kennedy multiple times kind of made the threat of full on nuclear war against the Soviets if they tried to encroach on West Berlin. It's sort of insane. It's kind of insane that that went well. But basically, I think that's going to be the only option for the data centers. It's a terrible option. This whole scheme is terrible, right? Like being, being, being in this like neck and neck race sort of at this point is terrible. And, you know, it's also, you know, I think I have some uncertainty basically on how easy that decisive advantage will be. I'm pretty confident that if you have superintelligence, you have two years, you have the robots, you're able to get that 30 year lead. Look, then you're in this like go for one situation. You have your like, you know, millions or billions of like mosquito sized drones that can just take it out. I think there's even a possibility you can kind of get a decisive advantage earlier. So, you know, there's these stories, you know, about these as well about, you know, like colonization.

• 1:23:00Speaker 01

  And like the sort of 1500s where it was, uh, you know, these like a few hundred kind of Spaniards. We were able to like topple the Aztec empire, you know, a couple, I think a couple of other empires as well. You know, each of these had a few million people and it was not like godlike technological advantage. It was some technological advantage. It was, I mean, it was some amount of disease. And then it was kind of like cunning strategic play. And so I think there's, there's a, the possibility that even sort of early on, you know, you haven't gone through the full industrial explosion yet. You have superintelligence, but you know, you're able to kind of like manipulate the opposing generals, claim you're allying with them. Then you have, you have some, you know, you have sort of like some crazy new bioweapons. Maybe, maybe there's even some way to like pretty easily get a paradigm that like deactivates enemy nukes. Anyway, so I think this stuff could get pretty wild. Here's what I think we should do. I really don't want this volatile period.

• 1:23:44Speaker 01

  And so a deal with China would be nice. It's going to be really tough if you're in this unstable equilibrium.

• 1:23:52Speaker 01

  I think, basically, we want to get in a position where it is clear that the United States, that a sort of coalition of democratic allies will win. It's clear to the United States, it's clear to China, you know, that will require having locked down the secrets. That will require having built the 100 gigawatt cluster in the United States and having done the natural gas and doing what's necessary. And then when it is clear that the democratic coalition is well ahead, then you go to China and then you offer them a deal. And, you know, China will know they're going to win. This is going to be, they're very scared of what's going to happen. We're going to know we're going to win, but we're also very scared of what's going to happen because we really want to avoid this kind of like breakneck, breakneck race right at the end. And where things could really go awry. And, you know, and then, and then so then we offer them a deal. I think there's an incentive to come to the table. I think there's a sort of more stable arrangement you can do.

• 1:24:36Speaker 01

  It's a sort of an atoms for peace arrangement. And we're like, look, we're going to respect you. We're not, we're not going to like, we're not going to use super intelligence against you. You can do what you want. You're going to get your like, you're going to get your slice of the galaxy. We're going to like, we're going to benefit share with you. We're going to have some like compute agreement where it's like, there's some ratio of compute that you're allowed to have. And that's like enforced with like opposing AIs or whatever. And we're just not going to do, we're just not going to do this kind of like volatile sort of WMD arms race to the death.

• 1:25:03Dwarkesh Patel?

  We're good.

• 1:25:03Speaker 01

  And sort of it's like a new world order that's US-led, that's sort of democratic-led, but that respects China, lets them do what they want.

• 1:25:09Dwarkesh Patel?

  Okay, there's so much to, there's so much there. First on the galaxies thing, I think it's just a funny anecdote, so I want to kind of want to tell it. And this, we were at an event and I'm respecting Chatham House rules here. I'm not revealing anything about it, but we were talking to somebody, or Leopold was talking to somebody influential. Afterwards, that person asked the group, Leopold told me that he wants, he's not going to spend any money on consumption until he's ready to buy galaxies. And he goes, the guy goes, I honestly don't know if he meant galaxies like the brand of private plane galaxy or the physical galaxies.

• 1:25:49Dwarkesh Patel?

  And there was an actual debate, like he, he went away to the restroom and there was an actual debate among people who are very influential about, he can't amend galaxies. And the other people who knew you better would be like, no, he means galaxies.

• 1:26:02Speaker 01

  I mean the galaxies, I mean the galaxies. I mean, I think it'll be interesting. I mean, I, yeah, I think there's a, I mean, there's two ways to buy the galaxies. One is like, at some point, you know, it's like post superintelligence, you know, there's some crazy.

• 1:26:12Dwarkesh Patel?

  I love, okay, so what happens is he's on the, I'm knocking my ass off. I'm not even saying a thing. People are like having this debate. And then so Leopold comes back and the guy, somebody's like, oh, Leopold, we're having this debate about whether you meant, you want to buy the galaxy or you want to buy the other thing. And Leopold assumes they must mean not the private plane, the galaxy versus the actual galaxy. But do you want to buy the property rights of the galaxy or actually just send out the probes right now?

• 1:26:42Speaker 01

  Exactly.

• 1:26:46Dwarkesh Patel?

  Oh, my God. All right. Back to China.

• 1:26:53Dwarkesh Patel?

  There's a whole bunch of things I could ask about that plan about whether you're going to get credible promise. You will get some part of galaxies, whether they care about that.

• 1:27:01Speaker 01

  You have AIs to help you enforce stuff.

• 1:27:03Dwarkesh Patel?

  Okay, sure. We'll leave that aside. That's a different rabbit hole. The thing I want to ask is.

• 1:27:07Speaker 01

  But it has to be the thing we need. The only way this is possible is if we lock it down.

• 1:27:12Dwarkesh Patel?

  I see.

• 1:27:12Speaker 01

  If we don't lock it down, we are in this fever struggle. Greatest peril mankind will have ever seen.

• 1:27:19Dwarkesh Patel?

  So, but given the fact that during this period, instead of just taking their chances and they don't really understand how this AI governance scheme is going to work, where they're going to check, whether we had to actually get the galaxies.

• 1:27:31Speaker 01

  Yeah.

• 1:27:32Dwarkesh Patel?

  The data centers, they can't be built underground. They have to be built above ground. Taiwan is right off the coast of us. They need the ships from there.

• 1:27:38Speaker 01

  Yep.

• 1:27:38Dwarkesh Patel?

  Why aren't we just going to invade? Listen, we don't want, like, worst case scenario is they win the superintelligence, which they're on track to do anyways. Wouldn't this instigate them to either invade Taiwan or blow up the data center in Arizona or something like that?

• 1:27:51Speaker 01

  Yeah. I mean, look, I mean, you talked about the data center one and then, you know, you probably have to like threaten nuclear retaliation to protect that. They might also just blow it up. There's also maybe ways they can do it without sort of attribution, right? Like you pay.

• 1:28:01Dwarkesh Patel?

  Suxnet, Suxnet.

• 1:28:02Speaker 01

  Suxnet. Yeah. I mean, this is, I mean, this is part of, we'll talk about this later, but you know, I think, um, look, I think we need to be working on the Suxnet for the Chinese project, but the, um.

• 1:28:10Dwarkesh Patel?

  But by the way, for the audience.

• 1:28:11Speaker 01

  Taiwan, I mean, Taiwan, the Taiwan thing, the, um, you know, you know, I talk about, you know, AGI by, you know, 27 or whatever. Do you know about the like terrible 20s?

• 1:28:21Dwarkesh Patel?

  No.

• 1:28:22Speaker 01

  Okay, well, I mean, sort of in the sort of Taiwan watcher circles, people often talk about like the late 2020s is like maximum period of risk for Taiwan because sort of like, you know, military modernization cycles and basically extreme fiscal tightening on the military budget in the United States over the last decade or two has meant that sort of we're in this kind of like, you know, trough in the late 20s of like, you know, basically overall naval capacity. And, you know, that's sort of when China is saying they want to be ready. So it's already kind of like, it's kind of pitching, you know, there's some sort of like, you know, parallel timeline there. Yeah, look, it looks appealing to invade Taiwan. I mean, maybe not because they, you know, basically remote cutoff of the chips. And so then it doesn't mean they get the chips, but it just means they, they, you know, it's just, it's, you know, the machines are deactivated. But, look, I mean, imagine if during the Cold

• 1:29:05Speaker 01

  War, you know, all of the world's uranium deposits had been in Berlin, you know, and Berlin was already, I mean, almost multiple times when this caused nuclear war. So, God help us all.

• 1:29:17Dwarkesh Patel?

  Well, the Groves had a plan after the, after the war.

• 1:29:19Speaker 01

  Yeah.

• 1:29:19Dwarkesh Patel?

  That the plan was that America would go around the world and getting the rights to every single uranium deposit because they didn't realize how much uranium there was in the world. And they thought this was the thing that was feasible. Not realizing, of course, that there's like huge deposits in the Soviet Union itself.

• 1:29:32Speaker 01

  Right.

• 1:29:34Dwarkesh Patel?

  Okay.

• 1:29:35Speaker 01

  East Germany too. There's a, there's always, there's a lot of East German workers who kind of got screwed.

• 1:29:39Dwarkesh Patel?

  Oh, interesting.

• 1:29:39Speaker 01

  And got cancer.

• 1:29:42Dwarkesh Patel?

  Okay, so the framing we've been talking about, that we've been assuming.

• 1:29:45Speaker 01

  Yeah.

• 1:29:45Dwarkesh Patel?

  And I'm not sure I buy yet, is that the United States.

• 1:29:50Speaker 01

  Yeah.

• 1:29:50Dwarkesh Patel?

  This is our leverage. This is our data center. China is a competitor. Right now, obviously, that's not the way things are progressing. Private companies control these AIs. They're deploying them. It's a market based thing.

• 1:30:02Dwarkesh Patel?

  Why, why will it be the case that the, it's like the United States has this leverage or is doing this thing versus China is doing this thing?

• 1:30:11Speaker 01

  Yeah, I mean, look, look, on the, on the project, you know, I mean, there's sort of descriptive and prescriptive claims or sort of normative positive claims. I think the main thing I'm trying to say is, you know, you know, look, we're at, we're at these SF parties or whatever. And I think people talk about AGI and they're always just talking about the private AI labs. And I think I just really want to challenge that assumption. It just seems like, seems pretty likely to me, you know, as we've talked about, for reasons we've talked about, that, look, like the national security state is going to get involved. And, you know, I think there's a lot of ways this could look like, right? Is it, is it like nationalization? Is it a public private partnership? Is it a kind of defense contractor, like a relationship? Is it a sort of government project that's supposed to help the people? And so there's a spectrum there. But I think people are just vastly underrating the chances of this more or less

• 1:30:56Speaker 01

  looking like a government project. And look, I mean, look, if, you know, it's sort of like, you know, do you think, you think like we all have literal, like, you know, when we have like literal superintelligence on our cluster, right? And it's like, you know, you have 100 billion, they're like, sorry, you have a billion like superintelligent scientists that they can like hack everything, they can like stuxnet the Chinese data centers, you know, they're starting to build the robo-armies, you know, you like, you really think that'll be like a private company? And the government would be like, oh my God, what is going on? You know, like, yeah.

• 1:31:22Dwarkesh Patel?

  Suppose there's no China. Suppose there's people like Iran, North Korea, who theoretically at some point will go to do superintelligence, but they're not on our heels and don't have the ability to be on our heels. In that world, are you advocating for the national project or do you prefer the private path or?

• 1:31:40Speaker 01

  Yes, I mean, two responses to this. One is, I mean, you still have like Russia, you still have these other countries, you know, you've got to have Russia-proof security, right? It's like you can't, you can't just have Russia steal all your stuff. And like, maybe their clusters aren't going to be as big, but like, they're still going to be able to make the crazy bioweapons and the, you know, the mosquito-sized drone swarm, you know, and so on. And so, I mean, I think, I think, I think the security component is just actually a pretty large component of the project in the sense of like, I currently do not see another way where we don't kind of like instantly proliferate this to everybody. And so, yeah, so I think it's sort of like, you still have to deal with Russia, you know, Iran, North Korea, and you know, like, you know, Saudi and Iran are going to be trying to get it because they want to screw each other, and you know, Pakistan and India because they want to screw each other. There's like this enormous destabilization still. That said, look, I agree with you.

• 1:32:23Speaker 01

  If, you know, if, you know, by some, somehow things had shaken out differently, I'm like, you know, AGI would have been in 2005, you know, sort of like unparalleled in American hegemony, I think there would have been more scope for less government involvement. But again, you know, as we were talking about earlier, I think that would have been sort of this like very unique moment in history. And I think basically, you know, almost all other moments in history, there would have been the sort of great power competitor.

• 1:32:48Dwarkesh Patel?

  So, okay, so let's get into this debate. So I, my position here is if you look at the people who are involved in the Manhattan Project itself, many of them regretted their participation, as you said. Now, we can infer from that that we should sort of start off with a cautious approach to the nationalized ASI project. Then you might say, well, listen, obviously the super-

• 1:33:12Speaker 01

  Did they regret their participation because of the project or because of the technology itself? I think people will regret it, but I think it's about the nature of the technology and it's not about the project.

• 1:33:23Dwarkesh Patel?

  I think they also probably had a sense that different decisions would have been made if it wasn't some concerted effort that everybody had agreed to participate in. That if it wasn't in the context of this, we need to race to beat Germany and Japan, you might not develop, so that's a technology part, but also like you wouldn't actually like hit them with-

• 1:33:40Speaker 01

  But it's still gonna be a weapon, you know, it's like the sort of destructive potential, the sort of military potential. It's not, it's not because of the project, it is because of the technology. And that will unfold regardless, you know?

• 1:33:50Dwarkesh Patel?

  But I think this underrates the power of muddling-

• 1:33:54Speaker 01

  Imagine you go through like the 20th century in like, you know, a decade, you know, it's just the sort of, the sort of, yes, great technological progress.

• 1:34:01Dwarkesh Patel?

  Actually, let's actually run that example. Suppose you actually, there was some reason that the 20th century would be run through in one decade. Do you think the cause of that should have been, should have been like the technologies that happened through the 20th century shouldn't have been privatized? That it should have been a more sort of concerted government-led project?

• 1:34:20Speaker 01

  You know, look, there is a history of just dual-use technologies, right? And so I think AI in some sense is gonna be dual-use in the same way. And so there's gonna be lots of civilian uses of it, right? Like, nuclear energy, it's like itself, right? It was like, you know, there's the government project, develop the military angle of it. And then, you know, it was like, you know, then the government worked with private companies, there's a sort of like real, like flourishing of nuclear energy until, you know, the environmental ists stopped it. You know, planes, right? Like Boeing, or actually, you know, the Manhattan Project wasn't the biggest defense R and D project during World War II. It was the B-29 bomber, right? Because they needed the bomber that had long enough range to reach Japan, to destroy their cities. And then, you know, Boeing made some, Boeing made that, Boeing made the B-47, made the B-52, you know, the plane the US military uses today. And then they used that technology later on to-

• 1:35:06Speaker 01

  To, you know, build the 707 and this sort of the

• 1:35:08Dwarkesh Patel?

  But what does later on mean in this context? Because in the other, like, I get what it means after a war to privatize. But if you have, the government has ASI. Maybe just let me back up and explain my concern.

• 1:35:20Speaker 01

  Yeah.

• 1:35:20Dwarkesh Patel?

  So you have the only institution in our society which has a monopoly on violence.

• 1:35:24Speaker 01

  Yes.

• 1:35:26Dwarkesh Patel?

  And then we're going to give the, give it some, in a way that's not broadly deployed, access to the ASI.

• 1:35:32Speaker 01

  Yeah.

• 1:35:33Dwarkesh Patel?

  The counterfactual, and this maybe sounds silly, but listen, we're going to go through higher and higher levels of intelligence.

• 1:35:39Speaker 01

  Yeah.

• 1:35:39Dwarkesh Patel?

  Private companies

• 1:35:41Speaker 01

  Yeah.

• 1:35:41Dwarkesh Patel?

  will be required by regulation to increase their security.

• 1:35:44Speaker 01

  Yeah.

• 1:35:44Dwarkesh Patel?

  But they'll still be private companies and they're deploy this. And they're going to release the AGI. Now McDonald's and JP Morgan and some random startup are now more effective organizations because they have a bunch of AGI workers. And it'll be sort of like the industrial revolution in the sense that the benefits were widely diffused. If you don't end up in a situation like that, then the, I mean, even backing up, like, what is it we're trying to, why do we want to win against China? We want to win against China because we don't want a top-down authoritarian system

• 1:36:13Speaker 01

  Yeah.

• 1:36:13Dwarkesh Patel?

  to win.

• 1:36:14Speaker 01

  Yeah.

• 1:36:14Dwarkesh Patel?

  Now, if the way to beat that is that the most important technology that humanity will have has to be controlled by a top-down government, like, what was the point? Like, why, maybe, so, let's, like, run our cards with privatization. That's the way we get to the classic liberal market-based system we want for the ASI.

• 1:36:32Speaker 01

  Yep. All right, so, a lot to talk about here.

• 1:36:34Dwarkesh Patel?

  Yeah.

• 1:36:35Speaker 01

  I think, yeah, maybe I'll start a bit about, like, actually looking at what the private world would look like. And I think this is part of where the sort of, there's no alternative comes from. And then let's look like, look at, like, what the government project looks like, what checks and balances look like, and so on. All right, private world. I mean, first of all, okay, so, right, like, a lot of people right now talk about open source. And I think there's this sort of misconception that, like, AGI development is going to be like, oh, it's going to be some, like, beautiful decentralized thing. And, you know, like, you know, some giddy community of coders who gets to, like, you know, collaborate on it. That's not how it's going to look like, right? You know, it's, you know, $100 billion, trillion dollar cluster, it's not going to be that many people that have it. The algorithms, you know, it's like, right now, open source is kind of good because people just use the stuff that was published. And so they basically, you know, the algorithms were published. Or, you know, as Mistral, they just kind of like leave DeepMind and, you know, take all the secrets like that. And they just kind of replicate it.

• 1:37:19Speaker 01

  But that's not going to continue to be in the case. And so, you know, the sort of like open source alternative. I mean, also, people say stuff like, you know, 1026 flops. It'll be in my phone or, you know, it's no, it won't. You know, it's like Moore's law is really slow. I mean, AI chips are getting better, but like, you know, the $100 billion computer will not cost, you know, like $1,000, you know, within your lifetime or whatever. Aside from me. So anyway, so it's gonna be, it's gonna be like two or three, you know, big players on the private world. And so, look, a few things. So first of all,

• 1:37:49Speaker 01

  you know, you talk about the sort of like enormous power that sort of superintelligence will have and the government will have.

• 1:37:57Speaker 01

  I think it's pretty plausible that the alternative world is that like one AI company has that power, right? And especially if we're talking about lead, you know, it's like, what? I don't know. OpenAI has a six month lead. And then, you know, so then you're not talking, you're talking about basically, you know, the most powerful weapon ever. And it's, you know, you're kind of making this like radical bet on like a private company CEO is the benevolent dictator.

• 1:38:16Dwarkesh Patel?

  No, no, this is not necessarily like any other thing that's privatized. We don't count on them being benevolent. We just, look, to think of, for example, somebody who manufactures industrial fertilizer.

• 1:38:25Speaker 01

  Yes.

• 1:38:26Dwarkesh Patel?

  Right? This, the person with this factory, if they went back to an ancient civilization, they could like blow up Rome.

• 1:38:33Speaker 01

  Yep.

• 1:38:33Dwarkesh Patel?

  It could probably blow up Washington DC.

• 1:38:35Speaker 01

  Indeed.

• 1:38:35Dwarkesh Patel?

  And I think in their series, you talk about Tyler Cowen's phrase of muddling through.

• 1:38:40Speaker 01

  Yeah.

• 1:38:40Dwarkesh Patel?

  And I think even with privatization, people sort of underrate that there are actually a lot of private actors who have the ability to like, there's a lot of people who control the water supply or whatever. And we can count on cooperation and market based incentives to basically keep a balance of power.

• 1:38:55Speaker 01

  Sure.

• 1:38:56Dwarkesh Patel?

  I gather things are proceeding really fast.

• 1:38:57Speaker 01

  Yes.

• 1:38:57Dwarkesh Patel?

  But we like, we have a lot of historical evidence that this is the thing that works best.

• 1:39:00Speaker 01

  So, look, I mean, I mean, what do we do with nukes? Right? The, the way we keep the sort of nukes in check is not like, you know, a sort of beefed up second amendment where like each state has their own like little nuclear arsenal and like, you know, Dario and Sam have their own little nuclear arsenal. No, no, it's like, it's institutions, it's constitutions, it's laws, it's, it's, it's courts. And so, so anyway, I don't actually, I'm not sure that this, you know, I'm not sure that the sort of balance of power analogy holds. In fact, you know, Sort of the government having the biggest guns was sort of like an enormous civilizational achievement, right? Like Landfrieden in the sort of Holy Roman Empire, right? You know, if somebody from the town over kind of committed a crime on you, you know, you didn't kind of start a sort of a, you know, a big battle between the two towns. No, you take it to a court of the Holy Roman Empire and they would decide. And it's a big achievement.

• 1:39:44Speaker 01

  Now, the thing about, you know, the industrial fertilizer, I think the key difference is kind of speed and offense defense balance issues, right? So it's like 20th century and, you know, 10 years in a few years. That is an incredibly scary period. And it is incredibly scary, you know, because it's, you know, you're going through just this sort of enormous array of destructive technology and the sort of like enormous amount of like, you know, basically military advance. I mean, you would have gone from, you know, kind of like, you know, you know, bayonets and horses to kind of like tank armies and fighter jets in like a couple of years. And then from, you know, like, you know, and then to like, you know, nukes and, you know, ICBMs and stuff, you know, just like in a matter of years. And so it is sort of that speed that creates, I think, basically the way I think about it is there's going to be this initial just incredibly volatile, incredibly dangerous

• 1:40:29Speaker 01

  period. And somehow we have to make it through that. And that's going to be incredibly challenging. That's where you need the kind of government project. If you can make it through that, then you kind of go to like, now we can, now, you know, the situation has been stabilized. You know, we don't face this imminent national security threat. You know, it's like, yes, there were kind of WMDs that came along the way, but either we've managed to kind of like have a sort of stable offense defense balance, right? Like, I think bioweapons initially are a huge issue, right? Like, an attacker can just create like a thousand different synthetic, you know, viruses and spread them. And it's like, going to be really hard for you to kind of like, make a defense against each. But maybe at some point, you figure out the kind of like, you know, universal defense against every possible virus. And then you're in a stable situation again on the offense defense balance. Or you do the thing, you know, you do with planes where it's there's like, you know, there's certain capabilities that the private sector isn't allowed to have. And you've like figured out what's going

• 1:41:13Speaker 01

  on, restrict those. And then you can kind of like, let, let, you know, you let the sort of civilian civilian uses.

• 1:41:19Dwarkesh Patel?

  So I'm skeptical of this because. Well, there's

• 1:41:23Speaker 01

  And then, sorry, I mean, the other important thing is, so I talked about the sort of, you know, maybe it's like, it's, it's a, you know, it's, you know, it's one company with all this power. And I think it is like, I think it is unprecedented because it's like the industrial fertilizer guy cannot overthrow the US government. I think it's quite plausible that like the AI company with superintelligence can overthrow the

• 1:41:40Dwarkesh Patel?

  But there'll be multiple AI companies, right? And I buy that one of them could be ahead.

• 1:41:43Speaker 01

  So it's not obvious that it'll be multiple. I think it's, again, if there's like a six month lead, maybe, maybe there's two or three.

• 1:41:48Dwarkesh Patel?

  I agree.

• 1:41:48Speaker 01

  But if there's two or three, then what you have is just like a crazy race between these two or three companies. You know, it's like, you know, whatever, Demis and Sam, they're just like, I don't want to let the other one win. And, um, and they're both developing their nuclear arsenals and the robot. It's just like, also like, come on, the government is not going to let these people, you know, are they going to let like, you know, is Dario going to be the one developing the kind of like, you know, you know, super hacking Stuxnet and like deploying against the Chinese data center? The other issue though, is it won't just, if it's two or three, it won't just be two or three. There'll be two or three and it'll be China and Russia and North Korea. Because the private, in the private lab world, there is no way they'll have security that is good enough.

• 1:42:22Dwarkesh Patel?

  I think we're also assuming that somehow if you nationalize it, like the security just, especially in the world where this stuff is priced in by the CCP, that now you've like got it nailed down. And I'm not sure why we would expect that to be the case, but on this

• 1:42:37Speaker 01

  Government's the only one who does this stuff.

• 1:42:38Dwarkesh Patel?

  So, so if it's not Sam or Dario, who's, uh, we don't trust them to be benevolent dictator or whatever.

• 1:42:44Speaker 01

  Well, we're just corporate governments.

• 1:42:45Dwarkesh Patel?

  Okay, okay, okay. So, but here we're counting on, if it's because you can cause a coup, the same capabilities are going to be true of the government project, right? And so the model president in 2020, uh, 2025, but Donald Trump will be the person that you don't trust Sam or Dario to have these capabilities. And why, okay, I agree that like, I'm worried if Sam or Dario have a one year lead on ASI in that, in that world, then I'm like concerned about this being privatized. But in that exact same world, I'm very concerned about Donald Trump having the capability. And potentially if we're living in a world where the takeoff is slower than you anticipate, in that world, I'm like, very much I want the private companies. So like, in no part of this matrix, is it obviously true that the government led project is better than the private project?

• 1:43:30Speaker 01

  Let's talk about the government project a little bit and check Checks and balances. In some sense, I think my argument is a sort of Berkeley argument, which is like, American checks and balances have held for, you know, over 200 years and through crazy technological revolutions. You know, the US military could kill like every civilian in the United States.

• 1:43:45Dwarkesh Patel?

  You're gonna make that argument. The private-public balance of power has held for hundreds of years and then like-

• 1:43:50Speaker 01

  Corporate, but yeah, why has it held? Because the government has the biggest guns. And has never before has a single CEO or a random nonprofit board had the ability to launch nukes. And so again, it's like, you know, what is the track record of the government checks and balances versus the track record of the private company checks and balances? Well, the iLab, you know, like, first stress test, you know, went really badly, you know, that didn't really work, you know?

• 1:44:12Speaker 01

  I mean, even worse in the, in the, in the sort of private company world. So it's both like, it is not just the two, it is, it is like the two private companies and the CCP, and they just like instantly have all the shit. And then it's, you know, they probably won't have good enough internal control. So it's like, not just like the random CEO, but it's like, you know, rogue employees that can kind of like use these super intelligences to do whatever they want.

• 1:44:29Dwarkesh Patel?

  And this, this won't be true of the government? Like, the rogue employees won't exist on the project?

• 1:44:34Speaker 01

  Well, the government actually like, you know, has decades of experience and like actually really cares about the stuff. I mean, it's like they deal, they deal with nukes, they deal with really powerful technology. And it's, you know, this is like, this is the stuff that the national security state cares about. You know, again, to the, let's talk about the government checks and balances a little bit. So, you know, what are, what are checks and balances in the government world? First of all, I think it's actually quite important that you have some amount of international coalition. And I talked about these sort of two tiers before. Basically, I think the inner tier is a sort of modeled on the Quebec agreement, right? This was like Churchill and Roosevelt. They kind of agreed secretly, we're gonna like pool our efforts on nukes, but we're not gonna use them against each other, and we're not gonna use them against anyone else with their consent. And I think basically, look, bring in, bring in the UK, they have DeepMind, bring in the kind of like Southeast Asian states who have the chip supply chain,

• 1:45:17Speaker 01

  bring in some more of kind of like NATO close democratic allies for, you know, talent and industrial resources. And you have this sort of like, you know, so you have, you have those checks and balances in terms of like more international countries at the table.

• 1:45:29Speaker 01

  Somewhat separately, but then you have the sort of second tier of coalitions, which is the sort of atoms for peace thing, where you go to a bunch of countries, including like the UAE, and you're like, look, we're going to basically like, you know, there's a deal similar to like the NPT stuff where it's like, you're not allowed to like, do the crazy military stuff, but we're going to share the civilian applications. We're in fact going to help you and share the benefits and, you know, sort of kind of like this new sort of post superintelligence world order. All right. US checks and balances, right? So obviously, Congress is going to have to be involved, right? Appropriate trillions of dollars. I think probably ideally, you have, Congress needs to kind of like confirm whoever's running this. So you have Congress, you have like different factions of the government, you have the courts. I expect the First Amendment to continue being really important. And maybe that, I think that sounds kind of crazy to people, but I actually think, again, I think these are like institutions that have held the test of time in a really sort

• 1:46:14Speaker 01

  of powerful way. You know, eventually, you know, this is why, honestly, alignment is important is like, you know, the AIs, you program the AIs to follow the Constitution. And it's like, you know, why does the military work? It's like generals, you know, are not allowed to follow unlawful orders. They're not allowed to follow unconstitutional orders. You have the same thing for the AIs.

• 1:46:33Dwarkesh Patel?

  So what's wrong with this argument? Where you say, listen, maybe you have a point in the world where we have extremely fast takeoff. It's like one year from AGI to ASI.

• 1:46:41Speaker 01

  Yeah. And then you have the like, eight years after of ASI, where you have this like extraordinary technological progress.

• 1:46:47Dwarkesh Patel?

  Maybe you have a point. We don't know. You have these arguments. We'll like get into the weeds on them about why that's a more likely world, but like, maybe that's not the world we live in. And in the other world, I'm like, very on the side of making sure that these things are privately held. Now, why, when you nationalize, so when you nationalize, that's a one-way function. You can't go back. Why not wait until we have more evidence on which of those worlds we live in? Why, I think like rushing on the nationalization might be a bad idea while we're not sure. And, okay, I'll just respond to that first.

• 1:47:19Speaker 01

  I mean, I don't, I don't expect us to nationalize tomorrow. If anything, I expect it to be kind of with COVID, where it's like kind of too late. Like ideally, you nationalize it early enough to like actually lock stuff down. It'll probably be kind of chaotic and like, you're going to be trying to like do this crash program to lock stuff down and it'll be kind of late. It'll be kind of clear what's happening. We're not going to nationalize when it's not clear what's happening.

• 1:47:37Dwarkesh Patel?

  I think the whole historically institutions have held up well. First of all, they've actually almost broken a bunch of times. It's like this is this is this is the argument that some people who are saying that we shouldn't be that concerned about nuclear war, say, or it's like, listen, we have the nuke for 80 years and like, we've been fine so far, so the risk must be low. And then the answer to that is no, actually, it is a really high risk. And the reason we've avoided is like, people have gone through a lot of effort to make sure that this thing doesn't happen. I don't think that giving government ASI without knowing what that implies is going through a lot of effort. And I think the base rate, like you can talk about America.

• 1:48:15Dwarkesh Patel?

  I think America is very exceptional, not just in terms of dictatorship, but in terms of every other country in history has had a complete drawdown of wealth because of war, revolution, something. America is very unique in not having that. And the historical base rate, we're talking about great power competition. I think that has a really big, that's something we haven't been thinking about the last 80 years, but it's really big.

• 1:48:34Speaker 01

  Yeah.

• 1:48:35Dwarkesh Patel?

  Dictatorship is also something that is just the default state of mankind.

• 1:48:39Speaker 01

  Yeah.

• 1:48:39Dwarkesh Patel?

  And I think relying on institutions, which in an ASI world, like there's fundamentally right now, if the government tried to overthrow, there's a, it's much harder if you don't have the ASI, right? Like there's people who have AK AR-1415s and I, there's like things that make it harder.

• 1:48:57Speaker 01

  Could crush the AR-15s.

• 1:48:59Dwarkesh Patel?

  No, I think it should be pretty hard. The reason was Vietnam and Afghanistan was pretty hard.

• 1:49:02Speaker 01

  Whole country.

• 1:49:04Dwarkesh Patel?

  Yeah, yeah, I agree, but like I'm.

• 1:49:05Speaker 01

  They could.

• 1:49:06Dwarkesh Patel?

  I agree.

• 1:49:06Speaker 01

  It's similar with ASI.

• 1:49:08Dwarkesh Patel?

  Um, yeah, I think it's just like easier if you have what you're talking about.

• 1:49:11Speaker 01

  But there are institutions, there are constitutions, there are legal restraints, there are courts, there are checks and balances. The crazy bet is the bet which are like private company CEOs.

• 1:49:20Dwarkesh Patel?

  The same thing, by the way, isn't the same thing true of nukes where we have these institutional agreements about non-proliferation and whatever? And we're still very concerned about that being broken and somebody getting nukes and like, you should stay up at night worrying about that.

• 1:49:31Speaker 01

  Situation. But ASI is going to be a really precarious situation as well. And like, given, given how precarious nukes are, we've done pretty well.

• 1:49:38Dwarkesh Patel?

  And so what does privatization in this world even mean?

• 1:49:40Speaker 01

  I mean, I think the other thing is.

• 1:49:41Dwarkesh Patel?

  Like what happens after?

• 1:49:42Speaker 01

  I mean, the other thing, you know, because we're talking about like whether the government project is good or not. And it's like, I have very mixed feelings about this as well. Again, I think my primary argument is like.

• 1:49:52Speaker 01

  You know, if you're at the point where this thing has like vastly superhuman hacking capabilities, if you're at the point where this thing can develop, you know, bioweapons, you know, like crazy bioweapons, ones that are like targeted, you know, can kill everybody but the hand Chinese or, you know, that, you know, you know, would, would wipe out, you know, entire countries where you're talking about like building robo armors, you're talking about kind of like drone swarms that are, um, you know, again, the mosquito sized drones that could take it out, you know.

• 1:50:19Speaker 01

  The United States national security state is going to be intimately involved with this. And this will, you know, the labs, whether, you know, and I think, again, the government, a lot of what I think is the government project looks like, it is basically a joint venture between like, you know, the cloud providers, between some of the labs and the government. And so I think there is no world in which the government isn't intimately involved in this like crazy period. The very least, basically, you know, like the intelligence agencies need to be running security for these labs. So they're already kind of like, they're controlling everything, they're controlling access to everything. Um, then they're going to be like, probably, again, if we're in this like really volatile international situation, like a lot of the initial applications, it'll, it'll suck. It's not what I want to use ASI for, will be like trying to somehow stabilize this crazy situation. Um, somehow we need to prevent like proliferation of like some crazy new WMDs and like the undermining of

• 1:51:01Speaker 01

  mutually assured destruction to kind of like, you know, North Korea and Russia and China.

• 1:51:07Speaker 01

  And so I think, you know, I basically think your world, you know, I think there's much more spectrum than you're acknowledging here. And I think basically the world in which it's private labs is like extremely heavy government involvement. And really what we're debating is like, you know, what form of government project, but it is going to look much more like, you know, the national security state than anything it does look like, like a startup as it is right now. And I think the, yeah.

• 1:51:30Dwarkesh Patel?

  Look, I, I think something like that makes sense. I would be, if it's like the Manhattan Project, then I'm very worried where it's like, this is part of the US military. Um, where if it's more like, listen, you gotta talk to Jake Sullivan before you like run the next training run.

• 1:51:44Speaker 01

  Like Lockheed Martin, Skunk Corps is part of the US military. It's like, they call the shots.

• 1:51:48Dwarkesh Patel?

  Yeah, I don't think that's great. I think that's, I think that's bad. I think it would be bad if that happened with ASI. And like, what is, what is the scenario?

• 1:51:54Speaker 01

  What is the alternative? What is the, what is the alternative?

• 1:51:56Dwarkesh Patel?

  Okay. So it's closer to my end of the spectrum where, yeah, you do have to talk to Jake Sullivan before you can launch the next training cluster.

• 1:52:03Speaker 01

  Yeah.

• 1:52:03Dwarkesh Patel?

  But there's many companies who are still going for it.

• 1:52:06Speaker 01

  Yeah.

• 1:52:06Dwarkesh Patel?

  And the government will be intimately involved in the security.

• 1:52:10Speaker 01

  Yeah.

• 1:52:11Dwarkesh Patel?

  The Bud, the like three different companies are trying to

• 1:52:13Speaker 01

  Is Darius launching the Stuxnet attack?

• 1:52:15Dwarkesh Patel?

  Yeah, what do you, what do you, is launching, launching, okay.

• 1:52:18Speaker 01

  Instead of watching. Darius is activating the Chinese data centers.

• 1:52:21Dwarkesh Patel?

  I think this is similar to the story you could tell about, there's a lot of, like, literally big tech right now.

• 1:52:25Speaker 01

  Yeah.

• 1:52:26Dwarkesh Patel?

  I think Satya, if he wanted to, he probably like could get his engineers, like, what are the zero days in Windows? And the companies, and like, well, how do we get infiltrate the president's computer so that like we can?

• 1:52:36Speaker 01

  They'd be shut down.

• 1:52:37Dwarkesh Patel?

  No, no, no, but like right now, I'm saying Satya could do that, right? Cause he knows.

• 1:52:40Speaker 01

  And maybe shut down.

• 1:52:41Dwarkesh Patel?

  What do you mean?

• 1:52:42Speaker 01

  Government wouldn't let them do that.

• 1:52:44Dwarkesh Patel?

  Yeah, I, I, I think there's a story you could tell where like they could pull out, pull off a coup, whatever. But like, I think there's like multiple AI companies.

• 1:52:49Speaker 01

  They could not pull off a coup.

• 1:52:49Dwarkesh Patel?

  Okay.

• 1:52:50Speaker 01

  They could not pull, come on.

• 1:52:51Dwarkesh Patel?

  Okay, fine, fine, fine. I agree. I'm just saying like something closer to, so what's wrong with the scenario where you, the government is, there's like multiple companies going for it.

• 1:53:01Speaker 01

  Yeah.

• 1:53:02Dwarkesh Patel?

  But the AI is still broadly deployed and alignment works in the sense that you can make sure that it's not, you, the system level prompt is like, you can't help people make bioweapons or something. But these are still broadly deployed, so that.

• 1:53:14Speaker 01

  I mean, I expect the AIs to be broadly deployed. I mean, first of all, yeah.

• 1:53:16Dwarkesh Patel?

  Even if it's a government project?

• 1:53:17Speaker 01

  Yeah, I mean, look, I think first of all, like, I think the metas of the world, you know, open sourcing their eyes, you know, that are two years behind or whatever. Yeah, super valuable role. They're gonna like, you know, and, and so there's gonna be some question of like, either the offense, defense balance is fine. And so like, even if they open source two-year-old AIs, it's fine, or it's like, there's some restrictions on the most extreme dual-use capabilities, like, you know, you don't let private companies sell kind of crazy weapons. And that's great, and that will help with the diffusion. And, you know, you know, after the government project, you know, there's gonna be this initial tense period, hopefully that's stabilized. And then, look, yeah, like Boeing, they're gonna go out and they're gonna like make, do all the flourishing civilian applications and, you know, like nuclear energy, you know, it'll like all, the civilian applications will have their day. I think part of my argument here is that.

• 1:53:56Dwarkesh Patel?

  And how does that proceed, right? Because in the other world, there's existing stocks of capital that are worth a.

• 1:54:01Speaker 01

  Yeah, the clusters, there'll be, still be Google clusters.

• 1:54:04Dwarkesh Patel?

  And so Google, because they got the contract from the government, they'll be the ones that control the ASI, but like, why, why are they trading with anybody else?

• 1:54:10Speaker 01

  It'll also be Microsoft.

• 1:54:11Dwarkesh Patel?

  Why is there some random startup getting like?

• 1:54:12Speaker 01

  It'll be the same, it'll be the same companies that would be doing it anyway, but in this, in this world, they're just contracting with the government or like they're DPA'd for all their compute goes to the government.

• 1:54:21Dwarkesh Patel?

  And but

• 1:54:22Speaker 01

  It's very natural.

• 1:54:24Dwarkesh Patel?

  So after you get the ESI and then we're building the robot armies and building fusion reactors or whatever, that that's

• 1:54:31Speaker 01

  Only the government will get to build robot armies.

• 1:54:34Dwarkesh Patel?

  Yeah, now I'm worried. Or like the fusion reactors and stuff.

• 1:54:37Speaker 01

  It's what we do with nukes. It's the same situation we have today.

• 1:54:40Dwarkesh Patel?

  Because if you already have the robot armies and everything, like the existing society doesn't have some leverage where it makes sense for the government to Yeah, they're giving the sense that there's like they have a lot of capital that the government wants and there's other things like, why was Boeing privatized after?

• 1:54:52Speaker 01

  Government has the biggest guns. Government has the biggest guns. And the way we regulate is institutions, constitutions, legal restraints.

• 1:54:56Dwarkesh Patel?

  Okay, so tell me what privatization should looks like in the ESI world afterwards.

• 1:54:59Speaker 01

  Afterwards, like the Boeing example, right? It's like you have this government

• 1:55:02Dwarkesh Patel?

  Who gets it? Like, the Google?

• 1:55:04Speaker 01

  Google, Microsoft.

• 1:55:05Dwarkesh Patel?

  And who are they selling it to? Like, they already have the robot factories. Like, why are they selling it to us? Like, they already have the, they don't need like our, this is chum change in the ESI world because we didn't get like the ESI broadly deployed throughout this takeoff. So we don't have the robot, we don't have like the fusion reactors and whatever advanced, decades of advanced science that you're talking about. So like it just, what are they trading with us for?

• 1:55:26Speaker 01

  Trading with whom for?

• 1:55:27Dwarkesh Patel?

  That everybody who was not part of the project. They've got that technology that's decades ahead.

• 1:55:30Speaker 01

  Yeah, I mean, look, that's a whole nother issue of like, well, how does like economic distribution work or whatever? I don't know. That'll be rough.

• 1:55:36Dwarkesh Patel?

  Yeah, I think

• 1:55:37Speaker 01

  I'm just saying, I don't, I don't, basically, I'm kind of like, I don't see the alternative. The alternative is you like overturn a 500 year civilization achievement of Lon Fleaton. You basically instantly leak the stuff to the CCP and either you like barely scrape out a head. And but you're in this fever struggle. You're like proliferating crazy WMDs. It's this like enormously dangerous situation, enormously dangerous on alignment because you're in this kind of like crazy race at the end and you don't have the ability to like take six months to get alignment right. The alternative is, you know, alternative is like you aren't actually bundling your efforts to kind of like win the race against the authoritarian powers. You know, yeah, and so.

• 1:56:16Speaker 01

  You know, I don't like it. You know, I wish, I wish the thing we use the ASI for is to like, you know, cure the diseases and do all the good in the world. But it is my prediction that sort of like by the, in the end game,

• 1:56:32Speaker 01

  what will be at stake will not just be kind of cool products, but what will be at stake is like whether liberal democracy survives, like whether the CCP survives, like what the world order for the next century will be. And when that is at stake, forces will be activated that are sort of way beyond what we're talking about now. And like, you know, in, in the sort of like crazy race at the end, like the sort of national security implications will be the most important. You know, sort of like, you know, World War II, it's like, yeah, you know, nuclear energy had its day, but in the initial kind of period when, you know, when this technology was first discovered, you had to stabilize the situation, you had to get nukes, you had to do it right. And then, and then the civilian applications had their day.

• 1:57:12Dwarkesh Patel?

  I think of closer analogy to what this is because nuclear, I agree that nuclear energy is a thing that happens later on and it's like dual use in that way. But it's, it's something that happened like literally a decade after nuclear weapons were developed.

• 1:57:21Speaker 01

  Yeah.

• 1:57:21Dwarkesh Patel?

  Whereas with AI, like the immediately all the applications are unlocked and it's closer to literally, I mean, this is analogy people actually listen to me in the context of AGI is like, assume your society had 100 million more John von Neumann.

• 1:57:34Speaker 01

  Yeah, yeah.

• 1:57:34Dwarkesh Patel?

  And I don't think like if that was literally what happened, if tomorrow you just have 100 million more of them, the approach should have been, well, some of them will convert to ISIS and we need to like, be really careful about that. And then like, oh, you know, like what if a bunch of them are born in China and then we like, if we got to nationalize the John von Neumanns, I'm like, though I think it will be generally a good thing. And I'd be concerned about one power had getting like all the John von Neumanns.

• 1:57:56Speaker 01

  I mean, I think the issue is the sort of like bottling up and the sort of intensely short period of time, like this enormous sort of like, you know, unfolding of technological progress of an industrial explosion. And I think we do worry about the 100 million John von Neumann. And it's like rise of China. Why are we worried about the rise of China? Because it's like 100 billion people and they're able to do a lot of industry and do a lot of technology. And but it's just like, you know, the rise of China times like, you know, 100, because it's not just 100, 1 billion people. It's like a billion, super intelligent, crazy, you know, crazy things. And, and in like, you know, very short period.

• 1:58:27Dwarkesh Patel?

  Let's talk practically, because if the goal is we need to beat China, part of that is protecting-

• 1:58:32Speaker 01

  I mean, that's one of the goals, right?

• 1:58:34Dwarkesh Patel?

  Yeah, I agree. Well, one of the goals is to beat China.

• 1:58:36Speaker 01

  And also manage this incredibly crazy, scary period.

• 1:58:39Dwarkesh Patel?

  Right. So part of that is making sure we're not leaking our own secrets to them.

• 1:58:43Speaker 01

  Yep. Part of that is all- Building the trillion dollar cluster. I mean, building the trillion dollar cluster, right?

• 1:58:47Dwarkesh Patel?

  Yeah, but using your whole point that Microsoft can release corporate bonds that are-

• 1:58:51Speaker 01

  I think Microsoft can do the like hundreds of billions of dollar cluster. I think the trillion dollar cluster is closer to a national effort.

• 1:58:56Dwarkesh Patel?

  I thought your earlier point was that American capital markets are deep and so forth.

• 1:59:01Speaker 01

  They're pretty good. I mean, I think it's possible it's private. It's possible. But it's gonna be like, you know-

• 1:59:06Dwarkesh Patel?

  By the way, at this point, we have AGI that's rapidly accelerating productivity.

• 1:59:10Speaker 01

  I think the trillion dollar cluster is gonna be planned before the AGI. I think it's sort of like you get the AGI on the like 10 gigawatt cluster. Like maybe you have like one more year where you're kind of doing some final un-hobbling to fully unlock it. Then you have the intelligence explosion. And meanwhile, the trillion dollar cluster is almost finished. And then you do your super intelligence on your trillion dollar cluster. Or you run it on your trillion dollar cluster. And by the way, you have not just your trillion dollar cluster, but like, you know, hundreds of millions of GPUs on inference clusters everywhere.

• 1:59:34Dwarkesh Patel?

  And this isn't result, like, I think private, in this world, I think private companies have the capital and can raise capital to do it.

• 1:59:40Speaker 01

  The thing you will need the government force to do it fast.

• 1:59:42Dwarkesh Patel?

  Well, I was just about to ask, like, wouldn't it be the, like, we know private companies are on track to be able to do this? And beat China if they're unhindered by climate pledges or whatever.

• 1:59:52Speaker 01

  Well, that's part of what I'm saying.

• 1:59:53Dwarkesh Patel?

  So if that's the case, and if it really matters that we beat China, there's all gonna be all kinds of practical difficulties of like, will the AI researchers actually join the AI effort? If they do, there's gonna be three different teams at least, who are currently doing pre-training on different, they're different companies. Now, who decides, at some point, you're gonna have to, like, YOLO the hyperparameters of the trillion dollar cluster.

• 2:00:20Speaker 01

  YOLO the hyperparameters of the thing.

• 2:00:21Dwarkesh Patel?

  Who decides that? Just like merging extremely complicated research and development processes across very different organizations. This is somehow supposed to speed up America against the Chinese? Like, why don't we just let-

• 2:00:33Speaker 01

  Brain and DeepMind merged and it was like a little messy, but it was fine.

• 2:00:36Dwarkesh Patel?

  It was pretty messy and it was also the same company and also much earlier on in the process.

• 2:00:39Speaker 01

  Pretty similar, right? Same code, different code bases and like lots of different infrastructure and different teams. And it was like, you know, it wasn't, it wasn't like, it wasn't the smoothest of all processes, but you know, DeepMind is doing, I think, very well.

• 2:00:48Dwarkesh Patel?

  I mean, look, you give the example of COVID and the COVID example was like, listen, we woke up to it, maybe it was late, but then we deployed all this money. And COVID response to government was a clusterfuck over. And like, the only part of it that was worked is, I agree WarpSpeed was like enabled by the government. It was literally just giving the permission that you can actually do.

• 2:01:07Speaker 01

  Well, no, it was also taking, making like the big-

• 2:01:09Dwarkesh Patel?

  Advance American commitments or whatever. But I agree, but it was like fundamentally, it was like a private sector led effort.

• 2:01:13Speaker 01

  Yeah.

• 2:01:13Dwarkesh Patel?

  That was the only part of COVID that worked.

• 2:01:14Speaker 01

  I mean, I think, I think, again, I think the project will look closer to Operation WarpSpeed. And it's not even, I mean, I think, I think you'll have all the companies involved in the government project. I'm not that sold that merging is that difficult. You know, you have one, okay, you select one code base and you know, you run pre-training on like GPUs with, you know, one code base. And then you do the sort of secondary step on the, you know, the other code base with TPUs. I don't know, I think it's fine.

• 2:01:36Speaker 01

  I mean, to the topic of like, will people sign up for it? They wouldn't sign up for it today. I think this would be kind of crazy to people. But also, you know, I mean, this is part of the like secrets thing, you know, people gather at parties or whatever, you know, you know this. You know, I don't think anyone has really gotten up in front of these people and been like, look, you know, the thing you're building is the most important thing for like the national security of the United States for like whether, you know, like, you know, the free world will have another century ahead of it. Like, this is, this thing you're doing is really important, like for your country, for democracy. And, um, you know, don't talk about the secrets and it's not just about, you know, oh, DeepMind or whatever. It's about, it's about, you know, these really important things.

• 2:02:15Speaker 01

  And so, you know, I don't know, like again, we're talking about the Manhattan Project, right? This stuff was really contentious initially. But, you know, at some point, it was like clear that this stuff was coming. It was clear that there was like sort of a real sort of like exigency on the military national security front. And, um, You know, I think a lot of people come around on the like whether it'll be competent. I I agree. I mean, this is again where it's like a lot of the stuff is more like predictive in the sense. I think this is like reasonably likely, and I think not enough people are thinking about it. You know, like a lot of people think about like AI lab, politics or whatever. But like, nobody has a plan for the project, you know, it's like, you know, like

• 2:02:46Dwarkesh Patel?

  Should that make you more pessimistic about it? And like, we don't have a plan for it. We need to do it very soon because AGI is upon us.

• 2:02:51Speaker 01

  Yeah.

• 2:02:52Dwarkesh Patel?

  Then fuck, the only capable competent and technical institutions capable of making AI right now are private companies.

• 2:02:57Speaker 01

  Yeah, and they're gonna play that leading role. It'll be a sort of a partnership, basically. But the other, you know, the other thing is like, you know, again, we talked about World War II and, you know, American unpreparedness, the beginning of World War II is complete, you know, complete shambles, right? And so there is a sort of like very company. I think America has a very deep bench of just like incredibly competent managerial talent. You know, I think that, you know, there's a lot of really dedicated people. And, um, you know, I think basically a sort of operational warp speed, public, private partnership, something like that, you know, is sort of what I imagine it would look like.

• 2:03:25Dwarkesh Patel?

  Yeah, I mean, the recruiting the talent is an interesting question because the same sort of thing where

• 2:03:33Dwarkesh Patel?

  initially for the Manhattan Project, you had to convince people, we've got to beat the Nazis and you got to get on board. I think a lot of them maybe regretted how much they accelerated the bomb. And I want, I think this is generally a thing of war where

• 2:03:47Speaker 01

  I mean, I think they're also wrong to regret it, but.

• 2:03:51Dwarkesh Patel?

  Yeah, I mean, why?

• 2:03:53Speaker 01

  What's the reason for regretting it?

• 2:03:55Dwarkesh Patel?

  I think there's a world in which you don't have, the the way in which nuclear weapons were developed after the war was pretty explosive because there was a precedent that you actually can use nuclear weapons. Then because of the race that was set up, you immediately go to the H-bomb.

• 2:04:11Speaker 01

  I mean, I think my view is, again, this is, this is related to the view on AI and maybe some of our disagreement is like, that was inevitable, like, of course, like, you know, there's this, you know, world war. And then obviously there was the, you know, Cold War right after. Of course, like, you know, the military and technology angle of this would be like, you know, pursued with ferocious intensity. And I don't really think there's a world in which that doesn't happen, which is like, ah, we're all not going to build nukes. And also just like, nukes went really well. I think that could have gone terribly, right? You know, like, and, you know, again, I mean, this sort of, I think this is like not physically possible with nukes, this sort of pocket nukes for everybody. But I think sort of like WMDs that are sort of proliferated and democratized and like all the countries have it like. The US leading on nukes and then sort of like building this new world order that was kind of US led, or at least sort of like a few great powers and a non-proliferation

• 2:04:55Speaker 01

  regime for nukes, a partnership and a deal that's like, look, no military sort of application of nuclear technology, but we're going to help you with the civilian technology. We're going to enforce safety norms on the rest of the world. That worked. It worked. And it could have gone so much worse.

• 2:05:08Dwarkesh Patel?

  Okay, so zooming out.

• 2:05:10Speaker 01

  And Hiroshima and Nagasaki, you know, they were, I mean, this is, I mean, I say this a bit in the piece, but it's like, actually, the A-bomb, you know, like the A-bomb on Hiroshima and Nagasaki was just like, you know.

• 2:05:17Dwarkesh Patel?

  The firebombing, yeah.

• 2:05:18Speaker 01

  Firebombing. The thing, I think the thing that really changed the game was like the super, you know, the H-bombs and ICBMs. And then I think that's really when it took it to like a whole new level.

• 2:05:27Dwarkesh Patel?

  I think part of me thinks when you say, we, we'll tell the people that for the free world to survive, we need to pursue this project. It sounds similar to World War II is, so World War II is a sad story, obviously, the fact that it happened, but also like the victory is sad in the sense that Britain goes in to protect Poland.

• 2:05:49Speaker 01

  Yeah.

• 2:05:50Dwarkesh Patel?

  And at the end, the USSR, which is, you know, as your family knows, is incredibly brutal, ends up occupying half of Europe.

• 2:06:01Speaker 01

  Right.

• 2:06:02Dwarkesh Patel?

  And the, like, part of like, we're protecting the free world, that's why we got to rush the AI. And like, if we end up with the American AI Leviathan, I think there's a world where we look back on this where it has the same sort of twisted irony that Britain going into World War II had about trying to protect Poland.

• 2:06:22Speaker 01

  Look, I mean, I think there's going to be a lot of unfortunate things that happen. I'm just like, I'm just hoping we make it through. I mean, to the, to the point of it's like, I really don't think the pitch will only be the sort of like, you know, the race. I think the race will be sort of a backdrop to it. I think the sort of general like, look, it's important that democracies shape this technology. We can't just like leak this stuff to, you know, North Korea. It's going to be important. I think also for the, just safety, including alignment, including the sort of like creation of new WMDs. I'm not currently sold there's another path, right? So it's like, if you just have the breakneck grace, both internationally, because you're just instantly leaking all this stuff, including the weights, and just, you know, the commercial race, you know, Demis and Dario and Sam, you know, just kind of like, they all want to be first. And then it's incredibly rough for safety. And then you say, okay, safety regulation. But you

• 2:07:06Speaker 01

  know, it's sort of like, you know, the safety regulation that people talk about, it's like, oh, well, NIST, and they take years and they figure out what the expert consensus is and then they.

• 2:07:13Dwarkesh Patel?

  Isn't that what's going to happen to the project as well?

• 2:07:16Speaker 01

  I think, I mean, I think the sort of alignment angle during the intelligence explosion, it's gonna, you know, it's not a process of like years of bureaucracy and then you can kind of write some standards. I think it looks much more like basically a war and like you have a fog of war. It's like, look, it's like, is it safe to do the next oom, you know? And it's like, ah, you know, like, you know, we're like three ooms into the intelligence explosion. We don't really understand what's going on anymore. You know, the, um, you know, like a bunch of our like generalization scaling curves are like kind of looking not great. You know, some of our like automated AI researchers that are doing alignment are saying it's fine, but we don't quite trust them. In this test, you know, the like, the eyes started doing naughty things and ah, but then we like hammered it out and then it was fine. And, and like, ah, should we, should we go ahead? Should we take, you know, another six months? Also, by the way, you know, like China just stole the weights or we, you know, they're about to like deploy the Romo army, like what do we do?

• 2:08:01Speaker 01

  I think it's this, I think it is this crazy situation. And, um, you know, basically you, you relying much more on kind of like a sane chain of command than you are on sort of some like, you know, deliberative regulatory scheme. I wish you had, you were able to do the deliberative regulatory scheme. And this is the thing about the private companies too. I don't think, you know, they all claim they're gonna do safety, but I think it's really rough when you're in the commercial race and they're startups, you know, and startups, startups are startups. You know, I think they're not fit to handle WMDs.

• 2:08:34Dwarkesh Patel?

  Yeah, I'm coming closer to your position. But part of me also, so with the responsible scaling policies, I was told by people who are advancing that, that the way to think about this, because they know I'm like a libertarian type of person.

• 2:08:49Speaker 01

  Yeah, yeah, yeah.

• 2:08:49Dwarkesh Patel?

  And the way they approached me about it was that fundamentally, this is a way to protect market-based development of AGI in the sense that if you didn't have this at all, then you would have this sort of misuse and then it would have to be nationalized.

• 2:09:06Speaker 01

  Yeah.

• 2:09:06Dwarkesh Patel?

  And the RSPs are a way to make sure that through this deployment, you can still have a market-based order, but then there's these safeguards that make sure that things don't go off the rails.

• 2:09:15Speaker 01

  Yeah.

• 2:09:16Dwarkesh Patel?

  And I wonder if the, if, it seems like your story seems self-consistent, but it does feel, I know this was never your position, so I'm not like, I'm not looping you into this, but they're a sort of, um, modern Bailey almost in the sense of.

• 2:09:35Speaker 01

  Well, look, here, here's what I think about RSP type stuff or sort of safety regulation that's happening now. I think they're important for helping us figure out what world we're in and like flashing the warning signs when we're close, right?

• 2:09:44Dwarkesh Patel?

  And so.

• 2:09:46Speaker 01

  The story we've been telling is sort of like, you know, sort of what I think the modal version of this decade is. But it's like, I think there's lots of ways it could be wrong. I really, you know, we should talk about the data a little more. I think there's like, again, I think there's a world where the stuff stagnates, right? There's a world where we don't have AGI. And so I basically, you know, the RSP thing is like preserving the optionality. Let's see how this stuff goes. But like, we need to be prepared. Like, if the red lights start flashing, if we're getting the automated eye researcher, then it's like, then it's crunch time. And then it's time to go.

• 2:10:11Dwarkesh Patel?

  I think, okay, I can be on the same page on that, that we should have a very, very strong prior on pursuing in a market-based way, unless you're right about what the explosion looks like, the intelligence explosion. And so like, don't move yet, but in that world where like, really does seem like Rutherford can be automated. And that is the only bottleneck to getting TSI. Okay, I think we can leave it at that. I can, I can, yeah, I am somewhat of the way there.

• 2:10:41Speaker 01

  Okay, okay. Yeah, I hope it goes well. It's gotta be, ah, very stressful. And again, right now is the chill time. Enjoy your vacation while it lasts.

• 2:10:52Dwarkesh Patel?

  It's like funny to look out over, just like, this is San Francisco.

• 2:10:57Speaker 01

  Yeah, yeah, yeah. Opening eyes right there, you know, Anthropic's there. I mean, again, this is kind of like, you know, it's like, you guys have this enormous power over how it's, how it's going to go for the next couple of years. And that power is depreciating.

• 2:11:08Dwarkesh Patel?

  Yeah. Who's you guys?

• 2:11:10Speaker 01

  Like, you know, people at labs.

• 2:11:11Dwarkesh Patel?

  Yeah, yeah, yeah.

• 2:11:13Speaker 01

  But it is a sort of crazy world. And you're talking about like, you know, I feel like you talk about like, oh, maybe they'll nationalize too soon. It's like, you know, almost nobody like, really like feels it, sees what's happening. And it's, I think this is the thing that I find stressful about all this stuff is like, look, maybe I'm wrong. Like, if I'm right, we're in this crazy situation where there's like, you know, like a few hundred guys that are like paying attention. And it's daunting.

• 2:11:36Dwarkesh Patel?

  I went to Washington a few months ago.

• 2:11:37Speaker 01

  Yeah.

• 2:11:38Dwarkesh Patel?

  And I was talking to some people who are doing AI policy stuff there.

• 2:11:41Speaker 01

  Yeah.

• 2:11:42Dwarkesh Patel?

  And I was asking them how likely they think nationalization is.

• 2:11:44Speaker 01

  Yeah.

• 2:11:45Dwarkesh Patel?

  And they said, oh, you know, like, It's really hard to nationalize stuff. It's been a long time since we've done it.

• 2:11:50Speaker 01

  Yeah.

• 2:11:50Dwarkesh Patel?

  There's these very specific procedural constraints on what kinds of things can be nationalized. And then I was asked, well, like ASI, so that means because there's, there's constraints at a Defense Production Act or whatever, that that won't be nationalized. There's, the Supreme Court would overturn that. And they're like, yeah, I guess that would be nationalized.

• 2:12:12Speaker 01

  That's the short summary of my post or my view on the project.

• 2:12:21Dwarkesh Patel?

  Okay, so, uh, but before we go further on the ASI stuff, let's just back off.

• 2:12:25Speaker 01

  Okay.

• 2:12:26Dwarkesh Patel?

  You, uh, we began the conversation, I think people will be confused, you graduated valedictorian of Columbia when you were 19.

• 2:12:31Speaker 01

  Uh huh.

• 2:12:32Dwarkesh Patel?

  So you got to college when you were 15.

• 2:12:33Speaker 01

  Right.

• 2:12:34Dwarkesh Patel?

  And you grew, so you were in Germany, then you got to college at 15.

• 2:12:36Speaker 01

  Yeah.

• 2:12:37Dwarkesh Patel?

  Uh, how the fuck did that happen?

• 2:12:40Speaker 01

  I, I really wanted out of Germany.

• 2:12:42Dwarkesh Patel?

  Mm hmm.

• 2:12:43Speaker 01

  Um, I, you know, I went to kind of a, you know, German public school. It was a, it was not a good environment for me.

• 2:12:51Speaker 01

  And, uh, you know, I mean-

• 2:12:52Dwarkesh Patel?

  In what sense? It's just like, no peers that are-

• 2:12:54Speaker 01

  Yeah, look, I mean, it wasn't, yeah, it was, you know, there's, I mean, there's also just a sense in which, um, sort of like, there's this particular sort of German cultural sense. I think in the US, you know, there's all these like amazing high schools and like sort of an appreciation of excellence. And in Germany, there's really this, um, sort of like Paul Poppy syndrome almost, right? Where it's, um, you know, you're the curious kid in class and you want to learn more instead of the teacher being like, ah, that's great. They're like, they kind of resent you for it and they're like trying to crush you and, um, I mean, there's also like, there's no kind of like elite universities for undergraduate, which is kind of crazy.

• 2:13:24Speaker 01

  So, you know, the sort of, you know, there's sort of like, basically like the meritocracy was kind of crushed in Germany at some point. Also, I mean, there's this sort of incredible sense of, you know, complacency, um, you know, across the board. I mean, one of the things that always puzzles me is like, you know, even just going to a US college was this kind of like radical act. And like, you know, it doesn't seem radical to anyone here because it's like, ah, this is obviously the thing you do. And you can go to Columbia, you go to Columbia, but it's, you know, it is very unusual and it's, it's, it's wild to me because it's like, you know, this is where stuff is happening. You can get so much of a better education and, you know, like America is where, you know, it's where, where, where, where all the stuff is. And, um, people don't do it. And, and so, um,

• 2:14:08Speaker 01

  Yeah, anyway, so I, you know, I skipped a few grades and, you know, I think at the time it seemed very normal to me to kind of like go to college at 15 and come to America. I think, you know, now my, one of my sisters is now like turning 15, you know, and so then I, you know, and I look at her and I'm like, now I understand how my mother

• 2:14:27Dwarkesh Patel?

  And then so you get to, you're presumably the only 15 year old there.

• 2:14:30Speaker 01

  Yeah, yeah.

• 2:14:31Dwarkesh Patel?

  As it was just like normal for you to be a 15 year old? Like, what was the initial

• 2:14:34Speaker 01

  It felt so normal at the time. Yeah, so again, it's like now I understand why my mother's worried, you know, I think, you know, I worked, I worked on my parents for a while, you know, eventually I was, you know, I persuaded them. No, but yeah, it felt, felt very normal at the time. And it was great. It was also great because I, you know, I actually really liked college, right? And in some sense, it sort of came at the right time for me, where, you know, I, I mean, I, you know, for example, I really appreciated the sort of like liberal arts education and, you know, like the core curriculum and reading sort of core works of political philosophy and, and literature and

• 2:15:02Dwarkesh Patel?

  You, you did what econ and?

• 2:15:04Speaker 01

  I mean, my majors were math and statistics and economics. But, you know, Columbia has a sort of pretty heavy core curriculum and liberal arts education. And honestly, like, you know, I shouldn't have done all the majors. I should have just, I mean, the best courses were sort of the courses where it's like, there's some amazing professor and it's some history class and it's, I mean, that's, that's honestly the thing I would recommend people spend their time on in college.

• 2:15:26Dwarkesh Patel?

  Was there one professor or class that stood out that way?

• 2:15:28Speaker 01

  I mean, a few. There's like a class by Richard Betts on war, peace and strategy. Adam II is obviously fantastic. And, you know, has written very riveting books.

• 2:15:40Dwarkesh Patel?

  Yeah.

• 2:15:40Speaker 01

  Yeah. You should have him on the podcast, by the way.

• 2:15:42Dwarkesh Patel?

  I've tried.

• 2:15:43Speaker 01

  Okay.

• 2:15:43Dwarkesh Patel?

  Tried. I think you tried for me.

• 2:15:45Speaker 01

  Yeah, you gotta give it on the pod, man. Oh, it'd be so good.

• 2:15:49Dwarkesh Patel?

  Okay, so then, in a couple of years,

• 2:15:52Speaker 01

  Yeah.

• 2:15:53Dwarkesh Patel?

  we were talking to Tyler Cowen recently, and he said that when, the, the way we, he first encountered you,

• 2:16:00Speaker 01

  Yeah.

• 2:16:00Dwarkesh Patel?

  was you wrote this paper on economic growth and existential risk. And he said, I, when I found, read it, I couldn't believe that a 17 year old had written it. I thought if this was a MIT dissertation, I'd be impressed. So you were like, what, what, how did you go from, you're, I guess we were in junior then, you're writing,

• 2:16:19Speaker 01

  I'm a sophomore in junior.

• 2:16:19Dwarkesh Patel?

  You're, you're writing, you know, for pretty novel economic papers. Why, why did you get interested in this, this kind of thing? And what, like, what was the process to get in that?

• 2:16:30Speaker 01

  I don't know. I just, you know, I get interested in things. In some sense, it's sort of like, it feels very natural to me. It's like, I get excited about a thing. I read about it. I immerse myself. I think I can, you know, I can learn information very quickly and understand it. The, I mean, I think to the paper, I mean, I think one actual, at least for the way I work, I feel like sort of moments of peak productivity matter much more than sort of average productivity. I think there's some jobs, you know, like COO or something, you know, like average productivity really matters. But I think there's sort of a, I often feel like I have periods of like, you know, there's some, there's a couple of months where there's sort of nephrosence and I'm like, you know, and the other times I'm sort of computing stuff in the background. And at some point, you know, like writing the series, this is also kind of similar. It's just like you, you write it and, and it's, it's like, it's really flowing. And that's sort of what ends up mattering.

• 2:17:14Dwarkesh Patel?

  I think even for CEOs, it might be the case that the peak productivity is very important. There's

• 2:17:18Speaker 01

  I don't know.

• 2:17:18Dwarkesh Patel?

  one of our, following Chatham House rules, one of our friends in a group chat has pointed out how many famous CEOs and founders have been bipolar manic.

• 2:17:28Speaker 01

  Right, right, right. Yeah.

• 2:17:29Dwarkesh Patel?

  Which is very much the peak, like the call option on your productivity is the most important thing. And you get it by just increasing the volatility through bipolar.

• 2:17:38Dwarkesh Patel?

  Okay. So that's interesting. And so you get interested in economics first. First of all, why economics? Like you could read about anything at this mo, like you, if you wanted, you know, you could, you kind of got a slow start on that, right?

• 2:17:50Dwarkesh Patel?

  You wasted all these years on econ. There's an alternate world where you're like on the super alchemist team at 17 instead of 21 or whatever it was.

• 2:18:03Speaker 01

  I mean, in some sense, I'm still doing economics, right? You know, what is, what is straight lines on a graph? I'm looking at the log, log plots and like figuring out what the trends are. And like thinking about the feedback loops and equilibrium arms control dynamics. And, you know, it's, I think it is a sort of a way of thinking that I find very useful. And, you know, like what, you know, Dario and Ilya seeing scaling early, in some sense, that is a sort of very economic way of thinking. Also the sort of physics, kind of like empirical physics, you know, a lot of them are physicists. I think the economists usually can't code well enough and that's their issue. But I think it's that sort of way of thinking. I mean, the other thing is, you know, I, I thought there were sort of.

• 2:18:38Speaker 01

  You know, I thought of a lot of the sort of like core ideas of economics I thought were just beautiful. And, you know, in some sense, I feel like I was a little duped, you know, where it's like, actually, econ academia is kind of decadent now. You know, I think that, you know, for example, the paper I wrote, you know, it's sort of, I think the takeaway, you know, it's a long paper, it's 100 pages of math or whatever. I think the core takeaway, I can, you know, kind of give the core intuition for in like, you know, 30 seconds and it makes sense. And it's, and it's like, you don't actually need the math.

• 2:19:03Dwarkesh Patel?

  Yeah.

• 2:19:03Speaker 01

  I think that's the sort of, the best pieces of economics are like that. Where you do the work, but you do the work to kind of uncover insights that weren't obvious to you before. Once, once you've done the work, it's like some sort of like mechanism falls out of it that like, makes a lot of crisp intuitive sense that like, explains some facts about the world that you can then use in arguments. I think, you know, I think, you know, like a lot of econ 101 like this, and it's great, a lot of econ in the, you know, in the 50s and the 60s, you know, was like this. And, you know, Chad Jones' papers are often like this. I really like Chad Jones' papers for this. You know, I think, you know, why did I ultimately not pursue econ academia was a number of reasons. One of them was Tyler Cowen. You know, he was kind of took me aside and he was kind of like, look, I

• 2:19:46Speaker 01

  think you're one of the like, top young economists I've ever met, but also you should probably not go to grad school.

• 2:19:50Dwarkesh Patel?

  Oh, interesting. Really? I didn't realize that.

• 2:19:52Speaker 01

  Well, yeah. And it was, it was, it was good because he kind of introduced me to the, you know, I don't know, like the Twitter weirdos or just like, you know, and I think the takeaway from that was kind of, you know, got to move out west one more time.

• 2:20:03Dwarkesh Patel?

  Wait, Tyler introduced you to the Twitter weirdos?

• 2:20:05Speaker 01

  A little bit, yeah. Or just kind of like the sort of, you know,

• 2:20:07Dwarkesh Patel?

  Like this 60 year old, old economist introduced you to that Twitter?

• 2:20:11Speaker 01

  Yeah. Well, you know, I, I had been, I so went from Germany, you know, completely, you know, on the periphery to kind of like, you know, in a US elite institution and sort of got, got some vibe of like, sort of, you know, meritocratic elite, you know, US society. And then sort of, yeah, basically this sort of like, there was a sort of directory then to being like, look, I, you know, find the true American spirit. I got to come out here. But the other, the other reason I didn't become economist was because, or at least econ academia, was because I think sort of econ academia has become a bit decadent. And maybe it's just ideas getting harder to find. And maybe it's sort of things, you know, and the sort of beautiful, simple things have been discovered. But, you know, like, what are econ papers these days? You know, it's like, you know, it's like, 200 pages of like empirical analyses on what happened when, you know, like Wisconsin bought, you know, 100,000 more textbooks on like educational outcomes. And I'm really happy that work

• 2:20:55Speaker 01

  happened. I think it's important work, but I think it is not in government, covering these sort of like fundamental insights and sort of mechanisms in society. Or, you know, it's like even the theory work is kind of like, here's a really complicated model and the model spits out, you know, if the Fed does X, you know, then Y happens. You have no idea what that had, why that happened because it's like gazillion parameters and they're all calibrated in some way and it's some computer simulation. You have no idea about the validity, you know? Yeah. So I think, I think the sort of, you know, the most important insights are the ones where you have to do a lot of work to get them, but then there's sort of this crisp intuition.

• 2:21:25Dwarkesh Patel?

  Yeah. The P versus NP of.

• 2:21:27Speaker 01

  Sure.

• 2:21:28Dwarkesh Patel?

  Yeah.

• 2:21:30Dwarkesh Patel?

  That's really interesting. So just going back to your time in college.

• 2:21:34Speaker 01

  Yeah.

• 2:21:35Dwarkesh Patel?

  You say that peak productivity kind of explains the, this paper and things, but the valedictorian, that's getting straight A's or whatever, is very much a, a, a average productivity phenomenon, right?

• 2:21:47Speaker 01

  So there's two, there's one award for the highest GPA, which I won, but the valedictorian is like among the people which have the highest GPA and then like selected by faculty.

• 2:21:54Dwarkesh Patel?

  Okay. Yeah. So it's just not, it's not just peak productivity.

• 2:21:58Speaker 01

  It's just, it's, it's just, it's, I generally just love this stuff, you know? I just, I was curious and I thought it was really interesting and I love learning about it and, and I love kind of like, it made sense to me and, you know, it was, it was very natural and, you know, I think I'm, you know, I'm not, you know, I think one of my faults is I'm not that good at eating glass or whatever. I think there's some people who are very good at it. I think the sort of like, the sort of moments of peak productivity come when I, you know, I'm just really excited and engaged and, and, and, and, and love it. And, you know, I, I, you know, if you take the like courses, you know, that's what you got in college.

• 2:22:29Dwarkesh Patel?

  Yeah. It's, it's, it's the Bruce Banner code in Avengers. You know, I'm always angry. I'm always excited. I'm always curious. That's why I'm always be productivity. So it's interesting, by the way, when you were in college, I was also in college. I think you were, despite being a year younger than me, I think a year, you're ahead in college than me, or at least two, maybe two years ahead.

• 2:22:53Dwarkesh Patel?

  And we met around this time.

• 2:22:55Speaker 01

  Yeah, yeah, yeah.

• 2:22:56Dwarkesh Patel?

  We also met, I think, through the Tyler Cowen universe.

• 2:23:00Speaker 01

  Yeah, yeah.

• 2:23:00Dwarkesh Patel?

  And it's very insane how small the world is.

• 2:23:03Speaker 01

  Yeah.

• 2:23:03Dwarkesh Patel?

  I think I, did I reach out to you? I must have.

• 2:23:06Speaker 01

  Yes, I don't know.

• 2:23:07Dwarkesh Patel?

  About when I had a couple of videos and they had a couple hundred views or something.

• 2:23:11Speaker 01

  Yeah, it's a small world.

• 2:23:13Dwarkesh Patel?

  Yeah.

• 2:23:13Speaker 01

  I mean, this is the crazy thing about the AI world, right? It's kind of like, it's the same few people at the KF parties and they're the ones, you know, running the models at DeepMind and, you know, OpenAI and Anthropic. And, and, you know, I mean, I think some other friends of ours have mentioned this who are now later in their career and very successful that, you know, they actually met all the people who are also kind of very successful in Silicon Valley now, like, you know, when they're, when they're in their, you know, when, before their twenties or when they're really twenties. The, I mean, look, I actually think, you know, and why is it a small world? I mean, I think one of the things is some amount of like, you know, some sort of agency. And I think in a funny way, this is a thing I sort of took away from the sort of Germany

• 2:23:53Speaker 01

  experience where it was, I mean, look, I, I, it was crushing. I really didn't like it. And it was like, it was such an unusual move to kind of skip grades and such an unusual move to come to the United States. And, you know, a lot of these things I did were kind of unusual moves. And, you know, there's some amount where like, just like, just trying to do it, and then it was fine and it worked, that kind of reinforced like, you know, you don't, you don't just have to kind of conform to what the Overton window is. You can just kind of try to do the thing, the thing that seems right to you. And like, you know, most people can be wrong. And, I don't know, things like that. And I think that was kind of a, you know, valuable, kind of like early experience that was sort of formative.

• 2:24:33Dwarkesh Patel?

  Okay, so after college, what did you do?

• 2:24:35Speaker 01

  I did econ research for a little bit, you know, at Oxford and stuff. And then, then I worked at Future Fund.

• 2:24:40Dwarkesh Patel?

  Mm hmm.

• 2:24:41Speaker 01

  Yeah.

• 2:24:42Dwarkesh Patel?

  Okay, so, and tell me about it.

• 2:24:44Speaker 01

  Yeah, yeah, okay, okay. Yeah, Future Fund was a, you know, was a foundation that was, you know, funded by Sam Bankman-Fried. I mean, we were our own thing, you know, we were based in the Bay. You know, at the time, this was in sort of early 22. It was this incredibly exciting opportunity, right? It was basically like a startup foundation, which is like, you know, it doesn't come along that often that, you know, we thought would be able to give away billions of dollars. Thought would be able to kind of like, you know, remake how philanthropy is done, you know, from first principles. That would be able to have, you know, this like great impact, you know, the causes we focused on were, you know, biosecurity, AI, you know, finding exceptional talent and putting them to work on hard problems.

• 2:25:28Speaker 01

  And, you know, like a lot of the stuff we did, I was, I was really excited about, you know, like academics who would, you know, usually take six months would send us emails like, ah, you know, this is great. This is so quick and, you know, and straightforward. You know, in general, I feel like I've often find that with like, you know, a little bit of encouragement, a little bit of sort of empowerment, kind of like removing excuses, making the process easy, you know, you can kind of like get people to do great things.

• 2:25:49Dwarkesh Patel?

  I think on the Future Fund, the thing is context for people who might not realize.

• 2:25:54Speaker 01

  Yeah.

• 2:25:54Dwarkesh Patel?

  Not only were you guys planning on deploying billions of dollars, but as a team of four people.

• 2:26:00Speaker 01

  Yeah, yeah, yeah.

• 2:26:00Dwarkesh Patel?

  So you at 18 are on a team of four people.

• 2:26:04Speaker 01

  Yeah.

• 2:26:05Dwarkesh Patel?

  That is in charge of deploying billions of dollars.

• 2:26:07Speaker 01

  Yeah. I mean, just, I mean, yeah, Future Fund, you know, the, yeah, I mean, so that was, that was sort of the heyday, right? And then obviously, you know, when, when in sort of, you know, November of 22, you know, it was kind of revealed that Sam was this, you know, giant fraud. And from one day to the next, you know, the whole thing collapsed.

• 2:26:26Speaker 01

  That was just really tough. I mean, you know, obviously it was devastating. It was devastating, obviously for the people at their money in FTX, you know, closer to home, you know, all the, you know, all these grantees, you know, we'd wanted to help them and we thought they were doing amazing projects. And so, but instead of helping them, we ended up saddling them with like a giant problem.

• 2:26:45Speaker 01

  You know, personally, it was, you know, it was a startup, right? And so I, you know, I'd worked 70 hour weeks every week for, you know, basically a year on this to kind of build this up, you know, we're a tiny team. And then from one day to the next, it was all gone and not just gone, it was associated with this giant fraud.

• 2:27:00Speaker 01

  And so, you know, that was incredibly tough.

• 2:27:02Dwarkesh Patel?

  Mm hmm.

• 2:27:03Speaker 01

  Yeah.

• 2:27:04Dwarkesh Patel?

  And then were there any signs early on that SBF was?

• 2:27:09Speaker 01

  Yeah, like, obviously, I didn't know he was a fraud and the whole, you know, I would have never worked there if I did, you know? And, you know, we weren't, you know, we were a separate thing. We weren't with the, working with the business. I mean, I think, I do think there were some takeaways for me. I think one takeaway was, you know, I think there's a, I had this tendency, I think people in general have this tendency to kind of like, you know, give successful CEOs a pass on their behavior because, you know, they're successful CEOs and that's how they are. And that's just successful CEO things. And, you know, I didn't know Sam Bankman-Fried was a fraud, but I knew SBF and I knew he was extremely risk-taking, right? I knew he, he was narcissistic.

• 2:27:49Speaker 01

  He didn't tolerate disagreement well, you know, sort of by the end, he and I just like, didn't get along well. And sort of, I think the reason for that was like, there's some biosecurity grants he really liked because they're kind of cool and flashy. And at some point I'd kind of run the numbers and it didn't really seem that cost-effective and I pointed that out and he was pretty unhappy about that. And so I knew his character. And I think, you know, I feel like one takeaway for me was,

• 2:28:14Speaker 01

  was, you know, like, I think it's really worth paying attention to people's character, including like people you work for and successful CEOs. And, you know, that can save you a lot of pain down the line.

• 2:28:26Dwarkesh Patel?

  Okay, so after that, FTX implodes and you're out. And then you got into,

• 2:28:35Dwarkesh Patel?

  you, you went to OpenAI, the super alignment team had just started. I think you were, you were like part of the initial team. And so what, what was the original idea? What was compelling about that for you to join?

• 2:28:47Speaker 01

  Yeah, totally. So, I mean, what was the goal of the super alignment team? You know, the alignment team at OpenAI, you know, at, you know, other labs, sort of like several years ago, kind of had done sort of basic research and they developed RHF, reinforcement learning from human feedback. And that was a sort of a, you know, ended up being really successful technique for controlling sort of current generation of AI models.

• 2:29:10Speaker 01

  What we were trying to do was basically kind of be the basic research bet to figure out what is the successor to RLHF. And the reason we needed that is, you know, basically, you know, RLHF probably won't scale to superhuman systems. RLHF relies on sort of human raiders who kind of thumbs up, thumbs down, you know, like the model said something, it looks fine, looks good to me. At some point, you know, the superhuman models, the superintelligence is going to write, you know, a million lines of, you know, crazy complex code. You don't know at all what's going on anymore. And so how do you kind of steer and control these systems? How do you add side constraints? You know, the reason I joined was I thought this was an important problem and I thought it was just a really solvable problem, right? I thought this was basically, you know, there's, I think there's a, I still do, I mean, even more so do, I think there's a lot of just really promising sort of ML research on alignment, on sort of aligning superhuman

• 2:29:55Speaker 01

  systems. And maybe we should talk about that a bit more later. But so

• 2:30:00Dwarkesh Patel?

  It was so solvable, you solved it in a year.

• 2:30:02Speaker 01

  Yeah, I think quite so.

• 2:30:06Speaker 01

  Anyway, so look, OpenAI wanted to do this like really ambitious effort on alignment and, you know, Elliot was backing it and, you know, I liked a lot of the people there. And so I was, you know, I was really excited and I was kind of like, you know, I think there was a lot of people, sort of, on alignment, there's always a lot of people kind of making hay about it. And, you know, I appreciate people highlighting the importance of the problem. And I was just really into like, let's just try to solve it. You know, let's do the ambitious effort, you know, let's do the, you know, operation warp speed for solving alignment. And it seemed like an amazing opportunity to do so.

• 2:30:34Dwarkesh Patel?

  Okay. And now, basically, the team doesn't exist. I think the head of it has left. Both heads have had to have left, Yan and Ilya. That's been the news of last week. What happened? Why did the thing break down?

• 2:30:48Speaker 01

  I think OpenAI sort of decided to take things in a somewhat different direction.

• 2:30:53Dwarkesh Patel?

  Meaning what? I mean, that super alignment isn't the best way to frame the

• 2:30:58Speaker 01

  No, I mean, look, obviously, sort of after the November board events, you know, there were personnel changes. I think Ilya leaving was just incredibly tragic for OpenAI. And You know, I think some amount of repartization, I think some amount of, you know, I mean, there's been some reporting on the SuperLime compute commitment, you know, there's this 20% compute commitment as part of, you know, how a lot of people recruited, you know, it's like, we're going to do this ambitious effort on alignment. And, um, you know, some amount of, uh, you know, not keeping that and deciding to go in a different direction.

• 2:31:25Dwarkesh Patel?

  Hmm. Okay. So now Jan has left, Ilya has left. So this team itself has dissolved, but you were the sort of first person who left or was forced to leave. You were, the information reported that you were fired for leaking. But what happened? Was this accurate?

• 2:31:41Speaker 01

  Yeah. Um, look, why don't I, why don't I tell you what they claim I leaked? And you can tell me what you think. Um, yeah. So OpenAI did claim to employees that I was fired for leaking. Uh, and you know, I and others have sort of pushed them to say what the leak is. And so here's their response in full. Um, you know, sometime last year, I had, um, uh, written a sort of brainstorming document on preparedness, on safety and security measures we'd need in the future on the past AGI. Um, and I'd shared that with three external researchers for feedback. So that's it. That's the leak. Um, you know, I think for context, it was totally normal at OpenAI at the time to share sort of safety ideas with external researchers for feedback. Um, you know, it happened all the time.

• 2:32:22Speaker 01

  Um, you know, the doc was sort of my ideas, you know, before I shared it, I reviewed it for anything sensitive. Um, the internal version had a reference to a future cluster, but I redacted that for the external copy. Um, you know, there's a link in there to some, to some slides of mine, internal slides. Um, but you know, that was a dead link to the external people I shared it with. You know, the slides weren't shared with them. And, um, so obviously I, um, I pressed them to sort of tell me, what is the confidential information in this document? And what they came back with was a line in the doc about planning for AGI by 27, 28 and not setting timelines for preparedness.

• 2:33:02Speaker 01

  You know, I wrote this doc, you know, a couple of months after the super alignment announcement. We had put out, you know, this sort of four year planning horizon. I didn't think that planning horizon was sensitive. You know, it's, it's the sort of thing Sam says publicly all the time.

• 2:33:15Speaker 01

  I think sort of John said on the podcast a couple of weeks ago. Anyway, so that's it.

• 2:33:20Dwarkesh Patel?

  That's it? So that seems pretty thin for, if, if the cause was leaking, that seems pretty thin. Was there anything else to it?

• 2:33:28Speaker 01

  Yeah, I mean, so that was, that was the leaking claim. I mean, say a bit more about sort of what happened in the firing. Yeah. So one thing was last year I had written a memo, internal memo about opening I security. I thought it was egregiously insufficient. You know, I thought was insufficient to protect, you know, the theft of Model Eights or, or key algorithmic secrets from foreign actors. So I wrote this memo. I shared it with a few colleagues, a couple of members of leadership who sort of mostly said it was helpful.

• 2:33:58Speaker 01

  But then, you know, a couple of weeks later, a sort of major security incident occurred. And that prompted me to share the memo with a couple of members of the board. And so after I did that, you know, days later, it was made very clear to me that leadership was very unhappy with me having shared this memo with the board. You know, apparently the board had hassled leadership about security. And then I got sort of an official HR warning for this memo, you know, for sharing it with the board. The HR person told me it was racist to worry about CCP espionage. And they said it was sort of unconstructive.

• 2:34:33Speaker 01

  And, you know, look, I think I probably wasn't at my most diplomatic. You know, I definitely could have been more politically savvy. But, you know, I thought it was a really, really important issue and, you know, the, the security incident had made me really worried. Anyway, and so I guess the reason I bring this up is when I was fired, it was sort of made very explicit that the security memo is a major reason for my being fired. You know, I think it was something like, you know, the reason that this is a firing and not a warning is because of the security memo.

• 2:34:59Dwarkesh Patel?

  But you sharing it with the board?

• 2:35:01Speaker 01

  The warning I'd gotten for the security memo.

• 2:35:05Speaker 01

  Anyway, and I mean, some other, you know, what might also be helpful context is the sort of questions they asked me when they fired me. So you know, this was a bit over a month ago. I was pulled aside for a chat with a lawyer, you know, that quickly turned very adversarial. And, you know, the questions were all about my views on AI progress, on AGI, on the level of security appropriate for AGI, on, you know, whether government should be involved in AGI, on, you know, whether I and super alignment were loyal to the company, on, you know, what I was up to during the OpenAI board events, you know, things like that. And, you know, then they chatted to a couple of my colleagues, and then they came back and told me I was fired.

• 2:35:48Speaker 01

  And, you know, they'd gone through all of my digital artifacts from the time at my, you know, time at OpenAI, you know, messages, docs, and that's when they found, you know, the leak. Yeah. And so, anyway, so the main claim they made was this leaking allegation, you know, that's what they told employees. They, you know, the security memo. There's a couple of other allegations they threw in. One thing they said was that I was unforthcoming during the investigation because I didn't initially remember who I'd shared the doc with, the sort of preparedness brainstorming doc, only that I had sort of spoken to some external researchers about these ideas. And, you know, look, the doc was over six months old, you know, I'd spent a day on it. You know, it was a Google Doc I shared with my OpenAI email. It wasn't a, you know, screenshot or anything I was trying to hide. It simply

• 2:36:30Speaker 01

  didn't stick because it was such a non-issue. And then they also claimed that I was engaging on policy in a way that they didn't like. And so what they cited there was that I had spoken to a couple of external researchers, you know, somebody at a think tank about my view that AGI would become a government project, you know, as we discussed. You know, in fact, I was speaking to lots of sort of people in the field about that at the time. I thought it was a really important thing to think about. Anyway, and so they found, you know, they found a DM that I'd written to like a friendly colleague, you know, five or six months ago, where I relayed this, and, you know, they cited that. And, you know, I had thought it was well within OpenAI norms to kind of talk about high-level issues on the future of AGI with external people in the field. So that's what they alleged. That's what happened.

• 2:37:15Speaker 01

  You know, I've spoken to kind of a few dozen former colleagues about this, you know, since I think the sort of universal reaction is kind of like, you know, that's insane. I was sort of surprised as well, you know, I I had been promoted just a few months before. I think, you know, I think Ilya's comment for the promotion case at the time was something like, you know, Leopold's amazing. We're lucky to have him.

• 2:37:40Speaker 01

  But look, I mean, I think the thing I understand, and I think in some sense is reasonable is like, you know, I think I ruffled some feathers and, you know, I think I was I was probably kind of annoying at times, you know, it's like I security stuff and I kind of like repeatedly raised that and maybe not always in the most diplomatic way. You know, I didn't sign the employee letter during the board events, you know, despite pressure to do so.

• 2:38:03Dwarkesh Patel?

  And you were what, one of like eight people or something?

• 2:38:05Speaker 01

  Not that many people. I guess the I think the sort of two senior most people didn't sign were Andre and Jan.

• 2:38:14Speaker 01

  And, you know, I mean, on the letter, by the way, I by the time on sort of Monday morning, when that letter was going around, I think probably it was appropriate for the board to resign. I think they'd kind of like lost too much credibility and trust with the employees. But I thought the letter had a bunch of issues. I mean, I think one of them was it just didn't call for an independent board. I think it's sort of like basics of corporate governance to have an independent board. Anyway, you know, there's other things, you know, I in sort of other discussions, I pressed leadership for sort of OpenAI to abide by its public commitments. You know, I raised a bunch of tough questions about whether it was consistent with the OpenAI mission and consistent with the national interest to sort of partner with authoritarian dictatorships to build the core infrastructure for AGI.

• 2:38:57Speaker 01

  So, you know, look, you know, it's a free country, right? That's what I love about this country, you know, we talked about it. And so they have no obligation to keep me on staff.

• 2:39:08Speaker 01

  And, you know, I think in some sense, I think it would have been perfectly reasonable for them to come to me and say, look, you know, we're taking the company in a different direction. You know, we disagree with your point of view. You know, we don't trust you enough to sort of toe the company line anymore. And, you know, thank you so much for your work at OpenAI, but I think it's time to part ways. I think that would have made sense. I think, you know, we did start sort of materially diverging on sort of views on important issues. I'd come in very excited and aligned with OpenAI, but that sort of changed over time.

• 2:39:38Speaker 01

  Look, I think I think there would have been a very amicable way to part ways. And I think it's a it's a bit of a shame that it sort of this is the way it went down.

• 2:39:46Speaker 01

  You know, all that being said, I think, you know, I really want to emphasize. There's just a lot of really incredible people at OpenAI and it was incredible privilege to work with them. And, you know, overall, I'm just extremely grateful for my time there.

• 2:40:00Dwarkesh Patel?

  When you left, now that there's now there's been reporting about an NDA that former employees have to sign in order to have access to their vested equity. Did you sign such NDA?

• 2:40:15Speaker 01

  No. My situation was a little different in that I was sort of I was basically right before my cleft. But then, you know, they still offered me the equity. But I didn't want to sign a non disparagement, you know, freedom is priceless.

• 2:40:28Dwarkesh Patel?

  And how much was how much was the equity?

• 2:40:29Speaker 01

  Like, close to a million dollars.

• 2:40:32Dwarkesh Patel?

  So it was definitely a thing you were you and others were aware of that this is like a choice that OpenAI is explicitly offering you.

• 2:40:40Speaker 01

  Yeah.

• 2:40:40Dwarkesh Patel?

  And presumably, the person on OpenAI's staff knew that we're offering them equity, but they had to sign this NDA that has these conditions that you can't, for example, give the kind of statements about your thoughts on AGI and OpenAI that you're giving on this podcast right now.

• 2:40:56Speaker 01

  Look, I don't know what the whole situation is. I certainly think sort of vested equity is pretty rough if you're conditioning that on some NDA. It might be a somewhat different situation if it's a sort of severance agreement.

• 2:41:05Dwarkesh Patel?

  Right. But an OpenAI employee who had signed it presumably could not give the podcast that you're giving today.

• 2:41:12Speaker 01

  Quite possibly not. Yeah, I don't know.

• 2:41:16Dwarkesh Patel?

  Okay, so analyzing the situation here, I guess if you were to yeah, the board thing is really tough because if you were trying to defend them, you would say, well, listen, you were just kind of going outside the regular chain of command, and maybe there's a point there. Although the way in which the person from HR thinks that you have an adversarial relationship with or you're you're supposed to have an adversarial relationship with the board where to take give the board some information which is relevant to

• 2:41:47Dwarkesh Patel?

  Whether OpenAI is fulfilling its mission and whether it can do that in a better way is part of the leak, as if the board is that is supposed to ensure that OpenAI is following its mission as some sort of external actor. That seems pretty...

• 2:42:00Speaker 01

  I mean, I think, I think, I mean, to be clear, the leak allegation was just that sort of document I shared for feedback. This is just sort of a separate thing that they cited, and they said, "I wouldn't have been fired if not for the security memo."

• 2:42:09Dwarkesh Patel?

  They said you wouldn't have been fired.

• 2:42:11Speaker 01

  They said, "The reason this is a firing and not a warning is because of the warning you had gotten for the security memo."

• 2:42:15Dwarkesh Patel?

  Oh. Before you left, the incidents with the board happened, where Sam was fired and then rehired as CEO, and now he's on the board. Now, Ilya and Yan, who are the heads of the Superalimen team, and Ilya, who is a co-founder of OpenAI, obviously the most significant in terms of stash or a member of OpenAI from a research perspective, they've left. Seems like, especially with regards to Superalimen stuff and just generally with OpenAI, a lot of this sort of personnel drama has happened over the last few months. What's going on?

• 2:42:49Speaker 01

  Yeah, there's a lot of drama about OpenAI. Yeah, so why is there so much drama?

• 2:42:56Speaker 01

  You know, I think there would be a lot less drama if all OpenAI claimed to be was sort of building ChatGPT or building business software.

• 2:43:04Speaker 01

  I think where a lot of the drama comes from is, you know, OpenAI really believes they're building AGI, right? And it's not just, you know, a claim they make for marketing purposes, you know, whatever. You know, there's this report that Sam is raising, you know, $7 trillion for chips. And it's like, that stuff only makes sense if you really believe in AGI.

• 2:43:23Speaker 01

  And so I think what gets people sometimes is sort of the cognitive dissonance between sort of really believing in AGI, but then sort of not taking some of the other implications seriously. You know, it's, this is going to be incredibly powerful technology, both for good and for bad. And that implicates really important issues, like the national security issues we spoke about. Like, you know, are you protecting the secrets from the CCP? Like, you know, does America control the core AGI infrastructure? Or does it, you know, a Middle Eastern dictator control the core AGI infrastructure?

• 2:43:50Speaker 01

  And then, I mean, I think the thing that, you know, really gets people is the sort of tendency to kind of then make commitments and sort of like, you know, they say they take these issues really seriously, they make big commitments on them, but then sort of frequently don't follow through, right? So, you know, again, as mentioned, there's this commitment around super alignment compute, you know, sort of 20% of compute for this long-term safety research effort. And I think, you know, you and I could have a totally reasonable debate about what is the appropriate level of compute for super alignment. But that's not really the issue. The issue is that this commitment was made and it was used to recruit people and, you know, it was, it was very public. And it was made because, you know, there's a recognition that there would always be something more urgent than a long-term safety

• 2:44:32Speaker 01

  research effort, you know, like some new product or whatever. And, but then in fact, they just, you know, really didn't keep the commitment. And so, you know, there was always something more urgent than long-term safety research. I mean, I think, I think another example of this is, um, you know, when I raised these issues about security, you know, they would, they would tell me, you know, security is our number one priority. But then, you know, invariably, when, when it came time to sort of invest serious resources, when it came time to make trade-offs, to sort of take some pretty basic measures, security would not be prioritized. And so, yeah, I think it's the cognitive dissonance and I think it's the, the sort of unreliability that causes a bunch of the drama.

• 2:45:12Dwarkesh Patel?

  So let's zoom out, talk about the part, a big part of the story and also a big motivation of the way in which it must proceed with regards to geopolitics and everything, is that once you have the AGI, pretty soon after you proceed to ASI because, superintelligence, because you have these AGIs which can function as researchers into further AI progress. And within a matter of years, maybe less, you go to something that is like superintelligence. And at the high, and then from there, then you do, according to your story, do all this research and development into robotics and pocket nukes and whatever other crazy shit.

• 2:45:52Speaker 01

  Yeah.

• 2:45:52Dwarkesh Patel?

  But at a high.

• 2:45:55Dwarkesh Patel?

  Okay, but there's, I was skeptical of this story for many reasons.

• 2:45:59Speaker 01

  Yes.

• 2:46:00Dwarkesh Patel?

  At a high level.

• 2:46:01Speaker 01

  Yes.

• 2:46:01Dwarkesh Patel?

  It's not clear to me this input-output model of research is how things actually happen in research. We can, you look at economy-wide, right? Patrick Collison and others have made this point that compared to 100 years ago, we have 100x more researchers in the world. It's not like progress is happening 100x faster. So it's clearly not the case that you can just pump in more population into research and you get higher research on the other end. I don't know why it would be different for the AI researchers themselves.

• 2:46:31Speaker 01

  Okay, great. So this is getting into some good stuff. This is the classic disagreement I have with Patrick and others. Okay, so, you know, obviously inputs matter, right? So it's like, the United States produces a lot more scientific and technological progress than, you know, Liechtenstein, right? Or Switzerland. And even if I made, you know, Patrick Collison dictator of like Liechtenstein or Switzerland, and Patrick Collison was able to implement his like, you know, utopia of ideal institutions, keeping the talent pool fixed. He's not able to like, do some crazy high school immigration thing, or like, you know, whatever, some like crazy genetic breeding scheme, or whatever he wants to do.

• 2:47:05Speaker 01

  Keeping the talent pool fixed, but amazing institutions. I claim that still, even if you made Patrick Collison dictator Switzerland, maybe you get some factor, but Switzerland is not going to be able to outcompete the United States in scientific and technological parts. Obviously, magnitude's matter. Okay.

• 2:47:19Dwarkesh Patel?

  No, actually, I'm not sure I agree with this. There's been many examples in history where you have small groups of people.

• 2:47:25Speaker 01

  Yes.

• 2:47:25Dwarkesh Patel?

  Who are part of like Bell Labs, or Skunk Works, or something. It's a couple hundred researchers. OpenAI, right? Couple hundred researchers, they do.

• 2:47:32Speaker 01

  Highly selected, though, right? You know, it's like, it's like saying, you know,

• 2:47:35Dwarkesh Patel?

  Well, that's part of, that's part of why Patrick Collison as a dictator is gonna do a good job of this.

• 2:47:39Speaker 01

  Well, yes, if you can highly select all the best AI researchers in the world, you might only need a few hundred. But if you, you know, that's, that's the talent pool. It's like you have the, you know, 300 best AI researchers in the world.

• 2:47:48Dwarkesh Patel?

  But, but there's, there has been, it's not a case that from 100 years to now, there haven't been, the population has increased massively. A lot of the, in fact, you would expect the density of talent to have increased in the sense that malnutrition and other kinds of, poverty, whatever, that have debilitated past talent at the same sort of level. No longer delivitated in the same way.

• 2:48:06Speaker 01

  Yeah, so to the 100x point, right? So I don't know if it's 100x. I think it's easy to inflate these things, probably at least 10x. And so people are sometimes like, ah, you know, like, you know, come on, ideas haven't gotten that much harder to find, you know? Why would you have needed this 10x increase in research effort? Whereas to me, I think this is an extremely natural story. And why is it a natural story? It's a straight line on a log log plot. This is sort of a, you know, deep learning researcher's dream, right? What is this log log plot? On the x-axis, you have log cumulative research effort. On the y-axis, you have log GDP or ooms of algorithmic progress, or, you know, log transistors per square inch, or, you know, in the sort of experience curve for solar, kind of like, you know, whatever the the log of, you know, the price for a gigawatt of solar. And it's extremely natural for that to be a straight line, you know? This is sort of a class, yeah, it's a classic.

• 2:48:50Speaker 01

  And, you know, it's basically the first thing is very easy. Then basically, you know, you have to have log increments of cumulative research effort to find the next thing. And so, you know, in some sense, I think this is a natural story. Now, one objection kind of people then make is like, oh, you know, isn't it suspicious, right? That like ideas, you know, well, we increased research effort 10x and ideas also just got 10x harder to find. And so it perfectly, you know, equilibrates. And to there, I say, you know, it's just, it's an equilibrium. It's an adage equilibrium, right? So this is like, you know, isn't it a coincidence that supply equals demand, you know, and the market clears, right? And that's, and the same thing here, right? So it's, you know, ideas getting, how much ideas have gotten harder to find is a function of how much progress you've made.

• 2:49:32Speaker 01

  And then, you know, what the overall growth rate has been is a function of how much ideas have gotten harder to find in ratio to how much you've been able to like increase research effort. What is the sort of growth of log cumulative research effort? So in some sense, I think the story is sort of like fairly natural. And you see this, you see this not just economy wide, you see it in kind of experience curve for all sorts of individual technologies.

• 2:49:51Speaker 01

  So I think there's some process like this. I think it's totally plausible that, you know, institutions have gotten worse by some factor. Obviously, there's some sort of exponent of diminishing returns on more people, right? So like serial time is better than just paralyzing. But still, I think it's like clearly inputs matter.

• 2:50:06Dwarkesh Patel?

  Yeah, I agree, but if the coefficient of

• 2:50:11Dwarkesh Patel?

  how fast it diminish as you grow, the input is high enough, then the, in the abstract, the fact that inputs matter isn't that relevant. Okay, so, I mean, we're talking at a very high level, but just like take it down to the actual concrete thing here. OpenAI has a staff of at most low hundreds who are directly involved in the algorithmic progress in future models. If it was really the case that you could just arbitrarily scale this number and you could have much faster algorithmic progress and that would result in much higher, much better AIs for OpenAI, basically, then it's not clear why OpenAI doesn't just go out and hire every single person with 150 IQ, of which there are hundreds of thousands in the world.

• 2:50:51Dwarkesh Patel?

  And my, my story there is there is transaction cost to managing all these people that don't just go away if you have a bunch of AIs, that there, these tasks aren't easy to parallelize. And I think you, I'm not sure how you would explain the fact of like, why doesn't OpenAI go on a recruiting binge of every single genius in the world?

• 2:51:12Speaker 01

  All right, great. So let's talk about the OpenAI example and let's talk about the automated AI researchers. So, I mean, in the OpenAI case, I mean, just, you know, just kind of like look at the inflation of like AI researcher salaries over the last year. I mean, I think like, I don't know, I don't know what it is, you know, 4X, 5X, it's kind of crazy. So they're clearly really trying to recruit the best AI researchers in the world. And, you know, I don't know, it's, they do find the best AI researchers in the world. And I think my response to your thing is like, you know, almost all of these 150 IQ people, you know, if you just hired them tomorrow, they wouldn't be good AI researchers. They wouldn't be an Alec Radford.

• 2:51:39Dwarkesh Patel?

  But they're willing to make investments that take yours to pan out of the, the data centers they're buying right now will come online in 2026 or something. Why wouldn't they be able to make every 150 IQ person, some of them won't work out, some of them won't have the traits we like.

• 2:51:53Speaker 01

  Yeah.

• 2:51:54Dwarkesh Patel?

  But some of them by 2026 will be amazing AI researchers. Why, why, why aren't they making that bet?

• 2:51:58Speaker 01

  Yeah. And so sometimes this happens, right? Like smart physicists have been really good at AI research, you know, it's like all the anthropocopians.

• 2:52:03Dwarkesh Patel?

  But, but like if you talk to, I had Daria on the podcast, I'm like, they have this very careful policy of like, we're not going to just hire arbitrarily, we're going to be extremely selective.

• 2:52:11Speaker 01

  Yep. Training is not as easily scalable, right? So training is very hard, you know, if you, if you just hired, you know, a hundred thousand people, it's like, I mean, you couldn't train them all. It'd be really hard to train them all. You know, you wouldn't be doing any AI research. Like, you know, there's, there's huge costs to bringing on a new person training them. This is very different with AIs, right? And I think this is, it's really important to talk about the sort of like advantages that AIs will have. So it's like, you know, training, right? It's like, what does it take to be an Alec Radford? You know, you need to be in a really good engineer, right? The AIs, they're going to be an amazing engineer. They're going to be amazing at coding. You can just train them to do that. They need to have, you know, not just be a good engineer, but have really good research intuitions and like really understand deep learning. And this is stuff that, you know, Alec Radford, or, you know, somebody like him has acquired over years of research over just like being deeply immersed

• 2:52:53Speaker 01

  in deep learning, having tried lots of things himself and failed. The AIs, you know, they're going to be able to read every research paper I've written, every experiment I've ever run at the lab, you know, like gain the intuitions from all of this. They're going to be able to learn in parallel from all of each other's experiments, you know, experiences. You know, I don't know. What else? You know, it's like, what does it take to be an Alec Radford? Well, there's a, there's a sort of cultural elimination aspect of it, right? You know, if you hire somebody new, and there's like politicking, maybe they don't fit in. Well, in the AI case, you just make replicas, right? There's a like motivation aspect for it, right? So it's like, you know, Alec, you know, if I could just like duplicate Alec Radford, and before I run every experiment, I have him spend like, you know, a decade's worth of human time, like double checking the code and thinking really careful, carefully about it. I mean, first of all, I don't have that many Alec Radfords, and you know, he wouldn't care, and he would not be motivated.

• 2:53:38Speaker 01

  But, you know, the AIs, they can just be like, look, I have 100 million of you guys. I'm just going to put you on, just like really making sure this code is correct. There are no bugs. This experiment was thought through. Every hyperparameter is correct. Final thing I'll say is, you know, the 100 million human equivalent AI researchers, that is just a way to visualize it. So that doesn't mean you're going to have literally 100 million copies. You know, so there's trade-offs you can make between serial speed and in parallel. So you might make the trade-off is, look, we're going to run them at, you know, 10x, 100x serial speed. It's going to result in fewer tokens overall because of sort of inherent trade-offs. But, you know, then we have, I don't know what the numbers would be, but then we have, you know, 100,000 of them running at 100x human speed and thinking and, you know, and there's other things you can do on coordination. You know, they can kind of like share latent space, attend to each other's

• 2:54:21Speaker 01

  context. There's basically this huge range of possibilities of things you can do. The 100 million thing is more, I mean, another illustration of this is, you know, if you kind of I run the math in my series and it's basically, you know, twenty seven, twenty eight. You have this automated AI researcher. You're going to be able to generate an entire internet's worth of tokens every single day. So it's clearly sort of a huge amount of like intellectual work they can do.

• 2:54:43Dwarkesh Patel?

  I, I, I think the analogous thing there is today we generate more patents in a year than during the actual physics revolution and the early twentieth century. They were generating across like half a century or something. And are you making more physics progress in a year today than you? So yeah, you're going to generate all these tokens.

• 2:55:00Speaker 01

  Yeah.

• 2:55:01Dwarkesh Patel?

  Are you generating as much codified knowledge as humanity has been able to generate in the?

• 2:55:07Speaker 01

  No, no, probably not. Internet tokens are usually final output, right?

• 2:55:11Dwarkesh Patel?

  Right.

• 2:55:11Speaker 01

  A lot of these tokens, if we talk, we talked about the unhobbling, right?

• 2:55:14Dwarkesh Patel?

  Right.

• 2:55:14Speaker 01

  I think of a kind of like, you know, a GPT N token is sort of like one token of my internal monologue.

• 2:55:19Dwarkesh Patel?

  Yeah.

• 2:55:19Speaker 01

  Right. And so that's how I do this math on human equivalents. You know, it's like a hundred tokens a minute and then, you know, humans working for X hours and, you know, what is the, what is the equivalent there?

• 2:55:27Dwarkesh Patel?

  I, I, I think this goes back to something we were talking about earlier where, why haven't we seen the huge revenues from? People often ask this question that if you took GPT four back ten years and you show people this and they think this is going to automate. This is already automated half the jobs. And so there's a sort of a modus ponens, modus tollens here where part of the explanation is like, oh, it's like just on the verge. You need to do these unhobblings and part of that is probably true.

• 2:55:51Speaker 01

  Right.

• 2:55:52Dwarkesh Patel?

  But there is another lesson to learn there, which is that just looking at face value at a set of abilities.

• 2:55:58Speaker 01

  Yeah.

• 2:55:59Dwarkesh Patel?

  There's probably more sort of hobblings that you don't realize.

• 2:56:01Speaker 01

  Yes.

• 2:56:02Dwarkesh Patel?

  That are hidden behind the scenes. I think the same will be true of the AGIs that you have running as AI researchers. I think a lot of things.

• 2:56:08Speaker 01

  Basically agree, right? I think, I think my story here is like, you know, I, you know, I talk about, I think there's going to be some long tail, right? And so, you know, maybe it's like, you know, twenty six, twenty seven, you're like the proto automated engineer and it's really good at engineering. He doesn't have the research intuition, yet you don't quite know how to put them to work. But you know, the sort of even the underlying pace of AI progress is already so fast, right? In three years from not being able to do any kind of like math at all to now crushing, crushing these math competitions. And so you have the initial thing in like 26, 27, maybe the sort of it's an automated research engineer. It speeds you up by 2X. You go through a lot more progress in that year. By the end of the year, you've figured out like the remaining kind of unhobblings. You've like got a smarter model and you know, maybe than that thing, or maybe it's two years, you know, and that thing, just like that thing really can do automate 100%. And again, you know, they don't need to be doing everything.

• 2:56:52Speaker 01

  They don't need to be making coffee, you know, they don't need to like, you know, maybe there's a bunch of, you know, tacit knowledge and a bunch of other fields. But you know, AI researchers at AI labs really know the job of an AI researcher. And it's in some sense, it's a sort of, there's lots of clear metrics. It's all virtual, there's code, it's things you can kind of develop and train for.

• 2:57:10Dwarkesh Patel?

  So I mean, another thing is, how do you actually manage a million AI researchers? Humans, the sort of comparative ability we have, that we've been especially trained for is like working in teams.

• 2:57:22Speaker 01

  Yeah.

• 2:57:22Dwarkesh Patel?

  And despite this fact, we have, for thousands of years, we've been learning about how we work together in groups. And despite this, management is a clusterfuck, right? It's like, most companies are badly managed. It's, it's really hard to do this stuff.

• 2:57:36Speaker 01

  Yeah.

• 2:57:37Dwarkesh Patel?

  For AIs, the, the sort of like, we talk about AGI, but it'll be some bespoke set of abilities. Some of which will be higher than humans, some of which will be at human level.

• 2:57:49Speaker 01

  Yeah.

• 2:57:49Dwarkesh Patel?

  And so it'll be some bundle and we'll need to figure out how to put these bundles together with their human overseers, with the equipment and everything.

• 2:58:00Speaker 01

  Yeah.

• 2:58:01Dwarkesh Patel?

  And the idea that as soon as you get the bundle, you'll figure out how to get like, just shove millions of them together and manage them. I'm just very skeptical of, like, any other revolution, technological revolution in history has been very piecemeal. Much more piecemeal than you would expect on paper. If you just thought about what is the industrial revolution? Well, we dig up coal that powers a steam engines. We use the steam engines to run these railroads. That helps us get more coal out. And there's sort of like factorial story you can tell where in like a six, six hours, you can be pumping thousands of times more coal. But in real life, it takes centuries often, right? In fact, the electrification, there's this famous study about how.

• 2:58:44Dwarkesh Patel?

  To initially to electrify factories. It was decades before after electricity to change from the pulleys and water wheel base system that we had for steam engines to one that's works with more spread out electrical motors and everything. I think this will be the same kind of thing. It might take like decades to actually get millions of AI researchers to work together.

• 2:59:04Speaker 01

  Okay, great. This is great. Okay, so a few responses to that. First of all, I mean, I totally agree with the kind of like real world bottlenecks type of thing. I think this is sort of, you know, it's underrated. You know, basically what we're doing is we're removing the labor constraint. We automate labor and we like kind of exploit technology, but you know, there's still lots of other bottlenecks in the world. And so I think this is part of why the story is it kind of like starts pretty narrow at the thing where you don't have these bottlenecks. And then only over time, as we let it, it kind of expands to sort of broader areas. AI, this is part of why I think it's like initially this sort of AI research explosion, right? It's like AI research doesn't run into these real world bottlenecks. It doesn't require, you know, like plow a field or dig up coal. It's just, you're just doing AI research. The other thing, you know, the other thing about-

• 2:59:42Dwarkesh Patel?

  I love how like in your model, AI research, it's not complicated like about flipping a burger, it's just AI research.

• 2:59:49Speaker 01

  Well, no, but I mean, I mean, this is because people make these arguments of like, oh, you know, AGI won't do anything because it can't flip a burger. I'm like, yeah, it won't be able to flip a burger, but it's going to be able to do algorithmic progress, you know? And then, and then, and then when it does algorithmic progress, it'll figure out how to flip a burger, you know? And then we'll have a burger flipping, you know, robot. You know, look, the, sorry, the other thing is about, you know, again, these are the sort of quantities are lower bound, right? So it's like, this is just like, we can definitely run 100 million of these. Probably what will happen is one of the first things we're going to try to figure out is how to like, again, run like, you know, translate quantity into quality, right? And so it's like, even at the baseline rate of progress, you're like quickly getting smarter and smarter systems, right? If we said it was like, you know, four years between the preschooler and the high schooler, right?

• 3:00:32Speaker 01

  So I think, you know, pretty quickly, you know, there's probably some like simple algorithmic changes you find, you know, if instead of one ale, like Radford, you have 100, you know, you don't even need 100 million. And then, and then you get even smarter systems. And now these systems are, you know, they're capable of sort of creative, complicated behavior you don't understand. Maybe there is some way to like use all this test time compute in a more unified way rather than all these parallel copies. And you know, so there won't just be quantitatively superhuman, they'll pretty quickly become qualitatively superhuman. You know, it's sort of like, um, look like, you know, you're a high school student, you're like trying to wrap yourself, wrap your mind around kind of standard physics. And then there's some like super smart professor who is like, quantum physics, it all makes sense to him. And you're just like, what is going on? And sort of, I think pretty quickly you kind of enter that regime.

• 3:01:13Speaker 01

  Just given even the underlying pace of AI progress, but even more quickly than that, because you have this sort of accelerated force of now this automated AI research.

• 3:01:20Dwarkesh Patel?

  I, I agree that over time you would, I'm not denying that ASI is a thing that's possible.

• 3:01:27Speaker 01

  You know?

• 3:01:27Dwarkesh Patel?

  I'm just like, you know, how, how is this happening in a year? Like you, okay, first of all.

• 3:01:31Speaker 01

  So I think, I think the story is sort of like, basically, I think it's a little bit more continuous, you know, right? Like, I think already, you know, like I talked about, you know, 25, 26, you're basically going to have models as good as a college graduate. And I, you know, I don't, I don't know how, where the unhobbling is going to be, but I think it's plausible that even then you have kind of the proto automated engineer. So there's, I think there is a bit of like a smear, kind of an AGI smear or whatever, where it's like, there's sort of unhobblings that you're missing. There's kind of like ways of connecting them you're missing. There's like some level of intelligence you're missing. But then at some point you are going to get the thing that is like a hundred percent automated Alec Bradford. And once you have that, you know, things really take off, I think.

• 3:02:05Dwarkesh Patel?

  Yeah. Okay. So let's go back to the unhobblings.

• 3:02:08Speaker 01

  Yeah.

• 3:02:08Dwarkesh Patel?

  Is there, we're going to get a bunch of models by the end of the year. Is there something, let's suppose we didn't get some capacity by the end of the year.

• 3:02:15Speaker 01

  Yeah.

• 3:02:16Dwarkesh Patel?

  Is there some such capacity which lacking would suggest that AI progress is going to take longer than you are projecting?

• 3:02:22Speaker 01

  Yeah. I mean, I think there's, there's two kind of key things. There's the unhobbling and there's the data wall. Right? I think we should talk about the data wall for a moment. I think the data wall is, um, you know, even though kind of like all this stuff has been about, you know, crazy AI progress. I think the data wall is actually sort of underrated. I think there's like a real scenario where we're just stagnating.

• 3:02:36Dwarkesh Patel?

  Yeah.

• 3:02:37Speaker 01

  You know, because we've been running this tailwind of just like, it's really easy to bootstrap and you just do unsupervised learning, next token prediction. It learns these amazing world models, like bam, you know, great model. And you just got to buy some more compute, you know, do some simple efficiency changes. You know, and, and again, like so much of deep learning, all these like big gains on efficiency have been like pretty dumb things, right? Like, you know, you add a normalization layer, you know, um, you know, you fix the scaling laws, you know, and these already have been huge things, let alone kind of like obvious ways in which these models aren't good yet. Anyway, so data wall, big deal. Um, you know, I don't know, some like. Put some numbers on this, you know, some like you do common crawl, you know, online is like, you know, 30 trillion tokens. Lama 3, you trained on 15 trillion tokens. So you're basically already using all the data.

• 3:03:18Dwarkesh Patel?

  Yeah.

• 3:03:19Speaker 01

  And then, you know, you can get somewhat further by repeating it. So there's an academic paper by, you know, Boaz Barak and some others that does scaling laws for this. And they're basically like, yeah, you can repeat it sometime. After 16 times of repetition, it's just like returns basically go to zero. You're just completely screwed. And so, I don't know, say you can get another 10x on data from, say like Lama 3 and GPT4, you know, Lama 3 is already kind of like at the limit of all the data. You know, maybe you can get 10x more by repeating data. You know, I don't know, maybe that's like at most 100x better model than GPT4, which is like, you know, 100x effective compute from GPT4 is, you know, not that much. You know, if you do half an order of magnitude a year of compute, half an order of magnitude a year of algorithmic progress, you know, that's kind of like two years from GPT4. So,

• 3:03:59Speaker 01

  you know, GPT4 finished pre-training in 22, you know, 24. So I think one thing that really matters, I think we won't quite know by the end of the year, but you know, 25, 26, are we cracking the data wall?

• 3:04:13Dwarkesh Patel?

  Okay, so suppose we had three orders of magnitude less data in common crawl on the internet than we just happen to have now. And for decades,

• 3:04:23Speaker 01

  Yeah.

• 3:04:23Dwarkesh Patel?

  the internet, other things, we've been rapidly increasing the stock of data that humanity has.

• 3:04:28Speaker 01

  Yeah.

• 3:04:29Dwarkesh Patel?

  Is it your view that for contingent reasons, we just happen to have enough data

• 3:04:34Speaker 01

  Yeah.

• 3:04:34Dwarkesh Patel?

  to train models that are just powerful enough

• 3:04:37Speaker 01

  Yeah.

• 3:04:37Dwarkesh Patel?

  at 4.5 level where they can kick off the self-play RL loop?

• 3:04:42Speaker 01

  Yeah.

• 3:04:43Dwarkesh Patel?

  Or is it just that we, you know, if it had been three rooms higher,

• 3:04:48Speaker 01

  Yeah.

• 3:04:48Dwarkesh Patel?

  then it would progress would have been slightly faster.

• 3:04:50Speaker 01

  Yeah.

• 3:04:51Dwarkesh Patel?

  In that world, we would have been looking back at like, oh, how hard it would have been to like, kick off the RL explosion with just 4.5, but we would have figured it out. And then, so in this world, we would have gone to GPT3 and then we'd have to kick off and start up RL explosion.

• 3:05:02Speaker 01

  Yeah.

• 3:05:03Dwarkesh Patel?

  But we would have, we would have still figured it out. The sort of, we didn't just like luck out on the amount of data we happen to have in the world.

• 3:05:08Speaker 01

  I mean, three rooms is pretty rough, right? Like three rooms, if less data means like six rooms smaller, six rooms, like less compute model and Chillo scaling laws, you know, that's, it's basically capping out at like GPT2, barely. But so I think that would be really rough. I think you do make an interesting point about the contingency. You know, I guess earlier we were talking about the sort of like when in the sort of human trajectory are you able to learn from yourself? And so, you know, if you go with that analogy, again, like if you'd only gotten the preschooler model, it can't learn from itself. You know, if you'd only gotten the elementary schooler model, can't learn from itself. And, you know, maybe GPT-4, you know, smart high school is really where it starts. Ideally, you have a somewhat better model, but then it really is able to kind of like learn from itself or learn by itself. So I, yeah, I think there's an interesting, I think, I mean, I think maybe one oomles data, I would be like more iffy, but maybe still doable.

• 3:05:52Speaker 01

  Yeah, I think it would feel chiller if we had, you know, like one or two.

• 3:05:56Dwarkesh Patel?

  It would be an interesting exercise to get probability distributions of AGI contingent done across like ooms of data.

• 3:06:01Speaker 01

  Yeah.

• 3:06:02Dwarkesh Patel?

  Okay. The thing that makes me skeptical of this story is that the things, it totally makes sense why pre-training works so well. These other things, they're stories of in principle, why they ought to work, like humans can learn this way and so on. And maybe they're true, but I worry that a lot of this case is based on sort of first principles evaluation of how learning happens that fundamentally, we don't understand how humans learn. And maybe there's some key thing we're missing.

• 3:06:32Speaker 01

  Yeah.

• 3:06:33Dwarkesh Patel?

  On the sort of sample efficiency, yeah, humans actually, maybe there's, you say, well, the fact that these things are way of a less sample efficient in terms of learning than humans are suggested there is a lot of room for improvement. Another perspective is that we are just on the wrong path altogether, right? That's why there's a sample inefficient when it comes to pre-training. So, yeah, I'm just like, there's a lot of like first principles arguments stacked on top of each other where you get these un-hobblings and then you get AGI. Then you, because of these reasons, why you can stack all these things on top of each other, you get to ASI. And I'm worried that there's too many steps of this, sort of first principles thinking.

• 3:07:09Speaker 01

  I mean, we'll see, right? I mean, on the, on the, on the sort of sample efficiency thing, again, sort of first principles, but I think, again, there's this clear sort of missing middle. And so, you know, and sort of like, you know, people hadn't been trying, now people are really trying, you know, and so it's sort of, you know, I think often again in deep learning, something like the obvious thing works and there's a lot of details to get right. So it might take some time, but it's now where people are really trying. So I think we get a lot of signal in the next couple of years. You know,

• 3:07:42Speaker 01

  On hobbling, I mean, what is the signal on hobbling that I think would be interesting? I think, I think the question is basically like, are you making progress on this test time compute thing? Right? Like, is this thing able to think longer horizon than just a couple hundred tokens? Right? That was unlocked by Chain of Thought.

• 3:07:54Dwarkesh Patel?

  And on that point in particular, the many people who have longer timelines have come on the podcast. Have made the point that the way to train this long horizon RL, it's not, I mean, I earlier talking about like, well, they can think for five minutes, but not for longer. Yeah. But it's not because they can't physically output in hours with their tokens. It's just really, at least from what I understand.

• 3:08:16Speaker 01

  They're not clear, right? Like even like Gemini has like a million contacts and the million of contacts is actually great for consumption. And it solves one important on hobbling, which is the sort of onboarding problem. Right? Which is, you know, a new coworker, you know, in your first five minutes, like a new smart high school intern, first five minutes, not useful at all. A month in, you know, much more useful, right? Because they've like, looked at the monorepo and understand how the code works. And they've looked at your internal docs. And so being able to put that in context, great, solves this onboarding problem. Yeah, but they're not good at sort of the production of a million tokens yet.

• 3:08:45Dwarkesh Patel?

  Yeah, right. But on the production of a million tokens,

• 3:08:49Dwarkesh Patel?

  there's no public evidence that there's some easy loss function where you can.

• 3:08:54Speaker 01

  GPT-4 has gotten a lot better since, it's actually, so the GPT-4 gains since launch, I think are a huge indicator that there's like, you know, so you talked about this with John on the podcast, John said this was mostly post-training gains. You know, if you look at the sort of LMCS scores, you know, it's like 100 elo or something. It's like a bigger gap than between Claude 3 Opus and Claude 3 Haiku. And the price difference between those is 60x.

• 3:09:16Dwarkesh Patel?

  But it's not more agentic. It's like better in the same chatbot way.

• 3:09:20Speaker 01

  Right? Like, you know, went from like, you know, 40% to 70%.

• 3:09:23Dwarkesh Patel?

  The crux is like whether like be able to like.

• 3:09:25Speaker 01

  But I think, I think it indicates that clearly there's stuff to be done on hobbling. I think, yeah, I think, I think the interesting question is like this time of year from now, you know, is there a model that is able to think for like, you know, a few thousand tokens, coherently, cohesively, agentically. And I think probably there's, you know, again, this is, I'd feel better if we had an um or two more data. Cause it's like, the scaling just gives you this sort of like tailwind, right? Where like, for example, tools, right? Tools, I think, you know, talking to people who try to. Make things work with tools, you know, actually sort of G P four is really when tools start to work. And it's like, you can kind of make them work with G P three point five, but it's just really tough. And so it's just like having G P four, you can kind of help it learn tools in a much easier way. And so just a bit more tailwind from scaling.

• 3:10:07Speaker 01

  And then, and then, yeah, and does, does, I don't know if it'll work, but it's a key question.

• 3:10:12Dwarkesh Patel?

  Okay, I, I think it's a good place to sort of close that part where we know what the crux is and what the progress, what, what evidence that would look like. On the A G I to super intelligence.

• 3:10:24Dwarkesh Patel?

  Maybe it's a case that the games are really easy right now and you can just sort of let loose an Alec Ratford, give him a compute budget and he comes out the other end with something that is an additive. Like change this part of the code, this is compute multiplier, change this other part. What other parts of the world?

• 3:10:43Dwarkesh Patel?

  Okay, maybe here's an interesting way to ask this. How many other domains in the world are like this, where you think you could get the equivalent of in one year, you just throw enough intelligence across multiple instances. And you would just come out the other end with something that is remarkably decades, centuries ahead.

• 3:11:03Speaker 01

  Yeah.

• 3:11:04Dwarkesh Patel?

  Like you start off with no flight and then you have the right brothers, a million instances of G P D six and you come out the other end with Starlink.

• 3:11:13Speaker 01

  Yeah.

• 3:11:15Dwarkesh Patel?

  Like, is that your model of how things work?

• 3:11:17Speaker 01

  I think, I think you're exaggerating the timelines a little bit, but, but, you know, I think, you know, a decade's worth of progress in a year or something. I think that's a reasonable prompt. So I think this is where, you know, basically the sort of automated A I researcher comes in because it gives you this enormous headwind on all the other stuff, right? So it's like, you know, you automate A I research with your sort of automated Alec Radfords. You come out the other end, you've done another five booms, you have a thing that is like vastly smarter. Not only is it vastly smarter, you like, you know, you've been able to make it good at everything else, right? You're like, you're solving robotics, the robots are important, right? Because like for a lot of other things, you do actually need to like try things in the physical world. I mean, I don't know, maybe you can do a lot in simulation, those are the really quick worlds. I don't know if you saw the like last Nvidia G T C and it was all about the like digital twins and just like having all your manufacturing processes in simulation.

• 3:12:02Speaker 01

  I don't know, like, again, if you have these like, you know, super intelligent, like cognitive workers, like, can they just like make simulations of everything, you know, kind of off of code style? And then, and then, you know, make a lot of progress in simulation possible. But I also just think you're going to get the robots.

• 3:12:17Speaker 01

  Again, I agree about like, there are a lot of real world bottlenecks, right? And so, you know, I don't know, it's quite possible that we're going to have, you know, crazy drone storms, but also, you know, like lawyers and doctors still need to be humans because of like, you know, regulation. But, you know, I think, you know, you kind of start narrowly, you broaden, and then the worlds in which you kind of let them loose, which again, because of, I think, these competitive pressures, we will have to let them loose in some degree on, you know, various national security applications. I think like, quite rapid progress is possible. The other thing, though, is it's sort of, you know, basically in this sort of explosion after, there's kind of two components. There's the A, right, in the production function, like growth of technology. And that's massively accelerated by, you know, you have a billion super intelligent scientists and engineers and technicians, you're superbly competent at everything.

• 3:13:01Speaker 01

  You also just automated labor, right? And so it's like, even without the whole technological explosion thing, you have this industrial explosion, at least if you let them, let them loose, which is like, now you can just build, you know, you can cover Nevada and like, you know, you start with one robot factory that's producing more robots. And basically this like, just cumulative process, because you've taken labor out of the equation.

• 3:13:21Dwarkesh Patel?

  Yeah, that's super interesting.

• 3:13:23Speaker 01

  Yeah.

• 3:13:23Dwarkesh Patel?

  Although, when you increase the K or the L without increasing the A, you can look at the Soviet Union or China, where they rapidly increased inputs.

• 3:13:35Speaker 01

  Yeah.

• 3:13:35Dwarkesh Patel?

  And that does have the effect of being geopolitically game changing, where you, it is remarkable, like you go to Shanghai over a six, course of decades.

• 3:13:43Speaker 01

  These crazy cities in a decade, right?

• 3:13:45Dwarkesh Patel?

  Right, right. I mean, the closest thing to like, people talk about 30% growth rate or whatever.

• 3:13:48Speaker 01

  Yeah, Asian tigers, 10%. It's totally possible, right?

• 3:13:51Dwarkesh Patel?

  But, but without productivity gains, it's not like the industrial revolution, where like, you're, from the perspective of, you're looking at a system from the outside, your goods have gotten cheaper, they can manufacture more things. But, you know, it's, it's not like the next century is coming at you.

• 3:14:07Speaker 01

  Yeah, it's both, it's both. So, you know, both are important. The other thing I'll say is like, all this stuff, I think the magnitudes are really, really important, right? You know, we talked about a 10x of research effort, or maybe 10, 30x over a decade. You know, even without any kind of like self-improvement type loop, you know, we talk the sort of even in the sort of GPT4 to AGI story, we're talking about an order of magnitude of effective compute increase a year, right? Half an order of magnitude of compute, half an order of magnitude of algorithmic progress. That sort of translates into effective compute. And so you're doing a 10x a year, right? Basically on your labor force, right? So it's like, it's a radically different world if you're doing a 10x or 30x in a century versus a 10x a year on your labor force. So the magnitudes really matter. They also really matter on the sort of intelligence explosion, right?

• 3:14:48Speaker 01

  So like just the automated AI research part. So you know, one story you could tell there is like, well, ideas get harder to find, right? Algorithmic progress is going to get harder. Yeah, right now you have the easy wins, but in like four or five years, there'll be fewer easy wins. And so the sort of automated AI researchers are just going to be what's necessary to just keep it going, right? Because it's gotten harder. But that's sort of, it's like a really weird knife-edge assumption economics where you assume it's just enough.

• 3:15:09Dwarkesh Patel?

  But isn't that the equilibrium story you were just telling with why the economy as a whole has 2% economic growth? Because you just proceeded on the equal, I guess you were saying by the time it gets to the equilibrium.

• 3:15:17Speaker 01

  The result of the equilibrium here is it's like way faster, at least, you know, and it's at least, and it depends on the sort of exponents, but it's basically, it's the increase. Like suppose you need to like 10x effective research effort in AI research in the last, you know, four or five years to keep the pace of progress. We're not just getting a 10x, you're getting, you know, a million x or 100,000 x. There's just magnitudes really matter. And the magnitude is just basically, you know, once, one way to think about this is that you have kind of two exponentials. You have your sort of like normal economy that's growing at, you know, 2% a year. And you have your like AI economy and that's going at like 10x a year. And it's starting out really small, but sort of eventually it's gonna, it's just, it's, it's, it's, it's way faster and eventually it's going to overtake, right? And even if you have, you can almost sort of just do the simple revenue extrapolation, right?

• 3:15:57Speaker 01

  If you think your AI economy, you know, that has some growth rate, I mean, it's a very simplistic way and so on, but there's, there's this sort of 10x a year process. And that will eventually kind of like, you're going to transition the sort of whole economy from, as it broadens, from the sort of, you know, 2% a year to the sort of much faster growing process. And I don't know, I think that's very like, um, consistent with historical change, you know, stories of right? There's this sort of like. You know, there's a sort of long run hyperbolic trend, you know, manifested in the sort of like sort of change and growth mode in the Austral, you know, revolution, but there's just this long run hyperbolic trend. And, um, you know, now you have this sort of, now you have that another sort of change in growth mode.

• 3:16:35Dwarkesh Patel?

  Yeah. Yeah. I mean, that was one of the questions I asked Tyler.

• 3:16:37Speaker 01

  Yeah.

• 3:16:38Dwarkesh Patel?

  When I had him on the podcast.

• 3:16:39Speaker 01

  Yeah, nice.

• 3:16:39Dwarkesh Patel?

  Is that you do go from, the fact that after 1776, you go from a regime of negligible economic growth to percent.

• 3:16:47Speaker 01

  Yeah.

• 3:16:47Dwarkesh Patel?

  Is really interesting. It shows that, I mean, from the perspective of somebody in the Middle Ages or before.

• 3:16:53Speaker 01

  Yeah.

• 3:16:53Dwarkesh Patel?

  Two percent is equivalent to the sort of 10%.

• 3:16:55Speaker 01

  Yeah.

• 3:16:56Dwarkesh Patel?

  I guess you're projecting even higher for the AI economy, but.

• 3:16:59Speaker 01

  Yeah. I mean, it depends. I think again, and it's all this stuff, you know, I have a lot of uncertainty, right? So a lot of the time I'm trying to kind of tell the modal story. I think it's important to be kind of concrete and visceral about it. And I, you know, I have, I have a lot of uncertainty basically over how the 2030s play out. And basically the thing I know is it's going to be fucking crazy. Um, but, but you know, exactly what, you know, where the bottlenecks are and so on. I think that'll be kind of like.

• 3:17:20Dwarkesh Patel?

  So, uh, let's talk through the numbers here. You hundreds of millions of AI researchers. So right now, G P D four O turbo is like 15 bucks for a million tokens outputted. And a human thinks 150 tokens a minute or something. And if you do the math on that, I think it's for an hour's worth of human output. You, it's like 10 cents or something. Now.

• 3:17:48Speaker 01

  Cheaper than a human worker.

• 3:17:49Dwarkesh Patel?

  Huh?

• 3:17:49Speaker 01

  Cheaper than a human worker.

• 3:17:50Dwarkesh Patel?

  Oh, yeah.

• 3:17:50Speaker 01

  But it can't do the job yet.

• 3:17:51Dwarkesh Patel?

  That's right. That's right. But by the time you're talking about models that are trained on the 10 gigawatt cluster.

• 3:17:56Speaker 01

  Yeah.

• 3:17:56Dwarkesh Patel?

  Then you have something that is four orders of magnitude more expensive via inference, three orders of magnitude, something like that. So that's like a hundred dollars an hour of labor. And now you're having hundreds of millions of such laborers. Is, is there enough compute to do with the model that is a thousand times bigger, this kind of labor?

• 3:18:17Speaker 01

  Great. Okay. Great question. So I actually don't think inference costs for sort of frontier models are necessarily going to go up that much. So, I mean, one historical data point is.

• 3:18:24Dwarkesh Patel?

  But isn't the test time sort of thing that it will go up even higher?

• 3:18:27Speaker 01

  I mean, we're just doing per token, right?

• 3:18:29Dwarkesh Patel?

  Yeah.

• 3:18:29Speaker 01

  And then I'm just saying, you know, if suppose each. Model token was the same as sort of a human token thing at 100 tokens a minute. So it's like, yeah, it'll use more, but the sort of if you just the token calculations is already pricing that in. The the question is like per token pricing, right? And so like GPT three, when it launched was like actually more expensive than GPT four now. And so over just like, you know, fast increases in capability gains inference costs has remained constant.

• 3:18:55Speaker 01

  That's sort of wild. I think it's worth appreciating, and I think it gestures at sort of an underlying pace of algorithmic progress.

• 3:19:02Speaker 01

  I think there's a sort of like more theoretically grounded way to why why inference costs would stay constant. And it's the following story, right? So on Chinchilla scaling laws, right? You, you know, half of the additional compute you allocate to bigger models and half of it you allocate to more data. Right? But also, if we go with the sort of basic story of half an order of a year more compute and half an order of magnitude a year of algorithmic progress. You're also kind of like, you're saving half an order of magnitude a year. And so that kind of would exactly compensate for making the model bigger. The caveat on that is, you know, obviously not all training efficiencies are also inference efficiencies. You know, a bunch of the time they are separately, you can find inference efficiencies. So I don't know, given this historical trend, given the sort of like, you know, baseline sort of theoretical reason.

• 3:19:43Speaker 01

  You know, I don't know, I, I think it's not crazy baseline assumption that actually these models, frontier models are not necessarily going to get more expensive per token.

• 3:19:51Dwarkesh Patel?

  Oh, really?

• 3:19:52Speaker 01

  Yeah.

• 3:19:52Dwarkesh Patel?

  Like, okay, that's, that's wild.

• 3:19:55Speaker 01

  We'll see, we'll see. I mean, the other thing, you know, maybe they get, you know, even if they get like 10x more expensive than, you know, you have 10 million instead of 100 million, you know? So it's like, it's not really, you know, like.

• 3:20:04Dwarkesh Patel?

  But, okay, so part of the intellectual explosion is that each of them has to run experiments. That are.

• 3:20:09Speaker 01

  Yeah, yeah, yeah.

• 3:20:10Dwarkesh Patel?

  GPT four sized. And the result, so that takes up a bunch of compute.

• 3:20:14Speaker 01

  Yes.

• 3:20:15Dwarkesh Patel?

  Then you need to consolidate the results of experiments and what is the synthesized.

• 3:20:20Speaker 01

  I mean, you have a much bigger influence street anyway than your training.

• 3:20:22Dwarkesh Patel?

  Sure, okay.

• 3:20:23Speaker 01

  But I think the experiment compute is a constraint.

• 3:20:25Dwarkesh Patel?

  Yeah, okay, I'm going back to maybe a sort of bigger fundamental thing we're talking about here. We're projecting, in a series you say we should denominate the probability of getting to AGI in terms of orders of magnitude of effective compute. Effective here, accounting for the fact that there's a quote unquote, compute multiplier if you have better algorithms.

• 3:20:49Speaker 01

  Yes.

• 3:20:50Dwarkesh Patel?

  And I'm not sure that it makes sense to be confident that this is a sensible way to project progress. It might be, but I'm just like, I have a lot of uncertainty about it. It seems similar to somebody trying to project when we're going to get to the moon. And they're like, looking at the Apollo program in the 50s or something, and they're like, we have some amount of effective jet fuel. And if we get more efficient engines, then we have more effective jet fuel. And so we're going to like probability of getting to the moon based on the amount of effective jet fuel we have. And I don't deny that jet fuel is important to launch rockets. But that seems like an odd way to denominate when you're going to get to the moon.

• 3:21:32Speaker 01

  Yeah. Yeah. So I mean, I think these cases are pretty different. I don't know. I didn't, I don't think there is a sort of clear, I don't know how rocket science works, but I didn't, I didn't get the impression that there's some clear scaling behavior with like, you know, the amount of jet fuel. I think the, I think in AI, you know, I mean, first of all, the scaling laws, you know, they've just helped, right? And so if you, a friend of mine pointed this out, and I think it's a great point. If you kind of concatenate both these sort of original Kaplan scaling laws paper that I think went from 10 to the negative nine to 10 petaflop days. And then, you know, concatenate additional compute to, from there to kind of GPT-4, you assume some algorithmic progress. You know, it's like the scaling laws have held, you know, like probably over 15 rooms, you know, I know it's rough calculation, probably maybe even more. Held for a lot of rooms.

• 3:22:14Dwarkesh Patel?

  They've held for the specific loss function which they're trained on, which is training next token. Whereas the progress you are forecasting, which will require for further progress

• 3:22:24Speaker 01

  Yes.

• 3:22:24Dwarkesh Patel?

  in capabilities.

• 3:22:26Speaker 01

  Yeah.

• 3:22:26Dwarkesh Patel?

  It was specifically, we know that scaling can't work because of the data wall. And so there's some new thing that has to happen. And I'm not sure whether the, you can extrapolate that same scaling curve to tell us whether these hobblings will also, like, is this not on the same graph?

• 3:22:39Speaker 01

  The hobblings are just a separate thing. So this is this is sort of like, you know, it's yeah. So I mean, a few few things here, right? Okay, so the on the on the effective compute scaling, the you know, in some sense, I think it's like people center the scaling laws because they're easy to explain and the sort of like, why, why is scaling matter? The scaling laws like came way after people, or at least, you know, like Dario, Ilya realized that scaling mattered. And I think, you know, I think the almost more important than the sort of loss curve is just like, just in general, make, you know, there's this great quote from Dario on your on your on your on your podcast. It's just like, you know, Ilya was like, the models, they just want to learn, you know, you make them bigger, they learn more. And and that just applied just across domains, generally, you know, all the capabilities. And so and you can look at this in benchmarks. Again, like you say, headwind data wall.

• 3:23:23Speaker 01

  And I'm sort of bracketing that and talking about that separately. The other thing is on hobblings, right? If you just put them on the effective compute graph, these on hobblings would be kind of huge, right? So like, I think But what does that even mean? Like, what is it? What is on the Y axis here? Like, say, MLPR on this benchmark or whatever, right? And so, you know, like, you know, we mentioned the sort of, you know, the LMCS differences, you know, RHF, you know, again, as good as 100x chain of thought, right? Chain of just going from this prompting chain, a simple algorithmic chain can be like 10x effective compute increases on like math benchmarks. I think this is like, you know, I think this is useful to illustrate that on hobblings are large. But I think they're like, I kind of think of them as like slightly separate things. And the kind of the way I think about is that, like, at a per token level, I think GP4 is not that far away from like a token of my internal monologue.

• 3:24:07Speaker 01

  Right? Even like 3.5 to 4 took us kind of from like the bottom of the human range to the top of the human range. Unlike a lot of, you know, on a lot of, you know, kind of like high school tests. And so it's like a few more 3.5 to 4 jumps per token basis, like per token intelligence. And then you've got to unlock the test time, you've got to solve the onboarding problem, make it use a computer. And then you're getting real close. I'm reminded of And again, the story might be wrong, but I think it is strikingly plausible. I agree. And so I'm just I think actually, I mean, the other thing I'll say is like, you know, I say this 2027 timeline. I think it's unlikely, but I do think there's worlds that are like AGI next year. And that's basically if the test and compute overhang is really easy to crack. If it's really easy to crack, then you do like four rooms of test and compute, you know, from a few hundred tokens to a few million tokens, you know, quickly.

• 3:24:52Speaker 01

  And then, you know, again, maybe it's maybe it only takes one or two, 3.5 to four jumps per token, like one or two of those jumps per token, plus uses test and compute. And you basically have the proto automated engineer.

• 3:25:03Dwarkesh Patel?

  So I'm reminded of Stephen Pinker releases his book on, what is it? The Better Angels of Our Nature. And it's like a couple of years ago or something. And he says, the secular decline in violence and war and everything. And you can just like plot the line from the end of World War II. In fact, before World War II, then these are just aberrations, whatever. And basically, as soon as it happens, Ukraine, Gaza, the everything is like.

• 3:25:30Speaker 01

  So impending ASI and crazy global conflict. Right, right. ASI and crazy new W and D.

• 3:25:39Dwarkesh Patel?

  I think this is a sort of thing that happens in history where you see a straight line and you're like, oh, my gosh. And then just like, as soon as you make that prediction.

• 3:25:45Speaker 01

  Yeah.

• 3:25:46Dwarkesh Patel?

  Who was that famous author?

• 3:25:47Speaker 01

  So, yeah, this, you know, again, people are predicting deep learning will hit a wall every year. Maybe one year they're right, but it's like gone a long way and it hasn't hit a wall. And we don't have that much more to go. And, and, you know, so, yeah.

• 3:25:58Dwarkesh Patel?

  I guess I think this is a sort of plausible story and let's just run with it.

• 3:26:01Speaker 01

  Yeah.

• 3:26:02Dwarkesh Patel?

  And see what it implies.

• 3:26:03Speaker 01

  Yeah.

• 3:26:04Dwarkesh Patel?

  So we were talking, in your series, you talk about alignment from

• 3:26:08Speaker 01

  Yeah.

• 3:26:08Dwarkesh Patel?

  the perspective of this is not about some doomer scheme to get the point zero one percent probability distribution where things don't go off the rails. It's more about just controlling the systems, making sure they do what we intend them to do. If that's the case, and we're going to be in this sort of geopolitical conflict with China. And part of that will involve, and what we're worried about is them making these CCP bots that go out and take the red flag of Mao across the galaxies or something. Then. Shouldn't we be worried about alignment as something that if in the wrong hands,

• 3:26:48Dwarkesh Patel?

  this is the thing that enables brainwashing, sort of dictatorial control? This is, this seems like a worrying thing. This should be part of the sort of algorithmic secrets we keep hidden, right? How to align these models, because that's also something the CCP can use to control their models.

• 3:27:03Speaker 01

  I mean, I think in the world where you get the democratic coalition, yeah. I mean, also just alignment is often dual use, right? Like, RHF, you know, it's like, alignment team developed, it was great, you know, it was a big win for alignment, but it's also, you know, obviously makes these models useful.

• 3:27:14Dwarkesh Patel?

  Right.

• 3:27:15Speaker 01

  The, um, but yeah, so yeah, alignment enables the CCP bots. Alignment also is what you need to get the, you know, get the sort of, you know, whatever, USAI, so like follow the constitution and like disobey, you know, unlawful orders and, you know, like respect separation of powers and checks and balances. And so, yeah, you need alignment for whatever you want to do. It's just, it's the sort of underlying technique.

• 3:27:37Dwarkesh Patel?

  Tell me what you make of this take. I've been starting with this a little bit.

• 3:27:40Speaker 01

  Okay.

• 3:27:40Dwarkesh Patel?

  So fundamentally, there's many different ways the future could go.

• 3:27:44Speaker 01

  Yeah.

• 3:27:45Dwarkesh Patel?

  There's one path in which the Eliezer-type crazy AIs with the nanobots take the future and to turn everything into gray goo or paperclips. And the more you solve alignment, the more that path of the decision tree is circumscribed. And then, so the more you solve alignment, the more it is just different humans and the visions they have. And of course, we know from history that things don't turn out the way you expect. So it's not like you can decide the future, but it will appear.

• 3:28:10Speaker 01

  It's part of the beauty of it, right?

• 3:28:11Dwarkesh Patel?

  Yeah.

• 3:28:11Speaker 01

  You want these mechanisms, the error correction, moralism.

• 3:28:13Dwarkesh Patel?

  Exactly. But from the perspective of anybody who's looking at the system, it'll be like, I can control where this thing is going to end up. And so the more you solve alignment and the more you circumscribe the different futures that are the result of AI will, the more that accentuates the conflict between humans

• 3:28:32Speaker 01

  Yes.

• 3:28:33Dwarkesh Patel?

  and their visions of the future.

• 3:28:34Speaker 01

  Yep.

• 3:28:34Dwarkesh Patel?

  And so in the world where alignment is solved, and the world in which alignment is solved, is the one, is the world in which you have the most sort of human conflict over where to take AI.

• 3:28:42Speaker 01

  Yeah, I mean, it's, you know, by removing the worlds in which the AIs take over, then like, you know, the remaining worlds are the ones where it's like, the humans decide what happens. And then, as we talked about, there's a whole lot of, you know, a whole lot of worlds and how that could go.

• 3:28:53Dwarkesh Patel?

  And I worry, so when you think about alignment, and it's just controlling these things.

• 3:28:57Speaker 01

  Yeah.

• 3:28:59Dwarkesh Patel?

  Just think a little forward and there's worlds in which hopefully, you know, human descendants or some version of things in the future merge with super intelligences and they have the rules of their own, but they're in some sort of law and market-based order.

• 3:29:14Speaker 01

  Uh huh.

• 3:29:15Dwarkesh Patel?

  I worry about if you have things that are conscious and should be treated with rights.

• 3:29:21Speaker 01

  Uh huh.

• 3:29:22Dwarkesh Patel?

  If you read about what alignment schemes actually are and then you read these books about what actually happened during the Cultural Revolution.

• 3:29:28Speaker 01

  Uh huh.

• 3:29:28Dwarkesh Patel?

  What happened when Stalin took over Russia? And you have very strong monitoring from different instances where one, everybody's tasked with watching each other. You have brainwashing, you have red teaming, where you have the spy stuff you were talking about, where you try to convince somebody you're on like a defector and you see if they defect with you.

• 3:29:48Speaker 01

  Yeah.

• 3:29:48Dwarkesh Patel?

  And if they do, then you realize they're an enemy and then you. And listen, maybe I'm stretching the analogy too far.

• 3:29:54Speaker 01

  Yeah.

• 3:29:55Dwarkesh Patel?

  But the, the way, like, the ease of these alignment techniques actually map on to something you could have read about during like Mao's Cultural Revolution is a little bit troubling.

• 3:30:06Speaker 01

  Yeah, I mean, look, I think sentient AI is a whole nother topic. I don't know if we want to talk about it. I agree that like, it's going to be very important how we treat them. You know, in terms of like, what you're actually programming these systems to do. Again, it's like alignment is just, it's a technical, it's a technical problem, a technical solution, enables the CCP bots. I mean, in some sense, I think the, um, you know, I almost feel like the sort of model and also about talking about checks and balances is sort of, you know, like the Federal Reserve or Supreme Court justices. And there's a funny way in which they're kind of this like very dedicated order, you know, Supreme Court justices. And it's amazing, they're actually quite high quality.

• 3:30:37Dwarkesh Patel?

  Yeah.

• 3:30:38Speaker 01

  Right? And they're like, really smart people, they really believe in the Constitution, they love the Constitution, they believe in their principles, they have, you know, these, these, these wonderful, um, you know, and yeah, they have different persuasions, but they have sort of, I think, very sincere kind of debates about what is the meaning of the Constitution, you know, what is the best actuation of these principles? You know, I guess, by the way, our recommendations, sort of SCOTUS or arguments is like the best podcasts, you know, but when I run out of high-quality content on the internet. I mean, I think there's going to be a process of like figuring out what the Constitution should be. I think, you know, this Constitution has like worked for a long time, you start with that, maybe eventually things change enough that you want edits to that. But anyway, you want them to like, you know, for example, for the checks and balances, they like, they really love the Constitution and they believe in it and they take it really seriously.

• 3:31:20Speaker 01

  And like, look, at some point, yeah, you are going to have like AI police and AI military, but I think sort of like. You know, being able to ensure that they like, you know, believe in it in the way that like a Supreme Court Justice does, or like in the way that like a Federal Reserve, you know, official takes their job really seriously.

• 3:31:35Dwarkesh Patel?

  Yeah. And I guess the big open question is whether if you do the project or something like the project.

• 3:31:40Speaker 01

  And sorry, the other important thing is like a bunch of different factions need their own AIs, right? And so it's really important that like each political party gets to like have their, you know, and like whatever crazy, you might totally disagree with their values, but it's like, it's really important that they get to like have their own kind of like superintelligence. And, and again, I think it's that these sort of like classical liberal processes play out, including like different people of different persuasions and so on. And I don't know, maybe the AI advisors might not make them, you know, wise. They might not follow the advice or whatever, but I think it's important.

• 3:32:07Dwarkesh Patel?

  Okay. So speaking of alignment, you seem pretty optimistic. So let's run through the source of the optimism.

• 3:32:14Speaker 01

  Yeah.

• 3:32:14Dwarkesh Patel?

  I think there you laid out different worlds in which we could get AI.

• 3:32:18Speaker 01

  Yeah.

• 3:32:19Dwarkesh Patel?

  There's one that you think is low probability of next year where GPT-4 plus scaffolding plus un-hobblings gets you to AGI.

• 3:32:26Speaker 01

  Not GPT-4, you know, like.

• 3:32:27Dwarkesh Patel?

  Oh, sorry, sorry. So GPT-5, yeah.

• 3:32:28Speaker 01

  Yeah, yeah, yeah.

• 3:32:29Dwarkesh Patel?

  And there's ones where it takes much longer. There's ones where it's something that's a couple of years.

• 3:32:34Speaker 01

  27 in a modal world, yeah.

• 3:32:35Dwarkesh Patel?

  So GPT-4 seems pretty aligned in the sense that I don't expect it to go off the rails.

• 3:32:41Speaker 01

  Yeah.

• 3:32:41Dwarkesh Patel?

  And maybe with scaffolding, things might change.

• 3:32:43Speaker 01

  Pretty good.

• 3:32:43Dwarkesh Patel?

  Yeah, exactly. So the, and you, maybe you will keep turn, there's cranks, you keep going up and one of the cranks gets to ASI.

• 3:32:51Speaker 01

  Yeah.

• 3:32:52Dwarkesh Patel?

  Is there any point at which the sharp left turn happens? Is it when you start, is it the case that you think plausibly when they act more like agents, this is a thing to worry about?

• 3:33:02Speaker 01

  Yeah.

• 3:33:03Dwarkesh Patel?

  Is there anything qualitatively that you expect to change with regards to the alignment perspective?

• 3:33:07Speaker 01

  Yeah, yeah, yeah. So I don't know if I believe in this concept of sharp left turn, but I do think there's basically, I think there's important qualitative changes that happen between now and kind of like somewhat superhuman systems, kind of like early on the intelligence explosion. And then important qualitative changes that happen from like early in intelligence explosion to kind of like true superintelligence and all its power and might. And let's talk about both of those. And so, okay, so the first part of the problem is one, you know, we're going to have to solve ourselves, right? We have to kind of have to align the like initial AIs and the intelligence explosion, you know, the sort of automated out of Bradford. I think there's kind of like, I mean, two important things that change from GPT-4, right? So one of them is, um, um, you know, if you believe the story on like, you know, synthetic data RL or self-play to get past the data wall. And if you believe this on hobbling story, you know, at the end, you're going to have things, you know, they're agents, right?

• 3:33:52Speaker 01

  Including. They do long-term plans, right? They have long, long, you know, they're somehow they're able to act over long horizons, right? But you need that, right? That's the sort of prerequisite to be able to do the sort of automated AI research. And so, you know, I think there's basically, you know, I basically think sort of pre-training is sort of alignment neutral in the sense of like, it has all these representations, it has good representations, you know, as representations of doing bad things, you know, but, but there's, there's, it's not like, you know, scheming against you or whatever. I think this sort of misalignment can arise once you're doing more kind of long horizon training, right? And so you're training, you know, again, too simplified example, but to kind of illustrate, you know, you're training an AI to make money. And, um, you know, if you're just doing that with reinforcement learning, you know, it's, you know,

• 3:34:35Speaker 01

  it might learn to commit fraud or lie or to deceive or seek power, simply because those are successful strategies in the real world, right? So maybe, you know, RL is basically it explores, maybe it figures out like, oh, it tries to like hack and then it gets some money and that made more money. You know, and then if that's successful, if that gets rewarded, that's just reinforced. So basically I think there's sort of more serious misalignment, kind of like misaligned long-term goals that could arise between now and, or that sort of necessarily have to be able to arise if you're able to get long horizon system. That's one. What you want to do in that situation is you want to add side constraints, right? So you want to add, you know, don't lie, don't deceive, don't commit fraud. And so how do you add those side constraints, right? The sort of basic idea you might have is like RHF, right? You're kind of like, yeah, it has this goal of like, you know, make money or whatever, but you're watching what it's doing.

• 3:35:20Speaker 01

  It starts trying to like, you know, lie or deceive or fraud or whatever, break the law. You're just kind of like thumbs down, don't do that. You anti-reinforce that. The sort of critical issue that comes in is that these AI systems are getting superhuman, right? And, and they're going to be able to do things that are too complex for humans to evaluate, right? So again, even early on, you know, in the intelligence explosion, the automated AI researchers and engineers, you know, they might write millions, you know, billions, trillions of lines of complicated code, you know, they might be doing all sorts of stuff. You just like, don't understand anymore. And so, you know, in the million lines of code, you know, is it somewhere kind of like, you know, hacking, hacking or like exfiltrating itself? Or like, you know, trying to go for the nukes or whatever, you know, like you don't know anymore, right? And so this sort of like, you know, thumbs up, thumbs down, pure RHF doesn't fully work anymore. Second part of the picture, and we should maybe talk more about this.

• 3:36:05Speaker 01

  First part of the picture, I think it's going to be like, there's a hard technical problem of what do you do sort of post RHF? But I think it's a solvable problem and it's like, you know, there's various things I'm bullish on. I think there's like ways in which deep learning has shaped that favorably. The second part of the problem is you're going from your like initial systems and intelligence explosion to like super intelligence and you know it's like many ums it ends up being like by the end of it you have a thing that's vastly smarter than humans.

• 3:36:29Speaker 01

  I think the intelligence explosion is really scary from an alignment point of view because basically if you have this rapid intelligence explosion, you know, less than a year, two years or whatever, you're going, say in the period of a year from systems where like, you know, failure would be bad, but it's not catastrophic to like, you know, saying a bad word, it's like, you know, it's, it's something goes awry to like, you know, failure is like, you know, it excreted itself, it starts hacking the military, it can do really bad things. You're going less than a year from sort of a world in which like, you know, it's some descendant of current systems and you kind of understand it and it's like, you know, has good properties, but something that potentially has a very sort of alien and different architecture, right, after having gone through another decade of a maladvances. I think one example there that's very salient to me is legible and faithful chain of thought, right? So a lot of the time when we're talking about these things, we're talking about, you

• 3:37:14Speaker 01

  know, it has tokens of thinking and then it uses many tokens of thinking and, you know, maybe we bootstrap ourselves by, you know, it's pre-trained, it learns to think in English, then we do something else on top so it can do the sort of longer chains of thought. And so, you know, it's very plausible to me that like for the initial automated alignment researchers, you know, we don't need to do any complicated mechanistic interpretability and just like literally you read what they're thinking, which is great, you know, it's like huge advantage, right?

• 3:37:41Speaker 01

  However, I'm very likely not the most efficient way to do it, right? There's like probably some way to have a recurrent architecture, it's all internal states is a much more efficient way to do it. That's what you get by the end of the year. You know, you're going this year from like RLHF plus plus some extension works to like, it's vastly superhuman, it's like, you know, it's, it's, it's, it's to us like, you know, you know, an expert in the field might be to like an elementary schooler, middle schooler. And so, you know, I think it's the sort of incredibly sort of like hairy period for alignment. Thing you do have is you have the automated AI researchers, right? And so you can use the automated AI researchers to also do alignment.

• 3:38:18Dwarkesh Patel?

  And so in this world,

• 3:38:20Speaker 01

  Yeah.

• 3:38:21Dwarkesh Patel?

  Why are we optimistic that the project is being run by people who are thinking? I think so here's, here's, here's, here's something to think about.

• 3:38:29Speaker 01

  Okay.

• 3:38:30Dwarkesh Patel?

  The OpenAI starts off with people who are very explicitly thinking about exactly these kinds of things.

• 3:38:38Speaker 01

  Yes.

• 3:38:39Dwarkesh Patel?

  Right?

• 3:38:39Speaker 01

  But are they still there?

• 3:38:40Dwarkesh Patel?

  No, no, but here's the thing. No, no, even the people who are there, even like the current leadership is like exactly these things you can find them in interviews in their blog posts talking about. And what happens is when, as you were talking about, when some sort of trivial, and Jan talked about it, this is not just you, Jan talked about it in his tweet thread, when there is some trade off that has to be made with, we need to do this flashy release this week and not next week because whatever Google I/O is the next week, so we need to get it. And then the trade off is made in favor of the less, the more careless decision. When we have the government or the national security advisor, the military or whatever,

• 3:39:23Dwarkesh Patel?

  which is much less familiar with this kind of discourse, is naturally thinking in this way about, ah, I'm worried the chain of thought is unfaithful and how do we think about the features that are represented here? Why should we be optimistic that a project run by people like that will be thoughtful about these kinds of considerations?

• 3:39:40Speaker 01

  I mean, it might not be. You know, I, I agree. I think, all right, a few thoughts, right? First of all, I think the private world, even if they sort of nominally care, is extremely tough for alignment. A couple of reasons. One, you just have the race between the sort of commercial labs, right? And it's like, you don't have any headroom there to like, be like, ah, actually, we're going to hold back for three months, like, get this right. And, you know, we're going to dedicate 90% of our compute to automate alignment research instead of just like pushing the next boom. The other thing, though, is like, in the private world, you know, China has stolen your ways, China has your secrets, they're right on your tails, you're in this fever struggle, no room at all for maneuver. They're like, the way, it's like, absolutely essential to get alignment right, and you

• 3:40:24Speaker 01

  get it during this intelligence explosion, you get it right, is you need to have that room to maneuver, and you need to have that clear lead. And, you know, again, maybe you've made the deal or whatever, but I think you're in incredibly tough space, tough spot if you don't have this clearly. So I think the sort of private world is kind of rough there. On like whether people will take it seriously, you know, I don't know, I have some faith in sort of, sort of, normal mechanisms of a liberal society.

• 3:40:46Dwarkesh Patel?

  Sure.

• 3:40:46Speaker 01

  That sort of, if, if, if alignment is an issue, which, you know, we don't fully know yet, but sort of the science will develop, we're going to get better measurements of alignment, you know, and, and the case will be clear and obvious. I worry that there's, you know, I worry about worlds where evidence is ambiguous. And I think a lot of, um, a lot of the most scary kind of intelligence explosion scenarios are worlds in which evidence is ambiguous. But again, it's sort of like, I, if evidence is ambiguous, then that's the worlds in which you really want the safety margins. And that's also the worlds in which kind of like running the intelligence explosion is sort of like, you know, running a war, right? It's like, ah, the evidence is ambiguous. You have to make these really tough trade offs and you like, you better have a really good chain of command for that. And it's not just like, you know, YOLOing at, ah, let's go, you know, it's cool. Um, yeah.

• 3:41:26Dwarkesh Patel?

  Let's talk a little bit about Germany.

• 3:41:28Speaker 01

  Uh huh.

• 3:41:28Dwarkesh Patel?

  We're making the analogy to World War II. And you made a really interesting point many hours ago.

• 3:41:38Speaker 01

  We should start after, you know, after the marathon.

• 3:41:43Dwarkesh Patel?

  The fact that throughout history, World War II is not unique, at least when you think in proportion to the size of the population.

• 3:41:51Speaker 01

  Yep.

• 3:41:52Dwarkesh Patel?

  But these other sorts of catastrophes where some significant portion of the population has been killed off.

• 3:42:00Speaker 01

  Yeah.

• 3:42:01Dwarkesh Patel?

  After that, the nation recovers and they get back to their heights.

• 3:42:06Speaker 01

  Uh huh.

• 3:42:07Dwarkesh Patel?

  So what's interesting after World War II is that Germany especially, and maybe Europe as a whole. Obviously they experienced fast economic growth in the direct aftermath because of catch up growth. But subsequently, we just don't think of Germany as, no, we're not talking about Germany potentially launching an intelligence explosion. And they're going to get and see at the AI table. We were talking about Iran and North Korea and Russia. We didn't talk about Germany, right?

• 3:42:32Speaker 01

  Well, because they're allies.

• 3:42:33Dwarkesh Patel?

  Yeah, yeah, yeah. But so what happened? I mean, the World War II and now it didn't like come back after the seven years war or something, right?

• 3:42:42Speaker 01

  Yeah, yeah, yeah. I mean, look, I'm generally very bearish on Germany. I think in this context, I'm kind of like, you know, it's a little bit, you know, I think you're underrating a little bit. I think it's probably still one of the, you know, top five most important countries in the world. You know, I mean, Europe overall, you know, it still has, I mean, it's a GDP that's like close to the United States, the size of the GDP, you know? And there's things actually that Germany is kind of good at, right? Like state capacity, right? Like, you know, the, the, you know, the roads are good and they're clean and they're well maintained. And, and, um, you know, in some sense, the sort of, um, a lot of this is the sort of flip side of things that I think are bad about Germany.

• 3:43:14Dwarkesh Patel?

  Yeah.

• 3:43:14Speaker 01

  Right? So in the US, it's a little bit like, there's a bit more of a sort of Wild West feeling to the United States, right? And it includes the kind of like crazy bursts of creativity, includes like, you know? Political candidates that are sort of, you know, there's there's a much broader spectrum. And, you know, much, you know, like both an Obama and a Trump is somebody you just wouldn't see in the sort of much more confined kind of German political debate. You know, I wrote this blog post at some point, Europe's political stupor about this.

• 3:43:39Speaker 01

  But anyway, and so there's this sort of punctilious sort of rule following that is like good in terms of like, you know, keeping your kind of state capacity functioning.

• 3:43:47Speaker 01

  But that is also, you know, I think I kind of, I think there's a sort of very constrained view of the world in some sense.

• 3:43:56Speaker 01

  You know, and that includes kind of, you know, I think after World War II, there's a real backlash against anything like elite, you know, and, you know, again, no, you know, no elite high schools or elite colleges and sort of

• 3:44:07Dwarkesh Patel?

  Why is that the logic?

• 3:44:08Speaker 01

  Excellence isn't cherished, you know, there's, yeah.

• 3:44:11Dwarkesh Patel?

  Why is that the logical intellectual thing to rebel against if what if you're trying to overcorrect from the Nazis?

• 3:44:19Speaker 01

  Yeah.

• 3:44:19Dwarkesh Patel?

  Was it because the Nazis were very much into elitism? What was, I don't understand why that's a logical sort of counter reaction.

• 3:44:25Speaker 01

  I know, maybe it was sort of a counter reaction against the sort of like whole like Aryan race and the sort of that sort of thing. I mean, I also just think there is a certain amount in what amount, certain, I mean, look, look at sort of World War One, end of World War One versus end of World War Two for Germany, right? And sort of, you know, a common narrative is that the peace of Versailles, you know, was was too strict on Germany. But, you know, the peace imposed after World War Two was like much more strict, right? It was a complete, you know, I mean, the whole country was destroyed, you know, it was, you know, and all the main, most of the major cities, you know, over half of the housing stock had been destroyed, right? Like, you know, in some birth cohorts, you know, like 40% of the men had died.

• 3:45:01Dwarkesh Patel?

  Half the population displaced.

• 3:45:03Speaker 01

  Oh, yeah. I mean, almost 20 million people, right, displaced, right? Huge, crazy, right? You know, like

• 3:45:08Dwarkesh Patel?

  And the borders are way smaller than the Versailles borders.

• 3:45:10Speaker 01

  Yeah, exactly. And, and, and sort of complete imposition of a new political system and, and, you know, on both sides, you know, and

• 3:45:20Speaker 01

  Yeah, so it was, but in some sense that worked out better than the post World War I piece, where then there was this kind of resurgence of German nationalism and, you know, in some sense, a thing that has been a pattern. So it's sort of like, it's unclear if you want to wake the sleeping beast. I do think that at this point, you know, it's gotten a bit too sleepy.

• 3:45:36Dwarkesh Patel?

  Yeah. I do think it's an interesting point about we underrate the American political system, and I've been making the same correction myself.

• 3:45:43Speaker 01

  Yeah.

• 3:45:44Dwarkesh Patel?

  There's, there was this book about, written by a Chinese economist called China's Worldview.

• 3:45:50Speaker 01

  Yeah.

• 3:45:51Dwarkesh Patel?

  And overall, I wasn't a big fan, but they made a really interesting point in there.

• 3:45:55Speaker 01

  Yeah.

• 3:45:56Dwarkesh Patel?

  Which was the way in which candidates rise up through the Chinese

• 3:46:01Speaker 01

  Yeah.

• 3:46:02Dwarkesh Patel?

  hierarchy for politics, for administration.

• 3:46:05Speaker 01

  Uh huh.

• 3:46:06Dwarkesh Patel?

  In some sense, it selects for, you're not going to get some Marjorie Taylor Greene or somebody running some

• 3:46:10Speaker 01

  Right, right, right, right. Don't get that in Germany either.

• 3:46:12Dwarkesh Patel?

  Right.

• 3:46:13Speaker 01

  Yeah.

• 3:46:14Dwarkesh Patel?

  But you're, he explicitly made the point in the book that that also means we're never going to get a Henry Kissinger or Barack Obama.

• 3:46:19Speaker 01

  Right.

• 3:46:20Dwarkesh Patel?

  In China.

• 3:46:21Speaker 01

  Totally, yeah.

• 3:46:22Dwarkesh Patel?

  We're going to get like, by the time they end up in the charge of the Politburo, on Politburo, there'll be like some 60-year-old bureaucrat who's never like ruffled any feathers.

• 3:46:28Speaker 01

  Yeah, yeah, yeah. I mean, I think, I think there's something really important about the sort of like very raucous political debate. And, I mean, yeah, in general, kind of like, you know, there's the sense in which in America, you know, lots of people live in their kind of like own world. I mean, like, we live in this kind of bizarre little like bubble in San Francisco and people, you know, and, and, but I, I think that's important for the sort of evolution of ideas, error correction, that sort of thing. You know, there's other ways in which the German system is more functional.

• 3:46:55Dwarkesh Patel?

  Yeah.

• 3:46:56Speaker 01

  But it is interesting that It makes major mistakes, right? Like the sort of defense spending, right?

• 3:47:00Dwarkesh Patel?

  Yeah.

• 3:47:01Speaker 01

  And, you know, then, you know, Russia invades Ukraine and, and you're like, wow, what did we do?

• 3:47:05Dwarkesh Patel?

  Right. No, that's a really good point, right?

• 3:47:06Speaker 01

  Yeah.

• 3:47:07Dwarkesh Patel?

  The main issues, there's, everybody agrees, but

• 3:47:09Speaker 01

  There's no debate about it, yeah, exactly.

• 3:47:10Dwarkesh Patel?

  Exactly. Yeah.

• 3:47:11Speaker 01

  It's a consensus blob kind of thing.

• 3:47:13Dwarkesh Patel?

  Right.

• 3:47:14Speaker 01

  And on the China point, you know, just having this experience of like reading German newspapers and I think how much, you know, how much more poorly I would understand the sort of German debate and sort of the sort of state of mind from just kind of afar. I worry a lot about, you know, or I think it is interesting just how kind of impenetrable China is to me. It's a billion people, right?

• 3:47:33Dwarkesh Patel?

  Right.

• 3:47:34Speaker 01

  And like, you know, almost everything else is really globalized, you have a globalized internet and I kind of, I kind of have a sense of what's happening in the UK. You know, I probably, even if I didn't read German newspapers, just sort of would have a sense of what's happening in Germany. But I really don't feel like I have a sense of what like You know, what is the state of mind? What is the state of political debate, you know, of a sort of average Chinese person or like an average Chinese leader?

• 3:47:55Dwarkesh Patel?

  Yeah.

• 3:47:56Speaker 01

  Yeah, I think that that I find that distance kind of worrying and I, you know, you know, and there's, and you know, there's some people who do this and they do really great work where they kind of go through the like party documents and the party speeches. And it seems to require a kind of a lot of interpretive ability where there's like very specific words and Mandarin that like mean we'll have one connotation, not the other connotation. But yeah, I think it's sort of interesting, given how globalized everything is.

• 3:48:17Dwarkesh Patel?

  Yeah.

• 3:48:18Speaker 01

  I mean, now we have basically perfect translation machines and it's still so, so impenetrable.

• 3:48:22Dwarkesh Patel?

  That's really interesting. I've been, I should, I'm sort of ashamed almost that I haven't done this yet.

• 3:48:27Speaker 01

  Yeah.

• 3:48:27Dwarkesh Patel?

  I think many months ago, I, when Alexi interviewed me on his YouTube channel, I said, I'm meaning to go to China to actually see for myself what's going on. And actually, I'm, I should, so by the way, if anybody listening has a lot of context on China, if I went to China, who could introduce me to people?

• 3:48:43Speaker 01

  Yeah.

• 3:48:43Dwarkesh Patel?

  Please email me.

• 3:48:44Speaker 01

  Well, you gotta, you gotta do some pods and you gotta, you gotta find some of the Chinese AI researchers, man.

• 3:48:47Dwarkesh Patel?

  I know. I was thinking at some point, again, this is.

• 3:48:51Speaker 01

  Can they speak freely, but you know?

• 3:48:52Dwarkesh Patel?

  Huh?

• 3:48:53Speaker 01

  I don't know if they can speak freely, but.

• 3:48:54Dwarkesh Patel?

  I was thinking of there's, so they had these papers and on the paper, they'll say who's a co-author.

• 3:48:59Speaker 01

  Yeah.

• 3:48:59Dwarkesh Patel?

  It's funny because, well, I was thinking of just emailing, cold emailing everybody, like, here's my calendar, like, can you, let's, let's just talk. I just want to see what is the vibe, even they don't tell me anything. I'm just like, what kind of person is this? Are they, how westernized are they?

• 3:49:11Speaker 01

  Yeah.

• 3:49:12Dwarkesh Patel?

  But I, as I was saying this, I just remembered that, in fact, ByteDance, according to mutual friends we have at Google, they cold emailed every single person on the Gemini paper and said, if you come work for ByteDance.

• 3:49:26Speaker 01

  Yeah.

• 3:49:26Dwarkesh Patel?

  We'll make you an L L D engineer, you'll report directly to the CTO. And in fact, this actually, I'm gonna go.

• 3:49:32Speaker 01

  That's how the secrets go over, right?

• 3:49:33Dwarkesh Patel?

  Right. No, I'm actually, I meant to ask this earlier, but suppose they hired what, if there's only a hundred or so people, maybe less, were working on the key algorithmic secrets.

• 3:49:41Speaker 01

  Yeah, yeah, yeah.

• 3:49:42Dwarkesh Patel?

  If they hired one such person.

• 3:49:43Speaker 01

  Yeah.

• 3:49:44Dwarkesh Patel?

  Is all the alpha gone that these labs have?

• 3:49:46Speaker 01

  If this person was intentional about it, they could get a lot. I mean, they couldn't get the sort of like, I mean, actually, they could probably just also exclude the code. They could get a lot of the key ideas. Again, like, you know, up until recently, stuff was published, but you know, they could get a lot of the key ideas if they tried, if they like, you know, I think there's a lot of people who don't actually kind of like look around to see what the other teams are doing, but you know, I think you kind of can.

• 3:50:05Speaker 01

  But yeah, I mean, they could. It's scary.

• 3:50:08Dwarkesh Patel?

  Right. I think the project makes more sense there where you can't just recruit a Manhattan project engineer and then just get.

• 3:50:14Speaker 01

  And it's like, these are secrets that can be used for like, probably every training run in the future. That'll be like, maybe are the key to the data wall that are like, they can't go on, or they can't go on, that are like, you know, they're going to be worth, you know, given sort of like the multipliers on compute, you know, hundreds of billions, trillions of dollars, you know, and all it takes is, you know, China to offer a hundred million dollars to somebody and be like, ah, come work for us. And then, and then.

• 3:50:36Dwarkesh Patel?

  Yep.

• 3:50:36Speaker 01

  I mean, yeah, I'm really uncertain on how sort of seriously China is taking AGI right now. One anecdote that was related to me on the topic of anecdotes, by another sort of like, you know, kind of researcher in the field was, at some point, they were at a conference with somebody, Chinese AI researcher, and he was talking to him and he was like, ah, I think it's really good that you're here. And like, you know, we got to have the international coordination and stuff. And apparently, this guy said that I'm the kind of most, senior most person that they're going to let leave the country to come to things like this.

• 3:51:06Dwarkesh Patel?

  Wait, what's the takeaway?

• 3:51:09Speaker 01

  As in, they're not letting really senior AI researchers leave the country.

• 3:51:12Dwarkesh Patel?

  Interesting.

• 3:51:12Speaker 01

  Kind of classic, you know, Eastern Bloc move.

• 3:51:14Dwarkesh Patel?

  Yeah.

• 3:51:16Speaker 01

  I don't know if this is true, but it's what I heard.

• 3:51:17Dwarkesh Patel?

  That's interesting. So I thought the point you made earlier about being exposed to German newspapers and also to, because earlier you were interested in economics and of law and national security, you have, the variety and intellectual diet there has exposed you to thinking about the geopolitical question here in ways, others, talking about AI. I mean, this is the first episode I've done about this, where we've talked about things like this, which is, now that I think about it, weird, given that this is an obvious thing in retrospect, I should have been thinking about. Anyways, so that's one thing we've been missing. What are you missing? And national security, you're thinking about, so you can't say national security. What like, perspective are you probably underexposed to as a result? And China, I guess you mentioned?

• 3:52:00Speaker 01

  Yeah, so I think the China one is an important one. I mean, I think another one would be a sort of very Tyler Cowen-esque take, which is like, you're not exposed to how, like, how will a normal person in America, like, you know, both like, use AI? Probably not. And that being kind of like bottlenecks to the fusion of these things. And I'm overrating the revenue because I'm kind of like, ah, you know, everyone at S F is adopting it. But, you know, kind of like, you know, Joe Schmo engineer at a company, you know, like, ah, will they, will they be able to integrate it? And then also the reaction to it, right? You know, I mean, I think this was a question, again, hours ago, where it was about like, you know, won't people kind of rebel against this?

• 3:52:36Dwarkesh Patel?

  Yeah.

• 3:52:36Speaker 01

  And they won't want to do the project. I don't know, maybe they will. Yeah.

• 3:52:41Dwarkesh Patel?

  Here's a political reaction that I didn't anticipate.

• 3:52:43Speaker 01

  Yeah.

• 3:52:44Dwarkesh Patel?

  So Tucker Carlson has recently ended the Joe Rogan episode. I already told you about this, but I'm just gonna tell the story again. So Tucker Carlson is on Joe Rogan.

• 3:52:53Speaker 01

  Yeah.

• 3:52:54Dwarkesh Patel?

  And they start talking about World War II.

• 3:52:56Speaker 01

  Ah.

• 3:52:57Dwarkesh Patel?

  And Tucker says, well, listen, I'm gonna say something that my fellow conservatives won't like, but I think nuclear weapons are immoral. I think it was obviously immoral that we use them on Nagasaki and Hiroshima. And then he says, in fact, nuclear weapons are always immoral.

• 3:53:12Speaker 01

  Uh huh.

• 3:53:14Dwarkesh Patel?

  Except when we would use them on data centers. In fact, it would be immoral not to use them on data centers. Because, look, these people in Silicon Valley, these fucking nerds are making A G, superintelligence, and they say that it could enslave humanity. We made machines to serve humanity, not to enslave humanity. And they're just going on and making these machines. And so we should, of course, be nuking the data centers.

• 3:53:40Dwarkesh Patel?

  And that is definitely.

• 3:53:41Speaker 01

  Real horseshoe, real horseshoe.

• 3:53:42Dwarkesh Patel?

  And that is definitely not a political reaction in 2024, I was expecting.

• 3:53:46Speaker 01

  Ah, who, I mean, who knows, man?

• 3:53:48Dwarkesh Patel?

  It's gonna be crazy.

• 3:53:49Speaker 01

  It's gonna be crazy.

• 3:53:50Dwarkesh Patel?

  The thing we learned with COVID is that also the left-right reactions that you would anticipate just based on hunches.

• 3:53:58Speaker 01

  It completely flipped multiple times, right? Initially, like, kind of the right was like, you know,

• 3:54:01Dwarkesh Patel?

  It's like so contingent.

• 3:54:02Speaker 01

  And then, and then, and then, and the left was like, this is racist. And then it flipped, you know, the left was really into the COVID. Yeah.

• 3:54:08Dwarkesh Patel?

  Yeah.

• 3:54:08Speaker 01

  And the whole thing also is just like so blunt and crude. And so, I don't know, I think, I think probably in general, you know, I think people are really under, you know, people like to make sort of complicated technocratic AI policy proposals. And I think, especially if things go kind of fairly rapidly on the left at A G I.

• 3:54:25Speaker 01

  You know, there might not actually be that much space for kind of like complicated, kind of like, you know, clever proposals. It might just be kind of much cruder reactions.

• 3:54:32Dwarkesh Patel?

  Yeah. Look, and then also when you mentioned the spies and national security getting involved and everything, and you can talk about that in the abstract, but now that we're living in San Francisco and we know many of the people who are doing the top AI research, it's also a little scary to think about people I personally know and friends with. It's not unfeasible if they have secrets in their head that are worth a hundred billion dollars or something, kidnapping, assassination, sabotage.

• 3:55:00Speaker 01

  Oh, their family or yeah, it's really bad. I guess it's to the point on security, you know, like right now it's just really foreign, but you know, at some point as it becomes like really serious, it's, you know, you're going to want the security cards.

• 3:55:13Dwarkesh Patel?

  Yeah.

• 3:55:13Speaker 01

  Yeah.

• 3:55:14Dwarkesh Patel?

  Yeah. So presumably you have thought about the fact that people in China will be listening to this and will be reading your series.

• 3:55:22Speaker 01

  Yeah.

• 3:55:23Dwarkesh Patel?

  And somehow you made the trade off that it's better to let the whole world know.

• 3:55:30Speaker 01

  Yeah.

• 3:55:31Dwarkesh Patel?

  And also including China.

• 3:55:32Speaker 01

  Yeah.

• 3:55:33Dwarkesh Patel?

  And make them up to AGI, which is part of the thing you're worried about is China waking up to AGI.

• 3:55:37Speaker 01

  Yeah.

• 3:55:37Dwarkesh Patel?

  Than to stay silent.

• 3:55:39Speaker 01

  Yeah.

• 3:55:39Dwarkesh Patel?

  I'm just curious, walk me through how you've thought about that trade off.

• 3:55:43Speaker 01

  Yeah, I actually, look, I think this is a tough trade off. I thought about this a bunch, you know, I think, you know, I think people in the PRC will read this.

• 3:55:56Speaker 01

  I think, you know, I think there's some extent to which sort of cat is out of the bag, you know, this is like not, you know, AGI being a thing people are thinking about very seriously is not new anymore. There's the sort of, you know, a lot of these takes are kind of old or, you know, I've had, I had, you know, similar views a year ago, might not have written it up a year ago, in part because I think this cat wasn't out of the bag enough. You know, I think the other thing is.

• 3:56:19Speaker 01

  I think to be able to manage this challenge, you know, I think much broader swaths of society will need to wake up, right? And if we're going to get the project, you know, we actually need sort of like, you know, a broad bipartisan understanding, the challenge is facing us. And so, you know, I think it's a tough trade off, but I think the sort of need to wake up people in the United States, in the sort of Western world, in the Democratic coalition, is ultimately imperative. And, you know, I think my hope is more people here will read it than in the PRC. You know, and I think people sometimes underrate the importance of just kind of like writing it up, laying out the strategic picture. And, you know, I think you've done actually a great service to sort of mankind in some sense by, you

• 3:56:59Speaker 01

  know, with your podcast. And, you know, I think it's overall been good.

• 3:57:06Dwarkesh Patel?

  Okay, so.

• 3:57:07Speaker 01

  By the way, you know, on the topic of, you know, Germany, you know, we were talking at some point about kind of immigration stories. I feel like you have a kind of interesting story you haven't told, and I think you should tell.

• 3:57:16Dwarkesh Patel?

  So, a couple of years ago, I was in college and I was 20. I was about to turn 21.

• 3:57:24Speaker 01

  Yeah.

• 3:57:24Dwarkesh Patel?

  I think it was.

• 3:57:25Speaker 01

  So you're, yeah, you came from India when you were really young.

• 3:57:27Dwarkesh Patel?

  Right, yeah, yeah. So until I was eight or nine, I lived in India, and then we moved around all over the place. But because of the backlog for Indians.

• 3:57:37Speaker 01

  Yeah, the green card backlog.

• 3:57:38Dwarkesh Patel?

  Yeah, it's, we were, we've been in the queue for like decades.

• 3:57:43Speaker 01

  Even though you came at eight, you're still on the, you know, H1B.

• 3:57:46Dwarkesh Patel?

  Yeah. And when you're 21, you get kicked off the queue and you had to restart the process. I'm on my dad's, my dad's a doctor and I'm on his H1B as a dependent. But when you're 21, you get kicked off. And so I'm 20 and it just like kind of dawns on me that this is my situation.

• 3:57:59Speaker 01

  Yeah, and you're completely screwed.

• 3:58:01Dwarkesh Patel?

  Right. And so I also had the experience that my dad, we've like moved all around the country, they have to prove that him as a doctor is like, you can't get native talent.

• 3:58:10Speaker 01

  Yeah, and you can't start a startup, you know, you're just like completely changed. And like even getting the H1B for you would have been like, you know, 20% lottery. So if you're lucky, you're in this.

• 3:58:19Dwarkesh Patel?

  And they had to prove that they can't get native talent, which means like for him, and like, we lived in North Dakota for three years, West Virginia for three years, Maryland, West Texas. And so it kind of dawned on me. This is my situation. As I turn 21, I'll be like on this lottery, even if I get the lottery, I'll be a fucking code monkey for the rest of my life because this this thing isn't going to let up.

• 3:58:37Speaker 01

  Yeah, can't do a startup.

• 3:58:38Dwarkesh Patel?

  Exactly. And so at the same time, I had been reading for the last year, I've been super obsessed with Paul Graham essays. My plan at the time was to make a startup or something. I was super excited about that. And it just occurred to me that I couldn't do this. That like this is just not in the cards for me. And so I was kind of depressed about it. I remember I kind of just,

• 3:58:59Dwarkesh Patel?

  I was in a daze through finals because I had like, it just occurred to me and I was really like anxious about it. And I remember thinking to myself at the time, that if somehow I end up getting my green card before I turn 21, there's no fucking way I'm turning, becoming a code monkey. Because the thing that I've like this feeling of dread that I have is this realization that I'm just gonna have to be a code monkey. And I realized that's my default path. If I if I hadn't sort of made a proactive effort not to do that, I would have graduated college as a computer science student and I would have just done that. And that's the thing I was super scared about.

• 3:59:36Dwarkesh Patel?

  So that was an important sort of realization for me. Anyway, so COVID happened because of that, since there weren't foreigners coming, the backlog cleared fast. And by the skin of my teeth, like a few months before I turned 21, extremely contingent reasons, I ended up getting a green card. Because I got a green card, I could,

• 3:59:55Speaker 01

  The whole podcast, right?

• 3:59:57Dwarkesh Patel?

  Exactly. I graduated college and I was like bumming around and I got was like, I graduated a semester early. I'm gonna like do this podcast, see what happens. And it didn't have a green card.

• 4:00:09Speaker 01

  You know, and it only existed because, yeah, it's actually because I think it's hard, it's probably, you know, what is the impact of like immigration reform? What is the impact of clearing, you know, like whatever 50,000 green cards in the backlog? And you're such like an amazing example of like, you know, all of this is only possible. And it's, yeah, it's, I mean, it's just incredibly tragic that this is so dysfunctional.

• 4:00:30Dwarkesh Patel?

  Yeah, yeah, yeah. No, it's insane.

• 4:00:33Speaker 01

  I'm glad you did it. I'm glad you kind of like, you know, tried the, you know, the unusual path.

• 4:00:39Dwarkesh Patel?

  Well, yeah, but I could only do it. Obviously, I was extremely fortunate that I got the green card. I was like, I had a little bit of saved up money. I got a small grant out of college. Thanks to the future fund to like do this for basically the equivalent of six months. And so it turned out really well. And then each time and I was like, oh, the okay. Podcast, come on, like I wasted a few months on this. Let's now go do something real. Something big would happen.

• 4:01:08Speaker 01

  And you kept with it.

• 4:01:10Dwarkesh Patel?

  Yeah, yeah. But there would always be just like the moment I'm about to quit the podcast. Something like Jeff Bezos would say that something nice about me on Twitter. The Aliya episodes gets like a half a million views, you know? And then now this is my career, but it was a sort of very, looking back on it incredibly contingent. That things worked out the right way.

• 4:01:27Speaker 01

  Yep. I mean, look, if the AGI stuff goes down, you know, it will be, uh, it'll be the most important kind of like, you know, source of, uh, it'll be how maybe most of the people who kind of end up feeling AGI.

• 4:01:39Dwarkesh Patel?

  Yeah, yeah. Also very much, uh, you're very linked with the story in many ways. First, the, um, the, I got like a $20,000 grant from a future fund. Right out of college and that sustained to me for six months, however long it was. And without that, I wouldn't.

• 4:01:58Speaker 01

  It's kind of crazy.

• 4:01:58Dwarkesh Patel?

  Yeah, 10 grand or what was it?

• 4:02:00Speaker 01

  No, it's just, it's tiny, but you know, it goes to show kind of how far small grants can go.

• 4:02:04Dwarkesh Patel?

  Yeah.

• 4:02:04Speaker 01

  It's sort of the emergency ventures too.

• 4:02:06Dwarkesh Patel?

  Exactly. The emergency ventures and the, um, well, the last year I've been in San Francisco. We've just been in close contact the entire time and just bouncing ideas back and forth. Learned just basically the alpha I have, I think people would be surprised by how much I got from you, Sholto, Trent and a couple others just.

• 4:02:26Speaker 01

  It's been, it's been an absolute pleasure.

• 4:02:27Dwarkesh Patel?

  Yeah, likewise, likewise. It's been super fun.

• 4:02:31Dwarkesh Patel?

  Okay. So some random questions for you. If you could convert to Mormonism.

• 4:02:35Speaker 01

  Yeah.

• 4:02:35Dwarkesh Patel?

  And you could really believe it.

• 4:02:36Speaker 01

  Yeah.

• 4:02:37Dwarkesh Patel?

  Would you do it? Would you push the button?

• 4:02:40Speaker 01

  Well, okay, okay. Uh, before I answer that question, one sort of observation about the Mormons. So actually there's a, there's an article that actually made a big impact on me. I think it was by McKee Coppins at some point, you know, in the Atlantic or whatever about the Mormons. And I think the thing he kind of. You know, and I think he even like interviewed Mitt Romney in it and so on. And I think the thing I thought was really interesting in this article was he kind of talked about how the experience of kind of growing up different, you know, growing up very unusual, especially if you grow up Mormon outside of Utah, you know, like the only person doesn't drink caffeine. You don't drink alcohol. You're kind of weird. How that kind of got people prepared for being willing to be kind of outside of the norm later on. And like, you know, Mitt Romney, you know, was willing to kind of take stands alone, you know, in his party because he believed, you know, what he believed is true.

• 4:03:23Speaker 01

  And I don't know, I mean, probably not to the same way, but I feel a little bit like this from kind of having grown up in Germany, having, you know, and really not having like this sort of German system and having been kind of an outsider or something. I think there's a certain amount in which kind of, yeah, growing up an outsider gives you kind of unusual strength later on to be kind of like, you know, willing to say what you think.

• 4:03:41Dwarkesh Patel?

  Yeah.

• 4:03:42Speaker 01

  And so that is one thing I really appreciate about Mormons, at least the ones that, you know, grew up outside of Utah. The fertility rates, they're good, they're important. They're going down as well, right?

• 4:03:51Dwarkesh Patel?

  Right.

• 4:03:51Speaker 01

  This is, this is the thing that really clinched the kind of fertility decline story.

• 4:03:54Dwarkesh Patel?

  Yeah.

• 4:03:56Speaker 01

  Yeah, even the Mormons, right?

• 4:03:57Dwarkesh Patel?

  You're like, oh, this is like a sort of a good start. Mormons will replace everybody.

• 4:04:00Speaker 01

  I don't know if it's good, but it's like, at least, you know, at least come on. You know, like it's at least some people retain high, you know, but it's no, no, you know, even the Mormons and sort of basically once, once the sort of these religious subgroups have high fertility rates.

• 4:04:09Dwarkesh Patel?

  Right.

• 4:04:09Speaker 01

  Once they kind of grow big enough, they become, they're too close in contact with sort of normal society and become normalized. Mormon fertility rates dropped from, I don't remember the exact numbers, maybe like four to two in the course of 10, 20 years. Anyway, so it's like, you know, now people point to the, you know, Amish or whatever, but I'm just like, it's probably just not scalable. And if you grow big enough, then there's just like, you know, the sort of like, you know, the sort of like overwhelming force of modernity kind of gets you.

• 4:04:31Dwarkesh Patel?

  Yeah.

• 4:04:32Speaker 01

  No, if I could convert to Mormonism, look, I think there's something, I don't believe it, right? If I believed it, I obviously would convert to Mormonism, right? Because it's you gotta, you gotta.

• 4:04:41Dwarkesh Patel?

  But you can choose a world in which you do believe it.

• 4:04:48Speaker 01

  I think there's something really valuable in believing in something greater than yourself and believing, having a certain amount of faith.

• 4:04:55Dwarkesh Patel?

  You do, right? That's what the series is.

• 4:04:57Speaker 01

  Yeah, and you know, there's a, you know, feeling some sort of duty to the thing greater than yourself. And you know, maybe my version of this is somewhat different. You know, I think I feel some sort of duty to like, I feel like there's some sort of historical weight on like how this might play out. And I feel some sort of duty to like make that go well. I feel some sort of duty to, you know, our country, to the national security of the United States. And, um, you know, I think, I think that, I think I can be a force for a lot of good.

• 4:05:28Dwarkesh Patel?

  I, the, going back to the opening, I think, just, uh, the, the thing that's especially impressive about that is, look, there's people who, at the company, who have through years and decades of building up savings from working in tech, have probably tens of millions liquid, more than that in terms of their equity. And the person, very many people were concerned about the clusters in the Middle East and the secrets leaking to China and all these things. But the person who actually made a hassle about it, and I think hassling people is so underrated.

• 4:06:09Dwarkesh Patel?

  I think that one person who made a hassle about it is the 22-year-old who has less than a year of the company, who doesn't have savings built up, who isn't like a solidified member of the, um, I think this is sort of like-

• 4:06:24Speaker 01

  Maybe, maybe it's me being naive and, you know, not having, knowing how big companies work and, you know. But like, there's a, you know, I think sometimes a bit of a speech geontologist, you know, I kind of believe in saying what you think. Sometimes friends tell me I should be more of a speech consequentialist.

• 4:06:38Dwarkesh Patel?

  No, I think, um, I, I, I really think-

• 4:06:43Dwarkesh Patel?

  The amount of people who, when they have the opportunity to talk to the person will just bring up the thing. I've been with you in multiple contexts, and I guess I shouldn't reveal who the person is or what the context was. But I've just been like very impressed that the dinner begins and by the end, somebody who has a major voice in how things go is seriously thinking about a worldview they would have found incredibly alien before the dinner or something. And I've been impressed with it, like, just like, just give them the spiel and hassle them.

• 4:07:16Speaker 01

  I mean, look, I just, I think, I think I feel this stuff pretty viscerally now. You know, I think there's a time, you know, there's a time when I thought about this stuff a lot, but it was kind of like econ models and like, you know, kind of like these sort of theoretical abstractions. And, you know, you talk about human brain size or whatever.

• 4:07:29Dwarkesh Patel?

  Right.

• 4:07:30Speaker 01

  And I think, you know, since, I think since at least last year, you know, I feel like, you know, I feel like I can see it, you know, and I, I just, I feel it. And I think I can like, you know, I can sort of see the cluster that AGI is going to be training on. And I can see the kind of rough combination of algorithms and the people that be involved and how this is going to play out. And, you know, I think, look, we'll see how it plays out. There's many ways this could be wrong. There's many ways it could go. But I think this could get very real.

• 4:07:56Dwarkesh Patel?

  Yeah. Should we talk about what you're up to next?

• 4:07:59Speaker 01

  Sure. Yeah.

• 4:08:00Dwarkesh Patel?

  Okay. So you're starting an investment firm.

• 4:08:02Speaker 01

  Yep.

• 4:08:02Dwarkesh Patel?

  Anchor investments from Nat Friedman, Daniel Gross, Patrick Collison, John Collison.

• 4:08:09Dwarkesh Patel?

  First of all, why is this thing to do? You believe the AGI is coming in a few years? Why the investment firm?

• 4:08:17Speaker 01

  A good question. Fair question. All right. So I mean, a couple of things. One is just, you know, I think we talked about this earlier, but it's like the screen doesn't go blank, you know, when sort of AGI or superintelligence happens. I think people really underrate the sort of, basically the sort of decade after. You have the intelligence explosion. That's maybe the most sort of wild period. But I think the decade after is also going to be wild. And, you know, this combination of human institutions, but superintelligence, you have crazy kind of geopolitical things going on. You have the sort of broadening of this explosive growth. And basically, yeah, I think it's going to be a really important period. I think capital will really matter. You know, eventually, you know, like, you know, going to go to the stars, you know, going to go to the galaxies. So anyway, so part of the answer is just like, look, I think done right, there's a lot of money to be made, you know. I think if AGI were priced in tomorrow, you could

• 4:08:58Speaker 01

  maybe make 100x. Probably you can make even way more than that because of the sequencing. And, and, and, you know, capital matters. I think the other reason is just.

• 4:09:10Speaker 01

  You know, some amount of freedom and independence. And I think, you know,

• 4:09:15Speaker 01

  you know, I think there's some people who are very smart about this AI stuff and who are kind of like see it coming. But I think almost all of them, you know, are kind of, you know, constrained in various ways, right? They're in the labs, you know, they're in some, you know, some other position where they can't really talk about this stuff. And, um, you know, in some sense, I've really admired sort of the thing you've done, which is I think it's really important that there's sort of voices of reason on this stuff publicly, or people who are in positions to kind of advise important actors and so on. And so I think there's a, you know, basically the thing this investment firm will be, will be kind of like, you know, a brain trust on AI. It's going to be all that situational awareness. We're going to have the best situational awareness in the business. You know, we're going to have way more situational business than any of the people who manage money in the, you know, New York.

• 4:09:51Dwarkesh Patel?

  Yeah.

• 4:09:52Speaker 01

  We're definitely going to, you know, we're going to do great on investing. But it's the same sort of situational awareness that I think is going to be important for understanding what's happening, being a voice of reason publicly and, and, and sort of being able to be in a position to advise.

• 4:10:05Dwarkesh Patel?

  Yeah. I, uh, there, the book about Peter Thiel.

• 4:10:10Speaker 01

  Yeah.

• 4:10:11Dwarkesh Patel?

  They had, um, an interesting quote about his hedge fund. I think it got terrible returns. So this isn't the example.

• 4:10:17Speaker 01

  Right, right. That's the, that's the sort of bear case, right? It's like too theoretical and yeah.

• 4:10:21Dwarkesh Patel?

  Sure, yeah. But they had an interesting quote that it's, um, it's, that it's like basically a think tank inside of a hedge fund.

• 4:10:27Speaker 01

  Yeah, that's what I'm trying to build.

• 4:10:29Dwarkesh Patel?

  Right.

• 4:10:30Speaker 01

  Yeah.

• 4:10:30Dwarkesh Patel?

  So presumably you've thought about the ways in which these kinds of things can blow. There's a very, there's a lot of interesting business history books about people who got the thesis right.

• 4:10:41Speaker 01

  Yeah.

• 4:10:41Dwarkesh Patel?

  But timed it wrong.

• 4:10:42Speaker 01

  Yeah.

• 4:10:43Dwarkesh Patel?

  Where they, they buy that internet's going to be a big deal.

• 4:10:47Speaker 01

  Yeah.

• 4:10:47Dwarkesh Patel?

  They sell at the wrong time and buy at the wrong time during the dot com boom.

• 4:10:50Speaker 01

  Yep.

• 4:10:51Dwarkesh Patel?

  And so they miss out on the gains, even though they're right about the.

• 4:10:53Speaker 01

  Yeah.

• 4:10:53Dwarkesh Patel?

  Anyways, yeah. Uh, what, what is that trick to preventing that kind of thing?

• 4:10:57Speaker 01

  Yeah. I mean, look, obviously you can't, you know, not blowing up is sort of like, you know, task number one and two or whatever. I mean, you know, I think this investment firm is going to just be betting on AGI, you know, betting on AGI and superintelligence before the decade is out. Taking that seriously, making the bets you would make.

• 4:11:12Dwarkesh Patel?

  Yeah.

• 4:11:12Speaker 01

  You know, if you took that seriously. So, you know, I think if that's wrong, you know, Firm is not going to do that well. The thing you have to be resistant to is like you have to be able to resist and get, you know, one or a couple or a few kind of individual calls, right? You know, it's like AI stagnates for a year because of the data wall. Or like, you know, you got, you got the call wrong on like when revenue would go up. And so anyway, that's pretty critical. You have to get timing right. I do think in general that the sort of sequence of bets on the way to AGI is actually pretty critical. And I think a thing people underrate. So, all right, I mean, yeah. So like, where does the story start, right? So like, obviously, the sort of only bet over the last year was Nvidia. And, you know, it's obvious now, very few people did it. This is sort of also, you know, a classic debate I and a friend had with another colleague of ours.

• 4:11:56Speaker 01

  Where this colleague was really into TSM, you know, TSMC. And he was just kind of like, well, you know, like these fabs are going to be so valuable. And also like Nvidia, there's just a lot of video synchronic risk, right? It's like maybe somebody else makes better GPUs. And that was basically right, but sort of only Nvidia had the AI beta, right? Because only Nvidia was kind of like a large fraction AI. The next few doublings would just like meaningfully explode their revenue. Whereas TSMC was, you know, a couple percent AI. So you, even though there's going to be a few doublings of AI, not going to make that big of an impact. All right, so it's sort of like the only place to find the AI beta basically was Nvidia for a while. You know, now it's broadening, right? So now TSM is like, you know, 20% AI by like 27 or something is what they're saying. One more doubling, it'll be kind of like a large fraction of what they're doing.

• 4:12:37Speaker 01

  And, you know, there's a whole, you know, whole stack, you know, there's like, you know, there's people making memory and COAS and, you know, power. You know, utility companies are starting to get excited about AI. And they're like, oh, it'll, you know, power production in the United States will grow, you know, not 2.5%, but 5% over the next five years. And I'm like, no, it'll grow more.

• 4:12:56Speaker 01

  You know, at some point, you know, you know, you know, like a Google or something becomes interesting. You know, people are excited about them with AI because it's like, oh, you know, AI revenue will be, you know, 10 billion or tens of billions. And I'm kind of like, ah, I don't really care about them before then. I care about it, you know, once it, you know, once you get the AI beta, right? And so at some point, you know, Google will get, you know, $100 billion of revenue from AI. Probably their stock will explode, you know, they're going to become, you know, 5 trillion, 10 trillion dollar company. Anyway, so the timing there is very important. You have to get the timing right. You have to get the sequence right. You know, at some point, actually, I think like, you know, there's going to be a real tailwind to equities from real interest rates, right? So basically in these sort of explosive growth worlds, you would expect real interest rates to go up a lot. Both on the sort of like, you know, basically both sides of the equation, right?

• 4:13:39Speaker 01

  On the supply side or on the sort of demand for money side, because, you know, People are going to be making these crazy investments, you know, initially in clusters and then in the robo factories or whatever, right? And so they're going to be borrowing like crazy. They want all this capital, higher ROI. And then on the sort of like consumer saving side, right, to like, you know, to, to, to give up all this capital, you know, the sort of like oil equation, standard sort of intratemporal transfer, you know, trade off of consumption.

• 4:14:05Dwarkesh Patel?

  So standard.

• 4:14:07Speaker 01

  Yeah.

• 4:14:08Dwarkesh Patel?

  Very standard.

• 4:14:09Speaker 01

  It's possible, you know, some of our friends have a paper on this, you know, basically, if you expect, you know, if consumers expect real growth rates to be higher, you know, interest rates are going to be higher because they're less willing to give up consumption, you know, consumption in the, there was, they're, they're less willing to give up consumption day for consumption in the future. Anyway, so at some point, real interest rates will go up if sort of ADA is greater than one, that actually means equities, you know, higher growth rate expectations mean equities go down because the sort of interest rate effect outweighs the growth rate effect. And so, you know, at some point, there's like the big bond short, you gotta get that right. You gotta get it right, that, you know, nationalization, you know, like, you got, you know, anyway. So there's this whole sequence of things, you gotta get that right.

• 4:14:44Dwarkesh Patel?

  And the unknown unknowns.

• 4:14:46Speaker 01

  Unknown unknowns, yeah. And so you've, look, you've got to be really, really careful about your like overall like risk positioning, right? And because, you know, you know, if you expect these kind of crazy events to play out, there's going to be crazy things you didn't foresee. You know, you do also want to make the sort of kind of bets that are tailored to your scenarios in the sense of like, you know, you want to find bets that are bets on the tails, right? You know, I don't think anyone is expecting, you know, interest rates to go above, you know, 10% like real interest rates, but you know, I think there's at least a serious chance of that, you know, before the decade is out. And so, you know, maybe there's some like cheap insurance you can buy on that, you know, that pays off.

• 4:15:18Dwarkesh Patel?

  Very silly question. In these worlds.

• 4:15:20Speaker 01

  Yeah.

• 4:15:21Dwarkesh Patel?

  Are financial markets where you make these kinds of bets going to be respected? And like, you know, like, is my fidelity account going to mean anything when we have the 50% economic growth? Like who's, who's like, we gotta respect his property rights.

• 4:15:34Speaker 01

  That's pretty deep into it. The bond short, the sort of 50, 50 second hour growth, that's pretty deep into it. I mean, again, there's this whole sequence of things, but yeah, no, I think property rates will be structured. Again, in the sort of modal world, the project, yeah. At some point, at some point, there's going to be figuring out the property rates for the galaxies, you know, that'll be interesting.

• 4:15:48Dwarkesh Patel?

  That will be interesting. So there's an interesting question about.

• 4:15:56Dwarkesh Patel?

  Going back to your strategy about, well, the 30s will really matter a lot about how the rest of the future goes.

• 4:16:02Speaker 01

  Yeah.

• 4:16:02Dwarkesh Patel?

  And you want to be in a position of influence by that point because of capital. It's worth considering, as far as I know, there's probably a whole bunch of literature on this, I'm just riffing, but the landed gentry during the, before the beginning of the industrial revolution, I'm not sure if they were able to leverage their position in a sort of Georgist or pickety type sense in order to accrue the returns that were realized through the industrial revolution.

• 4:16:36Speaker 01

  Yeah.

• 4:16:36Dwarkesh Patel?

  And I don't know what happened, at some point they just weren't the landed gentry. But I'd be concerned that even if you make great investment calls, you'll be like the guy who owned a lot of land, farmland before the industrial revolution. And like the guy who's actually going to make a bunch of money is the one that you mentioned. Even he doesn't make that much money, most of the benefits are widely diffused and so forth.

• 4:16:58Speaker 01

  I mean, I think the analog is like you sell your land, you put it all in sort of the, you know, the people who are building the new industry.

• 4:17:06Speaker 01

  I think the, I mean, I think the sort of like real depreciating asset, you know, for me is human capital, right? Yeah, no, look, I'm serious, right? It's like, you know, there's something about like, you know, I don't know, it was like valedictorian of Columbia, you know, the thing that made you special is you're smart.

• 4:17:19Dwarkesh Patel?

  Right.

• 4:17:20Speaker 01

  But actually, like, you know, that might not matter in like four years, you know, because it's actually automatable.

• 4:17:24Dwarkesh Patel?

  Right.

• 4:17:25Speaker 01

  And so anyway, a friend joked that the sort of investment firm was perfectly hedged for me. And it's like, you know, either like AGI this decade, and yeah, your human capital is depreciated, but you've turned that into financial capital. Or, you know, like no AGI this decade, in which case maybe the firm doesn't do that well, but you know, you're still in your twenties and you're still smart.

• 4:17:43Dwarkesh Patel?

  Excellent. And what's your story for why AGI hasn't been priced in? The story, financial markets are supposed to be very efficient, it's very, very hard to get an edge.

• 4:17:54Speaker 01

  Yeah.

• 4:17:56Dwarkesh Patel?

  Naively, you just say, well, I've looked at these scaling curves and they imply that we're going to be buying much more computed energy than the analysts realize.

• 4:18:05Speaker 01

  Yeah.

• 4:18:06Dwarkesh Patel?

  Shouldn't those analysts be broke by now? What's going on?

• 4:18:08Speaker 01

  Yeah, I mean, I used to be a true EMH guy. I was an economist, you know, I am. You know, I think the thing I, you know, changed my mind on is that I think there can be kind of groups of people, smart people, you know, who are, you know, stay there in San Francisco, who do just have off over the rest of society and kind of seeing the future. And so like COVID, right? Like, I think there's just honestly kind of similar group of people who just saw that and called it completely correctly. And, you know, they showed it the market, they did really well. You know, a bunch of other sort of things like that.

• 4:18:47Speaker 01

  So, you know, why is AGI not priced in? You know, it's sort of, you know, why, why hasn't the government nationalized the labs yet, right? It's like, you know, this, you know, society hasn't priced it in yet and sort of it hasn't completely diffused. And, you know, again, it might be wrong, right? But I just think sort of, you know, not that many people take these ideas seriously yet.

• 4:19:11Dwarkesh Patel?

  Yeah.

• 4:19:12Speaker 01

  Yeah.

• 4:19:13Dwarkesh Patel?

  Yeah. A couple of other sort of ideas that I was playing around with with regards to, we didn't get a chance to talk about, but the, the systems competition.

• 4:19:23Speaker 01

  Yeah.

• 4:19:24Dwarkesh Patel?

  There's a very interesting, one of my favorite books about World War II is Victor Davis Hansen's summary of everything. And he explains why the Allies made better decisions than the Axis.

• 4:19:39Speaker 01

  Why did they?

• 4:19:40Dwarkesh Patel?

  And so obviously, there were some decisions the Axis made that were pretty, like Blitzkrieg, whatever.

• 4:19:45Speaker 01

  Those were sort of by accident, though.

• 4:19:46Dwarkesh Patel?

  Well, in what sense? That they just had the infrastructure left over?

• 4:19:49Speaker 01

  Well, no, I mean, the sort of, I think, I mean, I don't, I mean, I think sort of my read of it is Blitzkrieg wasn't kind of some like a genius strategy. It was just kind of, it's like more like their hand was forced. I mean, this is sort of the very Adam Tuzian story of World War II, right? But it was, you know, there's sort of this long war versus short war. I think it's actually kind of an important concept. I think sort of Germany realized that if they were in a long war, including the United States, you know, they would not be able to compete industrially. So their only path to victory was like, make it a short war, right? And that- That sort of worked much more spectacularly than they thought, right? And sort of take over France and take over much of Europe. And so then, you know, the decision to invade the Soviet Union, it was, you know, it was, it was, um, look, if it was, it was about the Western Front in some sense, because it was like, we've got to get the resources.

• 4:20:33Dwarkesh Patel?

  Yeah.

• 4:20:33Speaker 01

  You know, we don't, we're actually, we don't actually have a bunch of the stuff we need, like, you know, oil and so on. You know, Auschwitz was actually just this giant chemical plant to make kind of like synthetic oil and a bunch of these things. It's the largest industrial project in Nazi Germany.

• 4:20:45Speaker 01

  And so, you know, and sort of they thought, well, you know, we completely crushed them in World War I, you know, it'll be easy, we'll invade them, we'll get the resources and then we can fight on the Western Front. And even during the sort of whole invasion of the Soviet Union, even though kind of like a large amount of the sort of, you know, the sort of deaths happened there, you know, like a large fraction of German industrial production was actually, you know, like planes and naval, you know, and so on. Those directed, you know, towards the Western Front and towards, you know, the Western Allies.

• 4:21:09Dwarkesh Patel?

  Well, and then so the point that Hanson was making was?

• 4:21:13Speaker 01

  By the way, I think this concept of like long war and short war is kind of interesting and with respect to thinking about the China competition. Which is like, you know, I worry a lot about kind of, you know, America, decline of sort of American, like late in American industrial capacity. You know, like, I think China builds like 200 times more ships than we do right now. You know, in some crazy way. And so it's like, maybe we have this superiority, say in the non-AI worlds, we have the superiority in military material, kind of like win a short war, at least, you know, kind of defend Taiwan in some sense. But like, if it actually goes on, you know, it's like, maybe China is much better able to mobilize, mobilize industrial resources. In a way that like, we just don't have that same ability anymore. I think this is also relevant to the AI thing in the sense of like, if it comes down to sort of a game about building, right?

• 4:21:58Speaker 01

  Including like, maybe AGI takes the trillion dollar cluster, not the hundred billion dollar cluster. Maybe, or even maybe AGI takes the, you know, is on the hundred billion dollar cluster. But, you know, it really matters if you can run, you know, 10x, you can, you can do one more order of magnitude of compute for your super intelligence or whatever.

• 4:22:13Speaker 01

  That, you know. Maybe right now they're behind, but they just have this sort of like raw latent industrial capacity to outbuild us.

• 4:22:19Dwarkesh Patel?

  Yeah.

• 4:22:19Speaker 01

  And that matters both in the run up to AGI and after, right? Where it's like, you have the superintelligence on your cluster. Now it's time to kind of like expand the explosive growth. And, you know, like, will we let the robo factories run wild? Like, maybe not, but like, maybe China will. We're like, you know, will we, well, yeah, will we produce the, how many, how many of the drones will we produce? And I think, yeah, so there's some sort of like outbuilding in the industrial explosion that I worry about.

• 4:22:40Dwarkesh Patel?

  You've got to be one of the few people in the world who is both concerned about alignment, but also wants to make sure that we'll let the robo factories proceed once we get the ASI to beat out China.

• 4:22:51Speaker 01

  Like it all is all a matter of. It's all, it's all part of the picture.

• 4:22:55Dwarkesh Patel?

  Yeah, yeah, yeah.

• 4:22:58Dwarkesh Patel?

  And by the way, speaking of the ASIs and the robot factories, one of the interesting things.

• 4:23:03Speaker 01

  Robo armies too.

• 4:23:03Dwarkesh Patel?

  Yeah. One of the interesting things, there's this question of what you do with industrial scale intelligence. And obviously it's not chatbots.

• 4:23:11Speaker 01

  Yeah.

• 4:23:12Dwarkesh Patel?

  But it's a, I think it's very hard to predict. Yeah, yeah.

• 4:23:17Dwarkesh Patel?

  But the, the, the history of oil is very interesting. Where in the, I think it's in the 1860s that we figure out how to refine oil, some geologist. And so Ben Standard Oil gets started, there's this huge boom. It changes American politics. Entire legislators are getting bought out by oil interests and presidents are getting elected based on the divisions about oil and breaking them up and everything. And all of this has happened.

• 4:23:46Speaker 01

  Yeah.

• 4:23:47Dwarkesh Patel?

  The world has never revolutionized before the car has been invented. And I, so when the light bulb was invented, I think it was like 50 years after oil refining had been discovered. It's majority of Standard Oil's history is before the cars invented.

• 4:24:03Speaker 01

  Kerosene lamps.

• 4:24:04Dwarkesh Patel?

  Exactly. So it's just used for lighting.

• 4:24:06Speaker 01

  Then they thought oil would just no longer be relevant.

• 4:24:08Dwarkesh Patel?

  Yeah, yeah. So there was a concern that Standard Oil would go bankrupt.

• 4:24:11Speaker 01

  Yeah.

• 4:24:13Dwarkesh Patel?

  When the label was amended.

• 4:24:14Speaker 01

  Yeah.

• 4:24:15Dwarkesh Patel?

  And but then there's sort of, you realize that there's immense amount of compressed energy here.

• 4:24:21Speaker 01

  Yeah.

• 4:24:22Dwarkesh Patel?

  You're gonna have billions of gallons of this stuff a year.

• 4:24:25Speaker 01

  Yeah.

• 4:24:25Dwarkesh Patel?

  And it's hard to sort of predict in advance what you can do with that.

• 4:24:28Speaker 01

  Yep, yep, that's right.

• 4:24:30Dwarkesh Patel?

  And then later on, it turns out, oh, transportation, cars with, that's what it'd be used for. Anyways, with intelligence, maybe one answer is the intelligence explosion.

• 4:24:41Speaker 01

  Right.

• 4:24:41Dwarkesh Patel?

  But even after that.

• 4:24:42Speaker 01

  Yeah.

• 4:24:42Dwarkesh Patel?

  So you have all these ASIs and you have enough compute, especially the compute they'll build.

• 4:24:46Speaker 01

  Yeah.

• 4:24:47Dwarkesh Patel?

  To run.

• 4:24:47Speaker 01

  Hundreds of millions of GPUs will hum.

• 4:24:49Dwarkesh Patel?

  Yeah, but what are we doing with that?

• 4:24:51Speaker 01

  Yeah.

• 4:24:51Dwarkesh Patel?

  And it's very hard to predict in advance. And I think it'll be very interesting to figure out what the Jupiter brains will be doing.

• 4:24:59Dwarkesh Patel?

  So, look, there's situational awareness.

• 4:25:01Speaker 01

  Yeah.

• 4:25:02Dwarkesh Patel?

  Of where things stand now. And we've gotten a good dose of that. The, obviously, a lot of the things we're talking about now, you couldn't have prejudged many years back in the past.

• 4:25:15Speaker 01

  Right.

• 4:25:16Dwarkesh Patel?

  And part of your worldview implies that things will accelerate.

• 4:25:19Speaker 01

  Yeah.

• 4:25:20Dwarkesh Patel?

  Because of AI getting to the process.

• 4:25:22Speaker 01

  Yeah.

• 4:25:23Dwarkesh Patel?

  But many other things that we are, that are unpredictable fundamentally.

• 4:25:26Speaker 01

  Yep.

• 4:25:26Dwarkesh Patel?

  Basically, how people will react, how the political system will react, how foreign adversaries will react.

• 4:25:30Speaker 01

  Yep.

• 4:25:31Dwarkesh Patel?

  That, those things will become evident over time.

• 4:25:34Speaker 01

  Yep.

• 4:25:34Dwarkesh Patel?

  So the situational awareness is not just knowing where the picture stands now, but being in a position to react appropriately to new information, to change your worldview as a result, to change your recommendations as a result.

• 4:25:48Speaker 01

  Yep.

• 4:25:49Dwarkesh Patel?

  What is the appropriate way to think about situational awareness as a continuous process rather than as a one-time thing you realized?

• 4:25:58Speaker 01

  Yep. No, I think this is great. Look, I think there's, there's a sort of mental flexibility and willing to change your mind. That's really important. I actually think this is sort of like how a lot of brains have been broken in the AGI debate, right? The doomers who actually, you know, I think were really prescient on AGI and thinking about the stuff, you know, like a decade ago. But, you know, they haven't actually updated on the empirical realities of deep learning. They're sort of like, the proposals are really kind of naive and unworkable. Doesn't really make sense. You know, there's people who come in with sort of a predefined ideology. There's kind of like, you know, the EAX a little bit, you know, like, they like to shitpost about technology, but they're not actually thinking through it. Like, you know, I mean, either the sort of stagnationists who think this stuff is only going to be, you know, a chatbot, and so of course it isn't risky, or they're just not thinking through the kind of like actually immense national security implications and how that's going to go. And you know, I actually think there's kind of a risk and kind of like having written

• 4:26:43Speaker 01

  the stuff down and like put it online and you know, there's a, there's, I think this sometimes happens to people is a sort of calcification of their worldview because now they've publicly articulated this position and you know, maybe there's some evidence against it, but they're clinging to it. And so I actually, you know, I want to give the big disclaimer on like, you know, I think it's really valuable to paint a sort of very concrete and visceral picture. I think this is currently my best guess on how this decade will go. I think if it goes anywhere like this, it will be wild. But you know, given, given the rapid pace of progress, we're going to keep getting a lot more information. And you know, I think it's important to sort of keep your head on straight about that.

• 4:27:24Speaker 01

  You know, I feel like the most important thing here is that, you know, and this relates to some of the stuff we've talked about and you know, sort of the world being surprisingly small and so on. You know, I feel like I used to have this worldview of like, look, there's important things happening in the world, but there's like people who are taking care of it, you know, and there's like the people in government and there's again, even like AI labs have idealized and people are on it, you know, surely they must be on it, right? And I think just some of this personal experience, even seeing how kind of COVID went, you know, people aren't necessar-- there's not some, not that somebody else is just kind of on it and making sure this goes well, however it goes. You know, the thing that I think will really matter is that there are sort of good people who take this stuff

• 4:28:09Speaker 01

  as seriously as it deserves and who are willing to kind of take the implication seriously, who are willing to, you know, who have situational awareness, are willing to change their minds, are willing to sort of stare the picture in the face. And you know, I'm counting on those good people.

• 4:28:26Dwarkesh Patel?

  All right. That's a great place to close Leopold.

• 4:28:29Speaker 01

  Thanks so much, Dark Ash.

• 4:28:30Dwarkesh Patel?

  Yeah, this is excellent. Hey, everybody. I hope you enjoyed that episode with Leopold. There's actually one more riff about German history that he had after a break and it was pretty interesting because I didn't want to cut it out. So I've just included it after this outro. You can advertise on the show now. So if you're interested, you can reach out at the form in the description below. Other than that, the most helpful thing you can do is just share the episode if you enjoyed it. Send it to group chats, Twitter, wherever else you think people who might like this episode might congregate. And other than that, I guess here's this riff on Frederick the Great. See you on the next one.

• 4:29:09Speaker 01

  I mean, I think the actual funny thing is, you know, a lot of this sort of German history stuff we've talked about is sort of like, not actually stuff I learned in Germany. It's sort of like stuff that I learned after. And there's actually, you know, funny thing where I kind of would go back to Germany over Christmas or whatever. And tell people. Suddenly understand the street names. You know, it's like, you know, Gneisnow and Schornhorst, and then all these like Prussian military reformers. And you're like, finally understood, you know, Sansa Sea. And you're like, Frederick, you know, Frederick the Great is this really interesting figure, where he's this sort of, in some sense, kind of like, gay lover of arts, right? Where he, he, you know, he hates speaking German. He only wants to speak French. You know, he like plays the flute, he composes, he has

• 4:29:49Speaker 01

  all this sort of great, you know, artists of his day, you know, over at Sansa Sea. And he actually had this sort of like really tough upbringing, where his father was this sort of like really stern sort of Prussian military man. And he had had a Frederick the Great as a child, as sort of a 17-year-old or whatever. He basically had a male lover. And what his father did was imprison his son, and then, I think, hang his male lover in front of him. And again, his father was this kind of very stern Prussian guy. He was this kind of gay, you know, lover of arts. But then later on, Frederick the Great turns out to be this like, you know, one of the most, kind of like, you know, successful

• 4:30:33Speaker 01

  kind of Prussian conquerors, right? Like, he gets Silesia, he wins the Seven Years' War. You know, also, you know, amazing military strategists, you know, amazing military strategy at the time consisted of like, he was able to like flank the army, and that was crazy, you know, and that was brilliant. And then, and then they like almost lose the Seven Years' War. And at the very end, you know, the sort of, the Russian Tsar changes, and he's like, "Ah, I'm actually kind of a Prussian stan." You know, "I think I'm like, I'm into this stuff." And then he lets, you know, lets Frederick the Great loose, and he had, "Let's, let's, let's let the army be okay." And anyway, sort of like, yeah, kind of bizarre, interesting figure in German history.