Ilya Sutskever — We're moving from the age of scaling to the age of research

Ilya & I discuss SSI’s strategy, the problems with pre-training, how to improve the generalization of AI models, and how to ensure AGI goes well. Watch on YouTube https://youtu.be/aR20FWCCjAs; read the transcript https://www.dwarkesh.com/p/ilya-sutskever-2. Sponsors * Gemini 3 https://gemini.google is the first model I’ve used that can find connections I haven’t anticipated. I recently wrote a blog post on RL’s information efficiency, and Gemini 3 helped me think it all through. It also generated the relevant charts and ran toy ML experiments for me with zero bugs. Try Gemini 3 today at gemini.google https://gemini.google * Labelbox https://labelbox.com/dwarkesh helped me create a tool to transcribe our episodes! I’ve struggled with transcription in the past because I don’t just want verbatim transcripts, I want transcripts reworded to read like essays. Labelbox helped me generate the exact data I needed for this. If you want to learn how Labelbox can help you (or if you want to try out the transcriber tool yourself), go to labelbox.com/dwarkesh https://labelbox.com/dwarkesh * Sardine https://sardine.ai/dwarkesh is an AI risk management platform that brings together thousands of device, behavior, and identity signals to help you assess a user’s risk of fraud & abuse. Sardine also offers a suite of agents to automate investigations so that as fraudsters use AI to scale their attacks, you can use AI to scale your defenses. Learn more at sardine.ai/dwarkesh https://sardine.ai/dwarkesh To sponsor a future episode, visit dwarkesh.com/advertise https://www.dwarkesh.com/advertise. Timestamps (00:00:00) – Explaining model jaggedness (00:09:39) - Emotions and value functions (00:18:49) – What are we scaling? (00:25:13) – Why humans generalize better than models (00:35:45) – SSI’s plan to straight-shot superintelligence (00:46:47) – SSI’s model will learn from deployment (00:55:07) – How to think about powerful AGIs (01:18:13) – “We are squarely an age of research company” (01:20:23) – Self-play and multi-agent (01:32:42) – Research taste 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.

• 0:00Speaker 00

  You know what's crazy? That all of this is real.

• 0:04Dwarkesh Patel?

  Yeah, meaning what?

• 0:04Speaker 00

  Don't you think so?

• 0:06Dwarkesh Patel?

  Meaning what?

• 0:06Speaker 00

  Like all this AI stuff and all this Bay Area, yeah, that it's hap- like, isn't it straight out of science fiction?

• 0:13Dwarkesh Patel?

  Yeah. Another thing that's crazy is like how normal the slow takeoff feels. The idea that we'd be investing 1% of GDP in AI, like I feel like it would have felt like a bigger deal, you know? Where right now it just feels like.

• 0:27Speaker 00

  We get used to things pretty fast, turns out, yeah. But also it's kind of like it's abstract, like, what does it mean? What it means that you see it in the news.

• 0:35Dwarkesh Patel?

  Yeah.

• 0:36Speaker 00

  That such and such company announced such and such dollar amount.

• 0:38Dwarkesh Patel?

  Right.

• 0:40Speaker 00

  That's, that's all you see.

• 0:41Dwarkesh Patel?

  Right.

• 0:42Speaker 00

  It's not really felt in any other way so far.

• 0:45Dwarkesh Patel?

  Now, should we actually begin here? I think this is an interesting discussion.

• 0:47Speaker 00

  Sure.

• 0:48Dwarkesh Patel?

  I think your point about, well, from the average person's point of view, nothing is that different will continue being true even into the singularity.

• 0:57Speaker 00

  No, I don't think so.

• 0:58Dwarkesh Patel?

  Okay, interesting.

• 0:59Speaker 00

  So the thing which I was referring to, not feeling different,

• 1:05Speaker 00

  is, okay, so such and such company announced some difficult to comprehend dollar amount of investment.

• 1:12Dwarkesh Patel?

  Right.

• 1:12Speaker 00

  I don't think anyone knows what to do with that.

• 1:14Dwarkesh Patel?

  Yeah.

• 1:15Speaker 00

  But I think that the impact of AI is gonna be felt. AI is going to be diffused through the economy. There are very strong economic forces for this. And I think the impact is going to be felt very strongly.

• 1:30Dwarkesh Patel?

  When do you expect that impact? I think the models seem smarter than their economic impact would imply.

• 1:38Speaker 00

  Yeah, this is one of the very confusing things about the models right now. How to reconcile the fact that they are doing so well on evals. And you look at the evals and you go, those are pretty hard evals.

• 1:56Dwarkesh Patel?

  Right.

• 1:57Speaker 00

  They are doing so well. But the economic impact seems to be dramatically behind. And it's almost like.

• 2:07Speaker 00

  It's it's very difficult to make sense of how can the model on the one hand do these amazing things? And then on the other hand, like, repeat itself twice in some situation in a kind of a An example would be, let's say you use vibe coding to do something. And you go to some place and then you get a bug. And then you tell the model, can you please fix the bug?

• 2:29Dwarkesh Patel?

  Yeah.

• 2:30Speaker 00

  And the model says, oh, my God, you're so right. I have a bug. Let me go fix that. And it introduces a second bug.

• 2:35Dwarkesh Patel?

  Yeah.

• 2:36Speaker 00

  And then you tell it, you have this new, the second bug.

• 2:39Dwarkesh Patel?

  Right.

• 2:39Speaker 00

  And it tells you, oh, my God, how could I have done it? You're so right again. And brings back the first bug.

• 2:44Dwarkesh Patel?

  Yeah.

• 2:44Speaker 00

  And you can alternate between those.

• 2:45Dwarkesh Patel?

  Yeah.

• 2:46Speaker 00

  And it's like, how is that possible?

• 2:48Dwarkesh Patel?

  Yeah.

• 2:48Speaker 00

  It's like, I'm not sure. But it does suggest that the Something strange is going on. I have two possible explanations. So here, this is the more kind of a Whimsical explanation is that maybe RL training makes the models a little bit too single minded and narrowly focused. A little bit too. I don't know, unaware.

• 3:14Speaker 00

  Even though it also makes them aware in some other ways. And because of this, they can't do Basic things, but there is another explanation, which is Back when people were doing pre-training. The question of what data to train on was answered. Because that that answer was everything.

• 3:35Dwarkesh Patel?

  Yeah.

• 3:36Speaker 00

  When you do pre-training, you need all the data.

• 3:41Speaker 00

  So you don't have to think it's going to be this data or that data.

• 3:44Dwarkesh Patel?

  Yeah.

• 3:44Speaker 00

  But when people do RL training, they do need to think. They say, OK, we want to have this kind of RL training for this thing and that kind of RL training for that thing. And from what I hear, All the companies have teams that just produce new RL environments and just add it to the training mix. And the question is, well, what are those? There are so many degrees of freedom. There is such a huge variety of RL environments you could produce. And one of the One thing you could do, and I think that's something that is done inadvertently, is that people take inspiration from the evals. You say, hey, I would love our model to do really well when we release it. I want the evals to look great.

• 4:28Speaker 00

  What would be RL training that could help on this task, right? I think that is something that happens, and I think it could explain a lot of what's going on. If you combine this with generalization of the models actually being inadequate, that has the potential to explain a lot of what we are seeing. This disconnect between eval performance and actual real world performance, which is something that we don't today exactly even understand what we mean by that.

• 5:00Dwarkesh Patel?

  I like this idea that the real reward hacking is the human researchers who are too focused on the evals. I think there's two ways to understand or to try to think about what you have just pointed out. One is, look, if it's the case that simply by becoming superhuman at a coding competition, a model will not automatically become more tasteful and exercise better judgment about how to improve your codebase. Well, then you should expand the suite of environments such that you're not just testing it on having the best performance in coding competition. It should also be able to make the best kind of application for X thing or Y thing or Z thing. And another, maybe this is what you're hinting at, is to say,

• 5:45Dwarkesh Patel?

  why should it be the case in the first place that becoming superhuman at coding competitions doesn't make you a more tasteful programmer more generally? Maybe the thing to do is not to keep stacking up the amount of environments and the diversity of environments to figure out an approach which lets you learn from one environment and improve your performance on something else.

• 6:07Speaker 00

  So I have a human analogy which might be helpful. So even the case, let's take the case of competitive programming, since you mentioned that. And suppose you have two students. One of them, work decided they want to be the best competitive programmer, so they will practice 10,000 hours for that domain. They will solve all the problems, memorize all the proof techniques and be very, very, you know...

• 6:35Speaker 00

  Be very skilled at quickly and correctly implementing all the algorithms. And by doing so, they became the best, one of the best. Student number two thought, oh, competitive programming is cool. Maybe they practiced for a hundred hours.

• 6:49Dwarkesh Patel?

  Yeah.

• 6:50Speaker 00

  Much, much less, and they also did really well. Which one do you think is going to do better in their career later on?

• 6:56Dwarkesh Patel?

  The second.

• 6:56Speaker 00

  Right? And I think that's basically what's going on. The models are much more like the first student, but even more, because then we say, okay. So the model should be good competitive programming. So let's get every single competitive programming problem ever. And then let's do some data augmentation. So we have even more competitive programming problems.

• 7:14Dwarkesh Patel?

  Yes.

• 7:14Speaker 00

  And we train on that. And so now you've got this great competitive programmer. And with this analogy, I think it's more intuitive. I think it's more intuitive with this analogy that, yeah, okay. So if it's so well trained, okay, it's like all the different algorithms and all the different proof techniques are like right at its fingertips. And it's more intuitive that with this level of preparation, it will not necessarily generalize to other things.

• 7:38Dwarkesh Patel?

  Hmm. But then what is the analogy for what the second student is doing before they do the hundred hours of fine tuning?

• 7:47Speaker 00

  I think it's like they have it. I think it's the it factor.

• 7:53Dwarkesh Patel?

  Yeah.

• 7:54Speaker 00

  Right? And like, I know, like when I was an undergrad, I remember there was, there was a student like this that studied with me. So I know, I know it exists.

• 8:01Dwarkesh Patel?

  Yeah. I think it's interesting to distinguish it from whatever pre-training does. So one way to understand what you just said about we don't have to choose the data in pre-training is to say, actually, it's not dissimilar to the 10,000 hours of practice. It's just that you get that 10,000 hours of practice for free because it's already somewhere in the pre-training distribution. But it's like, maybe you're suggesting actually, there's actually not that much generalization from pre-training. There's just so much data in pre-training. Everybody's like, it's not necessarily generalizing better than RL.

• 8:31Speaker 00

  Like the main, the main strength of pre-training is that there is A, so much of it.

• 8:35Dwarkesh Patel?

  Yeah.

• 8:36Speaker 00

  And B, you don't have to think hard about what data to put into pre-training. And it's a very kind of natural data and it does include in it a lot of what people do.

• 8:48Dwarkesh Patel?

  Yeah.

• 8:50Speaker 00

  People's thoughts and a lot of the features of, you know, it's like the whole world as projected by people onto text.

• 8:58Dwarkesh Patel?

  Yeah.

• 9:00Speaker 00

  And pre-training tries to capture that using a huge amount of data. It's, it's very, the pre-training is very difficult to reason about because it's so hard to understand the manner in which the model relies on pre-training data. And whenever the model makes a mistake, could it be because something by chance is not as supported by the pre-training data? You know, and support by pre-training is maybe a loose term.

• 9:30Speaker 00

  I, I don't know if I can add anything more useful on this, but

• 9:35Speaker 00

  I don't think there is a human analog to pre-training.

• 9:38Dwarkesh Patel?

  Hmm. Here's analogies that people have proposed for what the human analogy to pre-training is, and I'm curious to get your thoughts on why they're potentially wrong. One is to think about the first eighteen or fifteen or thirteen years of a person's life, when they aren't necessarily economically productive, but they are doing something that is making them understand the world better and so forth. And the other is to think about evolution as doing some kind of search for three billion years, which then results in a human lifetime instance. And then I'm, I'm curious if you think either of these are actually analogous to pre-training, or how, how would you think about at least what lifetime human learning is like if not pre-training?

• 10:22Speaker 00

  I think there are some similarities between both of these to pre-training.

• 10:27Dwarkesh Patel?

  Yeah.

• 10:27Speaker 00

  And pre-training tries to play the role of both of these. But I think there are some big differences as well. The amount of pre-training data is very, very staggering.

• 10:39Dwarkesh Patel?

  Yes.

• 10:41Speaker 00

  And somehow, a human being, after even fifteen years with a tiny fraction of that pre-training data, they know much less. Yeah. But whatever they do know, they know much more deeply, somehow. And the mistakes, like, like already at that age, you would not make mistakes that RAIs make. Yeah. There is another thing, you might say, could it be something like evolution? And the answer is maybe, but in this case, I think evolution might actually have an edge. Like, there is this, I remember reading about this case where some, you know, that's one thing that neuroscientists do, or rather, one way in which neuroscientists can learn about the brain, is by studying people with brain damage to different parts of the brain.

• 11:25Speaker 00

  And, and so, and some people have the most strange symptoms you could imagine. It's actually really, really interesting. And there was one case that comes to mind, it's relevant. I read about this person who had some kind of brain damage that took out, I think, a stroke or an accident, that took out his emotional processing. So he stopped feeling any emotion. And as a result of that, you know, he still remained very articulate and he could solve little puzzles and on tests, he seemed to be just fine. But he felt no emotion, he didn't feel sad, he didn't feel angry, he didn't feel animated.

• 12:06Speaker 00

  And he became somehow extremely bad at making any decisions at all. It would take him hours to decide on which socks to wear, and he would make very bad financial decisions. And that's very, but does, what, what does it say about the role of our built-in emotions in making us like a viable agent, essentially? And I guess to connect to your question about pre-training, it's like, maybe pre- like, maybe if you're good enough at, like, getting everything out of pre-training, you can get, you, you could get that as well, but that's the kind of thing which seems...

• 12:50Speaker 00

  Well, it may or may not be possible to get that from pre-training.

• 12:55Dwarkesh Patel?

  What is that? Clearly not just directly emotion. It seems like some almost value function like thing, which is giving, telling you which decision to be made, like what the end reward for any decision should be. And you think that doesn't sort of implicitly come from?

• 13:15Speaker 00

  I think it could. I'm just saying it's not one, it's not 100% obvious.

• 13:19Dwarkesh Patel?

  Yeah. But what is that? Like, what, how do you think about emotions? And what is the ML analogy for emotions?

• 13:26Speaker 00

  It should be some kind of a value function thing.

• 13:28Dwarkesh Patel?

  Yeah.

• 13:29Speaker 00

  But I don't think there is a great ML analogy because right now value functions don't play a very prominent role in the things people do.

• 13:36Dwarkesh Patel?

  It might be worth defining for the audience what a value function is, if you want to do that.

• 13:39Speaker 00

  I mean, certainly, I'll be very happy to do that. Right? So.

• 13:49Speaker 00

  So when people do reinforcement learning, the way reinforcement learning is done right now, how does it, how do people train those agents? So you have a neural net and you give it a problem. And then you tell the model, go solve it. And the model takes maybe thousands, hundreds of thousands of actions or thoughts or something, and then it produces a solution. The solution is graded. And then the score is used to provide a training signal for every single action

• 14:18Speaker 00

  in your trajectory.

• 14:20Speaker 00

  So that means that if you are doing something that goes for a long time, if you're training a task that takes a long time to solve, you will do no learning at all until you solve, until you come up with a proposed solution. That's how reinforcement learning is done naively. That's how O1, R1 ostensibly are done.

• 14:40Speaker 00

  The value function says something like, okay, look, maybe I could sometimes, not always, could tell you if you are doing well or badly. The notion of a value function is more useful in some domains than others. So for example, when you play chess and you lose a piece, you know, I messed up. You don't need to play the whole game to know that what I just did was bad and therefore whatever whatever preceded it was also bad. So the value function lets you short circuit the wait until the very end. Like, let's suppose that you started to pursue some kind of, um, okay, let's suppose that you are doing some kind of a math thing or a programming thing. And you're trying to explore a particular solution direction.

• 15:26Speaker 00

  And after, let's say after a thousand steps of thinking, you concluded that this direction is unpromising. As soon as you conclude this, you could already get a reward signal a thousand time steps previously. When you decided to pursue down this path, you say, oh, next time I shouldn't pursue this path in a similar situation. Long before you actually came up with a proposed solution.

• 15:51Dwarkesh Patel?

  Hmm. This was in the Deep Sea R1 paper is that the space of trajectories is so wide that maybe it's hard to learn a mapping from an intermediate trajectory and value. And also given that, you know, in coding, for example, you will have the wrong idea, then you'll go back, then you'll change something.

• 16:12Speaker 00

  This sounds like such lack of faith in deep learning.

• 16:15Dwarkesh Patel?

  Yeah.

• 16:15Speaker 00

  Like, I mean, sure, it might be difficult, but nothing deep learning can do.

• 16:21Dwarkesh Patel?

  Yeah.

• 16:22Speaker 00

  So my expectation is that like value function should be useful and and I fully, I fully expect that they will be used in the future, if not already. What was I alluding to with the person whose emotional center got damaged is more that.

• 16:45Speaker 00

  Maybe what it suggests is that the value function of humans is modulated by emotions in some important way that's hardcoreed by evolution. And maybe that is important for people to be effective in the world.

• 16:59Dwarkesh Patel?

  That's the thing I was actually planning on asking you. There's something really interesting about emotions as a value function, which is that it's impressive that they have this much utility while still being rather simple to understand.

• 17:15Speaker 00

  So I have two responses. I do agree that compared to the kind of things that we learn and the things we are talking about, the kind of as we are talking about, emotions are relatively simple. They might even be so simple that maybe you could map them out in a human understandable way. I think it would be cool to do. In terms of utility, though, I think there is a thing where, you know, there is this complexity, robustness, trade-off. Where complex things can be very useful, but simple things are very useful in a very broad range of situations.

• 17:59Speaker 00

  And so I think what we've got one way to interpret what we are seeing is that we've got these emotions that essentially evolved mostly from our mammal ancestors and then fine-tuned a little bit while we were hominids, just a bit. We do have like a decent amount of social emotions, though, which mammals may lack. But they're not very sophisticated. And because they're not sophisticated, they serve us so well in this very different world compared to the one that we've been living in. Actually, they also make mistakes. For example, our emotions, well, I don't know, does hunger count as an emotion? It's debatable, but I think, for example, our intuitive feeling of hunger

• 18:42Speaker 00

  is not succeeding in guiding us correctly in this world with an abundance of food.

• 18:48Dwarkesh Patel?

  Yeah. People have been talking about scaling data, scaling parameters, scaling compute. Is there a more general way to think about scaling? What are the other scaling axes?

• 19:03Speaker 00

  The thing, so here is a perspective. Here's a perspective I think might be, might be true.

• 19:10Speaker 00

  So the way a mail used to work is that people would just tinker with stuff and try to,

• 19:20Speaker 00

  and try to get interesting results. That's what's been going on in the past.

• 19:25Speaker 00

  Then the scaling insight arrived, right? Scaling laws, GPT three. And suddenly everyone realized we should scale. And it's just this, this is an example of how language affects thought. Scaling is just one word, but it's such a powerful word because it informs people what to do. They say, okay, let's, let's try to scale things. And so you say, okay, so what are we scaling? And pre-training was a thing to scale. It was a particular scaling recipe.

• 20:01Dwarkesh Patel?

  Yes.

• 20:02Speaker 00

  The big breakthrough of pre-training is the realization that this recipe is good. So you say, hey, if you mix some compute with some data into a neural net of a certain size, you will get results and you will know that it will be better if you just scale the recipe up. And this is also great. Companies love this because it gives you a very low risk way of investing your resources.

• 20:31Dwarkesh Patel?

  Yeah.

• 20:32Speaker 00

  Right? It's much harder to invest your resources in research. Compare that, you know, if you research, you need to have like go for researchers and research and come up with something versus get more data, get more compute, you know, you'll get something from pre-training. And indeed, you know, it looks like I, based on various. Things people say on, some people say on Twitter, maybe it appears that Gemini have found a way to get more out of pre-training. At some point, though, pre-training will run out of data. The data is very clearly finite. And so then, okay, what do you do next? Either you do some kind of a souped-up pre-training, different recipe from the one you've done before, or you're doing RL, or maybe something else.

• 21:15Speaker 00

  But now that compute is big, computers now very big, in some sense, we are back to the age of research. So maybe here's another way to put it. Up until 2020, from 2015, from 2012 to 2020, it was the age of research. Now, from 2020 to 2025, it was the age of scaling, or maybe plus minus, let's add error bars to those years, because people say, this is amazing, you've got to scale more, keep scaling, the one word, scaling. But now the scale is so big, like, is the belief really that, oh, it's so big, but if you had 100x more, everything would be so different? Like, it would be different, for sure, but like, is the belief that if you just 100x

• 21:58Speaker 00

  the scale, everything would be transformed? I don't think that's true. So it's back to the age of research again, just with big computers.

• 22:06Dwarkesh Patel?

  That's a very interesting way to put it. But let me ask you the question you just posed then. What are we scaling? And what would it mean to have a recipe? Because I guess I'm not aware of a very clean relationship that almost looks like a law of physics, which existed in pre-training, was a power law between data or computer parameters and loss. What is the kind of relationship we should be seeking? And how should we think about what this new recipe might look like?

• 22:38Speaker 00

  So we've already witnessed a transition from one type of scaling to a different type of scaling, from pre-training to RL.

• 22:51Speaker 00

  Now people are scaling RL. Now, based on what people say on Twitter, they spend more compute on RL than on pre-training at this point. Because RL can actually consume quite a bit of compute.

• 23:03Dwarkesh Patel?

  Yeah.

• 23:03Speaker 00

  You know, you do very, very long rollouts.

• 23:06Dwarkesh Patel?

  Yes.

• 23:07Speaker 00

  So it takes a lot of compute to produce those rollouts. And then you get relatively small amount of learning per rollout. So you really can't spend, you really can't spend a lot of compute. And I could imagine, like, I wouldn't, at this, at this, it's more like, I wouldn't even call it a scale, scaling. I would say, hey, like, what are you doing? And is the thing you are doing the, the, the, the most productive thing you could be doing?

• 23:31Dwarkesh Patel?

  Yeah.

• 23:31Speaker 00

  Can you find a most more productive way of using your compute? We've discussed the value function business earlier. And maybe, once people get good at value functions, they will be using their, their resources more productively.

• 23:47Speaker 00

  And if you find a whole other way of training models, you could say, is this scaling? Or is it just using your resources? I think it becomes a little bit ambiguous. In a sense that when people were in the age of research, back then, it was like, people say, hey, let's try this and this and this. Let's try that and that and that. Oh, look, something interesting is happening. And I think there will be a return to that.

• 24:10Dwarkesh Patel?

  So if we're back in the era of research, stepping back, what is the part of the recipe that we need to think most about? When you say value function, people are already trying the current recipe, but then having a little M as a judge and so forth. You could say that's a value function, but it sounds like you have something much more fundamental in mind. Do we need, do we need to go back to, should we even rethink pre-training at all and not just add more steps to the end of that process?

• 24:35Speaker 00

  Yeah. So the, the, the, the discussion about value function, I think it was interesting. I want to like emphasize that I think the value function is something like, it's going to make RL more efficient.

• 24:50Speaker 00

  And I think that makes a difference. But I think that anything you can do with a value function, you can do without, just more slowly.

• 24:57Dwarkesh Patel?

  Hmm.

• 24:59Speaker 00

  The thing which I think is the most fundamental is that these models somehow just generalize dramatically worse than people.

• 25:06Dwarkesh Patel?

  Yes.

• 25:07Speaker 00

  And it's super obvious. That's that seems like a very fundamental thing.

• 25:12Dwarkesh Patel?

  Okay, so this is the crux, generalization. And there's two sub-questions. There's one, which is about sample efficiency, which is, why should it take so much more data for these models to learn than humans? There's a second about, even separate from the amount of data it takes, there's a question of, why is it so hard to teach the thing we want to a model than to a human? Which is to say, for to a human, we don't necessarily need a verifiable reward to be able to, you're probably mentoring a bunch of researchers right now, and you're, you know, talking with them, you're showing them your code, and you're showing them how you think, and from that, they're picking up your way of thinking and how they should do research. You don't have to set like a verifiable reward for them.

• 25:57Dwarkesh Patel?

  That's like, okay, this is the next part of their curriculum, and now this is the next part of your curriculum, and, oh, it was, this training was unstable, and we gotta, there's not this schleppy bespoke process. So perhaps these two issues are actually related in some way. But I'd be curious to explore this, this second thing, which was more like continual learning, and this first thing, which feels just like sample efficiency.

• 26:18Speaker 00

  Yeah, so you know, you could actually wonder, one, one possible explanation for the human sample efficiency that needs to be considered is evolution. And evolution has given us a small amount of the most useful information possible. And for things like vision, hearing, and locomotion, I think there's a pretty strong case that evolution actually has given us a lot. So, for example, human dexterity far exceeds, I mean...

• 26:55Speaker 00

  Robots can become dexterous too if you subject them to like a huge amount of training and simulation. But to train a robot in the real world to quickly like pick up a new skill like a person does seems very out of reach.

• 27:06Dwarkesh Patel?

  Yeah.

• 27:06Speaker 00

  And here you could say, oh yeah, like locomotion. All our ancestors needed great locomotion squirrels like. So locomotion maybe like we've got like some unbelievable prior.

• 27:19Dwarkesh Patel?

  Yeah.

• 27:19Speaker 00

  You could make the same case for vision, you know, I believe Jan Le Can made the point, oh, like um children learn to drive after sixteen hour, after ten hours of practice, which is true. But our vision is so good. At least for me, when I remember myself being five year old, my, I was, I was very excited about cars back then. And I'm pretty sure my car recognition was more than adequate for self-driving already as a five-year-old. You don't get to see that much data as a five-year-old. You spend most of your time in your parents' house. So you have very low data diversity.

• 27:52Dwarkesh Patel?

  Yeah.

• 27:53Speaker 00

  But you could say maybe that's evolution too. But in language and math and coding, probably not.

• 28:00Dwarkesh Patel?

  It still seems better than models. I mean, obviously, models are better than the average human at language and math encoding. But are they better at the average human at learning?

• 28:09Speaker 00

  Oh, yeah. Oh, yeah, absolutely. What I meant to say is that language math encoding and especially math encoding suggests that whatever it is that makes people good at learning is probably not so much a complicated prior, but something more, some fundamental thing.

• 28:29Dwarkesh Patel?

  Wait, I'm not sure understood. Why should that be the case?

• 28:32Speaker 00

  So consider a skill that people exhibit some kind of great reliability or, you know, um.

• 28:39Dwarkesh Patel?

  Yeah.

• 28:41Speaker 00

  If the skill is one that was very useful to our ancestors for many millions of years, hundreds of millions of years. You could say, you could argue that maybe humans are good at it because of evolution, because we have a prior. An evolutionary prior that's encoded in some very non-obvious way.

• 29:03Dwarkesh Patel?

  Yeah.

• 29:04Speaker 00

  That somehow makes us so good at it.

• 29:06Dwarkesh Patel?

  Yeah.

• 29:07Speaker 00

  But if people exhibit great ability, reliability, robustness, ability to learn in a domain that really did not exist until recently. Then this is more an indication that people might have

• 29:26Speaker 00

  just better machine learning period.

• 29:28Dwarkesh Patel?

  Hmm. But then how should we think about what that is? Is it a matter of?

• 29:34Dwarkesh Patel?

  Yeah, what is the ML analogy for?

• 29:38Dwarkesh Patel?

  There's a couple of interesting things about it. It takes fewer samples. It's more unsupervised. You don't have to set a very like a child learning to drive a car. A child, children are not learning to drive a car. A teenager learning how to drive a car is like not exactly getting some pre-built verifiable reward there. It comes from their interaction with the machine and with the environment.

• 30:02Dwarkesh Patel?

  And yeah, it takes much fewer samples. It seems more unsupervised. It seems more robust.

• 30:07Speaker 00

  Much more robust. The robustness of people is really staggering.

• 30:12Dwarkesh Patel?

  Yeah, so it's like, OK, and do you have a unified way of thinking about? Why are all these things happening at once? What is the ML analogy that would? That could be, it could realize something like this?

• 30:23Speaker 00

  So, so, so this is where, you know, one of the things that you've been asking about is how can, you know, the teenage driver kind of self-correct and learn from their experience without an external teacher? And the answer is, well, they have their value function. Right? They have a general sense, which is also, by the way, extremely robust in people. Like, whatever it is, the human value function, whatever the human value function is, with a few exceptions around addiction, it's actually very, very robust.

• 30:59Speaker 00

  And so for something like a teenager that's learning to drive, they start to drive and they already have a sense of how they're driving. Immediately, how badly they're unconfident. And then they see, OK. And they, and then, of course, the, the learning speed of any teenager is so fast. After 10 hours, you're good to go.

• 31:17Dwarkesh Patel?

  Yeah, it seems like humans have some solution. And I'm curious about like, well, how are they doing it? And like, why is it so hard to like, how do we need to reconceptualize the way we're training models? To make something like this possible?

• 31:27Speaker 00

  You know, that is a great question to ask. And it's a question I have a lot of opinions about. But unfortunately, we live in a world where not, not all machine learning ideas are discussed freely. And this is, this is one of them. So there's probably a way to do it.

• 31:49Speaker 00

  I think it can be done. The fact that people are like that, I think it's a proof that it can be done. There may be another blocker, though, which is there is a possibility

• 32:02Speaker 00

  that the human neurons actually do more compute than we think. And if that is true, and if that plays an important role, then things might be more difficult. But regardless, I do think it points to the existence of some

• 32:19Speaker 00

  machine learning principle. But I have opinions on, but unfortunately, circumstances make it hard to to discuss in detail.

• 32:28Dwarkesh Patel?

  Nobody listens to this podcast, Ilya.

• 32:31Speaker 00

  Yeah.

• 32:32Dwarkesh Patel?

  So I have to say that prepping for Ilya was pretty tough. Because neither I nor anybody else had any idea what he's working on and what SSI is trying to do. I had no basis to come up with my questions. And the only thing I could go off, honestly, was trying to think from first principles about what are the bottlenecks to AGI? Because clearly, Ilya is working on them in some way. Part of this question involved thinking about RL scaling. Because everybody's asking how well RL will generalize and how we can make it generalize better. As part of this, I was reading this paper that came out recently on RL scaling. And it showed that actually, the learning cover on RL looks like a sigmoid. I found this very curious. Why should it be a sigmoid? Where it learns very little for a long time? And then it quickly learns a lot, and then it asymptotes.

• 33:17Dwarkesh Patel?

  This is very different from the power law you see in pre-training, where the model learns a bunch at the very beginning, and then less and less over time. And it actually reminded me of a note that I had written down after I had a conversation with a researcher friend, where he pointed out that the number of samples that you need to take in order to find a correct answer scales exponentially with how different your current probability distribution is from the target probability distribution. And I was thinking about how these two ideas are related. I had this vague idea that they should be connected, but I really didn't know how. I don't have a math background, so I couldn't really formalize it. But I wondered if Gemini 3 could help me out here. And so I took a picture of my notebook, and I took the paper, and I put them both in the context of Gemini 3, and I asked it to find the connection.

• 34:00Dwarkesh Patel?

  And it thought a bunch, and then it realized that the correct way to model the information you gain from a single yes or no outcome in RL is as the entropy of a random binary variable. It made a graph which showed how the bits you gain for a sample in RL versus supervised learning scale as the pass rate increases. And as soon as I saw the graph that Gemini 3 made, immediately a ton of things started making sense to me. Then I wanted to see if there was any empirical basis to this theory. So I asked Gemini to code an experiment to show whether the improvement in loss scales in this way with pass rate. I just took the code that Gemini outputted. I copy pasted it into a Google Colab notebook and I was able to run this toy ML experiment

• 34:43Dwarkesh Patel?

  and visualize its results without a single bug. It's interesting because the results look similar, but not identical to what we should have expected. And so I downloaded this chart and I put it into Gemini and I asked it, what is going on here? And it came up with a hypothesis that I think is actually correct, which is that we're capping how much supervised learning can improve in the beginning by having a fixed learning rate. And in fact, we should decrease the learning rate over time. It actually gives us an intuitive understanding for why in practice we have learning rate schedulers that decrease the learning rate over time. I did this entire flow from coming up with this vague initial question to building a theoretical understanding, to running some toy ML experiments, all with Gemini 3. This feels like the first model where it can actually come up

• 35:27Dwarkesh Patel?

  with new connections that I wouldn't have anticipated. It's actually now become the default place I go to when I want to brainstorm new ways to think about a problem. If you want to read more about RL scaling, you can check out the blog post that I wrote with a little help from Gemini 3. And if you want to check out Gemini 3 yourself, go to Gemini.google. I am curious, if you say we are back in the era of research, you were there from 2012 to 2020. And do you have... Yeah, what is now the vibe going to be if we go back to the era of research? For example, even after AlexNet, the amount of compute that was used to run experiments kept increasing and the size of frontier systems kept increasing.

• 36:12Dwarkesh Patel?

  And do you think now that this era of research will still require tremendous amounts of compute? Do you think it will require going back into the archives and reading old papers? What is, maybe, what was the vibe of like, you were at Google and OpenAI and Stanford, these places, when there was like a more of a vibe of research? What kind of things should we be expecting in the community?

• 36:38Speaker 00

  So, one consequence of the age of scaling is that there was this.

• 36:47Speaker 00

  Scaling sucked out all the air in the room.

• 36:49Dwarkesh Patel?

  Yeah.

• 36:51Speaker 00

  And so, because scaling sucked out all the air in the room.

• 36:57Speaker 00

  Everyone started to do the same thing. We got to the point where we are in a world where there are more companies than ideas, by quite a bit. Actually, on that, you know, there is this Silicon Valley saying that says that ideas are cheap, execution is everything. And people say that a lot.

• 37:19Dwarkesh Patel?

  Yeah.

• 37:19Speaker 00

  And there is truth to that. But then I saw, I saw someone say on Twitter, something like, if ideas are so cheap, how come no one's having any ideas? And I think it's true, too. I think, like, if you think about research progress in terms of bottlenecks.

• 37:40Speaker 00

  There are several bottlenecks. If you go back to the, if, and one of them is ideas, and one of them is your ability to bring them to life.

• 37:48Dwarkesh Patel?

  Yeah.

• 37:48Speaker 00

  Which might be compute, but also engineering.

• 37:52Speaker 00

  So if you go back to the 90s, let's say, you had people who had pretty good ideas. And if they had much larger computers, maybe they could demonstrate that their ideas were viable, but they could not. So they could only have very, very small demonstration and not convince anyone.

• 38:06Dwarkesh Patel?

  Yeah.

• 38:07Speaker 00

  So the bottleneck was compute. Then in the age of scaling, computers increased a lot. And of course, there is a question of how much compute is needed, but compute is large. So, compute is large enough such that.

• 38:25Speaker 00

  It's like not obvious that you need that much more compute to prove some idea. Like, I'll give you an analogy. AlexNet was built on two GPUs. That was the total amount of compute used for it. The transformer was built on eight to sixty-four GPUs. No single transformer paper experiment used more than sixty-four GPUs of 2017, which would be like, what, two GPUs of today? So the ResNet, right? Many, like even the, you could argue that the, like, O1 reasoning was not the most compute-heavy thing in the world.

• 39:07Speaker 00

  So they're definitely, for, for research,

• 39:13Speaker 00

  you need, like, definitely some amount of compute. But it's far from obvious that you need the absolutely largest amount of compute ever for research. You might argue, and I think it is true, that if you want to build the absolutely best system, if you want to build the absolutely best system, then it helps to have much more compute. And especially if everyone is within the same paradigm, then compute becomes one of the big differentiators.

• 39:42Dwarkesh Patel?

  Yeah, I guess, while it was possible to develop these ideas, I'm asking you for the history because you were actually there. I'm not sure what actually happened. But it sounds like it was possible to develop these ideas using minimal amounts of compute. But it wasn't, the transformer didn't immediately become famous. It became the thing everybody started doing and then started experimenting on top of and building on top of because it was validated at higher and higher levels of compute.

• 40:06Speaker 00

  Correct.

• 40:07Dwarkesh Patel?

  And if you at SSI have fifty different ideas, how will you know which one is the next transformer and which one is, you know, brittle without having the kinds of compute that other frontier labs have?

• 40:22Speaker 00

  So I can, I can comment on that, which is, the short comment is that, you know, you mentioned SSI. Specifically for us, the amount of compute that SSI has for research is really not that small. And I want to explain why. Like, a simple math can explain why the amount of compute that we have is actually a lot more comparable for research than one might think. Now explain. So.

• 40:55Speaker 00

  SSI has raised $3 billion, which is like, not small, but it's like a lot by any absolute sense. But you could say, but look at the other companies raising much more. But a lot of what their, a lot of their compute goes for inference.

• 41:13Speaker 00

  Like these big numbers, these big loans, it's earmarked for inference. That's number one. Number two, you need, if you want to have a product on which you do inference, you need to have a big staff of engineers, of salespeople. A lot of the research needs to be dedicated for producing all kinds of product-related features. So then when you look at what's actually left for research, the difference becomes a lot smaller.

• 41:41Speaker 00

  Now, the other thing is, is that if you're doing something different, do you really need the absolute maximal scale to prove it? I don't think that's true at all. I think that in our case, we have sufficient compute to prove, to convince ourselves and anyone else that what we're doing is correct.

• 42:02Dwarkesh Patel?

  There's been public estimates that, you know, companies like OpenAI spend on the order of five, six billion dollars a year, even just so far, on experiments. This is separate from the amount of money they're sending on inference and so forth. So it seems like they're spending more a year running experiments, like research experiments, than you guys have in total funding.

• 42:22Speaker 00

  I think it's a question of what you do with it. It's a question of what you do with it. Like they have, like, is the more, I think, in their case, in the case of others, I think there is a lot more demand on the training compute. There's a lot more different work streams. There is, there are different modalities. There is just more stuff, and so it becomes fragmented.

• 42:44Dwarkesh Patel?

  How will SSI make money?

• 42:46Speaker 00

  You know,

• 42:48Speaker 00

  my answer to this question is something like...

• 42:52Speaker 00

  We just for right now, we just focus on the research and then the answer to that question will reveal itself. I think there will be lots of possible answers.

• 43:00Dwarkesh Patel?

  Hmm. Is exercise plan still to straight shot superintelligence?

• 43:04Speaker 00

  Maybe. I think that there is merit to it. I think there's a lot of merit because I think that it's very nice to not be affected by the day to day market competition. But I think there are two reasons that may cause us to change the plan. One is pragmatic if timelines turned out to be long, which they might. And second, I think there is a lot of value in the best and most powerful AI being out there impacting the world.

• 43:42Dwarkesh Patel?

  Yeah.

• 43:43Speaker 00

  I think this is a meaningfully valuable thing.

• 43:46Dwarkesh Patel?

  But then, so why is your default plan to straight shot superintelligence? Because it sounds like, you know, OpenAI, Anthropic, all these other companies, their explicit thinking is, look, we have weaker and weaker intelligences that the public can get used to and prepare for. And why is it potentially better to build a superintelligence directly?

• 44:05Speaker 00

  So I'll make the case for and against.

• 44:08Dwarkesh Patel?

  Yeah.

• 44:08Speaker 00

  The case for is that you are, so one of the challenges that people face when they're in the market is that they have to participate in the rat race. And the rat race is quite difficult in that it exposes you to two, to difficult trade offs which you need to make. And there is, it is, it is nice to say, we'll insulate ourselves from all this and just focus on the research and come out only when we are ready and not before. But the counterpoint is valid too. And those are, those are opposing forces. The counterpoint is, hey, it is useful for the world to see powerful AI. It is useful for the world to see powerful AI,

• 44:53Speaker 00

  because that's the only way you can communicate it.

• 44:55Dwarkesh Patel?

  Well, I guess not even just that you can communicate the idea.

• 44:57Speaker 00

  Communicate the AI, not the idea. Communicate the AI.

• 45:02Dwarkesh Patel?

  What do you mean communicate the AI?

• 45:03Speaker 00

  So, okay, so let's suppose you read an essay about AI.

• 45:07Dwarkesh Patel?

  Yeah.

• 45:07Speaker 00

  And the essay says AI is gonna be this, and AI is gonna be that, and it's gonna be this.

• 45:11Dwarkesh Patel?

  Yeah, yeah.

• 45:11Speaker 00

  And you read it and you say, okay, this is an interesting essay.

• 45:14Dwarkesh Patel?

  Right.

• 45:15Speaker 00

  Now suppose you see an AI doing this, an AI doing that. It is incomparable. Like, basically, I think, I think that there is a big benefit from AI being in the public. And that would be a reason for us to not be quite straight shot.

• 45:35Dwarkesh Patel?

  Yeah. Well, I guess it's not even that, but I do think that is an important part of it. The other big thing is, I can't think of another discipline in human engineering and research where the end artifact was made safe-eraged mostly through just thinking about how to make it safe, as opposed to why are airplane crashes per mile so much lower today than they were decades ago? Why is it so much harder to find a bug in Linux than it would have been decades ago? And I think it's mostly because these systems were deployed to the world, you noticed failures, those failures were corrected, and the systems became more robust. And I'm not sure why AGI and superhuman intelligence would be any different, especially given,

• 46:18Dwarkesh Patel?

  and I hope we can, we're gonna get to this.

• 46:23Dwarkesh Patel?

  It seems like the harms of superintelligence are not just about like having some malevolent paperclip out there, but it's just like, this is a really powerful thing, and we don't even know how to conceptualize, how people will interact with it, what people will do with it. And having gradual access to it seems like a

• 46:40Dwarkesh Patel?

  better way to maybe spread out the impact of it and to help people prepare for it.

• 46:45Speaker 00

  Well, I think, I think on this point, even in the straight shot scenario, you would still do a gradual release of it. Is how I would imagine it.

• 46:57Speaker 00

  The gradualism would be an inherent inherent component of any plan. It's just a question of what is the first thing that you get out of the door? That's number one. Number two, I also think, you know, I believe you have advocated for continual learning more than other people. And I actually think that this is an important and correct thing. And here is why. So one of the things, so I'll give you another example of how thinking, how language affects thinking. And in this case, it is will be two words, two words that have shaped everyone's thinking, I maintain. First word, AGI. Second word, pre-training.

• 47:42Speaker 00

  Let me explain. So the word, the term AGI. Why does this term exist? It's a very particular term. Why does it exist? There's a reason. The reason that the term AGI exists is, in my opinion, not so much because it's like a very important, essential descriptor of some end state of intelligence, but

• 48:09Speaker 00

  because it is a reaction to a different term that existed, and the term is narrow AI. If you go back to ancient history of game playing AI, of checkers AI, chess AI, computer games AI, everyone would say, look at this narrow intelligence. Sure, the chess AI can beat Kasparov, but it can't do anything else. It is so narrow, artificial narrow intelligence. So in response, as a reaction to this, some people said, well, this is not good. It is so narrow. What we need is general AI. General AI, an AI that can just do all the things.

• 48:50Speaker 00

  The second, and that term just got a lot of traction.

• 48:55Dwarkesh Patel?

  Yeah.

• 48:56Speaker 00

  The second thing that got a lot of traction is pre-training. Specifically, the recipe of pre-training. I think the current, the way people do RL now is maybe

• 49:08Speaker 00

  is undoing the conceptual imprint of pre-training. But pre-training had the property, you do more pre-training and the model gets better at everything, more or less uniformly.

• 49:20Dwarkesh Patel?

  Yeah.

• 49:21Speaker 00

  General AI. Pre-training gives AGI.

• 49:27Speaker 00

  But the thing that happened with AGI and pre-training is that in some sense they overshot the target. Because by the kind, if you think about the term AGI, you will realize, and especially in the context of pre-training, you will realize that a human being is not an AGI. Because a human being, yes, there is definitely a foundation of skills. A human being, a human being lacks a huge amount of knowledge. Instead, we rely on continual learning. We rely on continual learning. And so then, when you think about, okay, so let's suppose that we achieve success and we produce a safe, some kind of safe superintelligence.

• 50:12Speaker 00

  The question is, but how do you define it? Where on the curve of continual learning is it going to be? I will produce like a superintelligent 15-year-old that's very eager to go and you say, okay, I'm going to, they don't know very much at all, the great student, very eager. You go and be a programmer. You go and be a doctor. Go and learn. So you could imagine that the deployment itself will involve some kind of a learning, trial and error period. It's a process as opposed to you drop the finished thing.

• 50:43Dwarkesh Patel?

  Okay, I see. So you're suggesting that the thing you're pointing out with superintelligence is not some finished mind which knows how to do every single job in the economy. Because the way, say, the original, I think, OpenAI charter or whatever defines AGI is like, it can do every single job, that every single thing a human can do. You're proposing instead a mind which can learn to do any single, every single job.

• 51:13Speaker 00

  Yes.

• 51:14Dwarkesh Patel?

  And that is superintelligence. And then, but once you have the learning algorithm. It gets deployed into the world the same way a human laborer might join an organization. And it seems like one of these two things might happen, maybe neither of these happens. One, this super efficient learning algorithm becomes superhuman, becomes as good as you, and potentially even better at the task of ML research. And as a result, the algorithm itself becomes more and more superhuman. The other is, even if that doesn't happen, if you have a single model, I mean, this is explicitly your vision, if you have a single model or instances of a model which are deployed through

• 51:58Dwarkesh Patel?

  the economy, doing different jobs, learning how to do those jobs, continually learning on the job, picking up all the skills that any human could pick up, but actually picking them all up at the same time and then amalgamating the learnings. You basically have a model which functionally becomes super intelligent even without any sort of recursive self-improvement in software, right? Because you now have one model that can do every single job in the economy, and humans can't merge our minds in the same way. And so do you expect some sort of like intelligence explosion from broad deployment?

• 52:28Speaker 00

  I think that it is likely that we will have rapid economic growth. I think the broad deployment, like, there are two arguments you could make which are conflicting. One is that, look, if indeed you get, once indeed you get to a point where you have an AI that can learn to do things quickly, and you have many of them, then they will, then there will be a strong force to deploy them in the economy unless there will be some kind of a regulation that stops it, which, by the way, there might be.

• 53:13Speaker 00

  But I think the idea of very rapid economic growth for some time, I think it's very possible from broad deployment. The other question is how rapid it's going to be.

• 53:25Speaker 00

  So I think this is hard to know because on the one hand, you have this very efficient worker. On the other hand, there is, the world is just really big. And there's a lot of stuff.

• 53:36Speaker 00

  And that stuff moves at a different speed. But then on the other hand, now the AI could, you know, so I think very rapid economic growth is possible. And we will see like all kinds of things like different countries with different rules and the ones which have the friendlier rules, the economic growth will be faster. Hard to predict.

• 53:54Dwarkesh Patel?

  Some people in our audience like to read the transcripts instead of listening to the episode. And so we put a ton of effort into making the transcripts read like they are standalone essays. The problem is that if you just transcribe a conversation verbatim using a speech to text model, it'll be full of all kinds of fits and starts and confusing phrasing. We mentioned this problem to Label Box and they asked if they could take a stab. Working with them on this is probably the reason that I'm most excited to recommend Label Box to people. It wasn't just, oh, hey, tell us what kind of data you need and we'll go get it. They walked us through the entire process from helping us identify what kind of data we needed in the first place to assembling a team of expert aligners to generate it. Even after we got all the data back, Label Box stayed involved.

• 54:38Dwarkesh Patel?

  They helped us choose the right base model and set up auto QA on the model's output so that we could tweak and refine it. And now we have a new transcriber tool that we can use for all our episodes moving forward. This is just one example of how Label Box meets their customers at the ideas level and partners with them through their entire journey. If you want to learn more or if you want to try out the transcriber tool yourself, go to labelbox.com/thwarkesh.

• 55:07Dwarkesh Patel?

  It seems to me that this is a very precarious situation to be in where look in the limit, we know that this should be possible because if you have something that is as good as a human at learning, but which can merge its brains, merge their different instances in a way that humans can't merge. Already, this seems like a thing that should physically be possible. Humans are possible. Digital computers are possible. You just need both of those combined to produce this thing. And it also seems like this kind of thing is Extremely powerful. And economic growth is one way to put it. I mean, Dyson Spears, a lot of economic growth. But another way to put it is just like, you will have potentially a very short period of time because a human on the job can,

• 55:52Dwarkesh Patel?

  you know, you're hiring people to SSI and six months they're like net productive probably, right? A human like learns really fast and so this thing is becoming smarter and smarter very fast. What is, how do you think about making that go well? And why is SSI positioned to do that well? What is SSI's plan there, basically, is what I'm trying to ask?

• 56:07Speaker 00

  Yeah.

• 56:10Speaker 00

  So, one of the, one of the ways in which my thinking has been changing is that

• 56:19Speaker 00

  I now place more importance on AI being deployed incrementally and in advance. One very difficult thing about AI is that we are talking about systems that don't yet exist. And it's hard to imagine them. I think that one of the things that's happening is that in practice, it's very hard to feel the AGI.

• 56:52Speaker 00

  It's very hard to feel the AGI. We can talk about it. But it's like, it's like talking about like the long future. Like imagine like having a conversation about like, how is it like to be old? When you're like old and frail and you can have a conversation, you can try to imagine it, but it's just hard and you come back to reality where that's not the case. And I think that a lot of the issues around AGI and its future power stem from the fact that it's very difficult to imagine. Future AI is going to be different. It's going to be powerful.

• 57:35Speaker 00

  Indeed, the whole problem, what is the problem of AI and AGI? The whole problem is the power.

• 57:43Speaker 00

  The whole problem is the power.

• 57:46Speaker 00

  When the power is really big, what's gonna happen? And one of the ways in which I've changed my mind over the past year and so that that change of mind may back may, I'll say, I'll hedge a little bit, may back propagate into the plans of our company is that

• 58:08Speaker 00

  so if it's hard to imagine,

• 58:12Speaker 00

  what do you do? You gotta be showing the thing. You gotta be showing the thing. And I maintain that, I think, I think most people who work on AI also can't imagine it. Because it's too different from what people see on a day-to-day basis.

• 58:28Speaker 00

  I do maintain, here's something which I predict will happen. That's a prediction. I maintain that as AI becomes more powerful, then people will change their behaviors.

• 58:45Speaker 00

  And we will see all kinds of unprecedented things which are not happening right now. And I'll give some examples. I do, like, I think, I think for better or worse, the frontier companies will play a very important role in what happens, as will the government. And the kind of things that I think we'll see, which you see the beginnings of, companies that are fierce competitors, starting to collaborate on AI safety. You may have seen OpenAI and Anthropic doing a first small step, but that did not exist. That's actually something which I predicted in one of my talks about three years ago, that such a thing will happen.

• 59:29Speaker 00

  I also maintain that as AI continues to become more powerful, more visibly powerful, there will also be a desire from governments and the public to do something.

• 59:42Speaker 00

  And I think that this is a very important force

• 59:46Speaker 00

  of showing the AI. That's number one. Number two, okay, so then the AI is being built. What needs to be done?

• 59:55Speaker 00

  So one thing that I maintain that will happen is that right now, people who are working on AI, I maintain that the AI doesn't feel powerful because of its mistakes. I do think that at some point, the AI will start to feel powerful, actually. And I think when that happens, we will see a big change in the way all AI companies approach safety. They'll become much more paranoid. I think, I say this as a prediction that we will see happen. We'll see if I'm right. But I think this is something that will happen because they will see the AI becoming more powerful. Everything that's happening right now, I maintain, is because people look at today's AI and it's hard to imagine the future AI.

• 1:00:41Speaker 00

  And there is a third thing which needs to happen. And I think this is, and I'm talking about it in broader terms, not just from the perspective of SSI, because you asked me about our company. But the question is, okay, so then what should, what should the companies aspire to build?

• 1:00:57Dwarkesh Patel?

  Yeah.

• 1:00:58Speaker 00

  What should they aspire to build? And there has been one big idea that actually, that everyone has been locked in, locked into, which is the self-improving AI. And why, why did it happen? Because there is fewer ideas than companies. But I maintain that there is something that's better to build. And I think that everyone will actually want that. It's like the AI that's robustly aligned to care about sentient life specifically. I think in particular, it will be, there is a case to be made that it will be easier to build an AI that cares about sentient life than an AI that cares about human life alone, because the AI itself will be sentient.

• 1:01:44Speaker 00

  And if you think about things like mirror neurons and human empathy for animals, which is, you know, you might argue it's not big enough, but it exists. I think it's an emergent property from the fact that we model others with the same circuit that we use to model ourselves, because that's the most efficient thing to do.

• 1:02:03Dwarkesh Patel?

  So even if you got an AI to care about sentient beings... And it's not actually clear to me that that's what you should try to do if you solve the alignment. It would still be the case that most sentient beings will be AIs. There will be trillions, eventually quadrillions of AIs. Humans will be a very small fraction of sentient beings. So it's not clear to me if the goal is some kind of human control over this future civilization, that this is the best criterion.

• 1:02:35Speaker 00

  It's true. I, I think that it's possible it's not the best criterion. I'll say two things. I think that thing number one.

• 1:02:47Speaker 00

  I think that if there, so I think that care for sentient life, I think there is merit to it. I think it should be considered. I think that it will be helpful if there was some kind of a shortlist of ideas that then the companies, when they are in this situation, could use. That's number two. Number three, I think it would be really materially helpful if the power of the most powerful superintelligence was somehow capped.

• 1:03:20Speaker 00

  Because it would address a lot of these concerns. The question of how to do it, I'm not sure. But I think that would be materially helpful. When you're talking about really, really powerful systems.

• 1:03:32Dwarkesh Patel?

  Yeah, um, before we continue the alignment discussion, I, I want to double click on that. How much room is there at the top? How do you think about superintelligence? Do you think, I mean, using this learning efficiency idea, maybe it's just extremely fast at learning new skills or new knowledge? And does it just have a bigger pool of strategies? Is there a single cohesive it?

• 1:03:54Dwarkesh Patel?

  In the center that's more powerful or bigger? And if so, do you, do you imagine that this will be sort of godlike in comparison to the rest of human civilization? Or does it just feel like another agent or another cluster of agents?

• 1:04:08Speaker 00

  So this is an area where different people have different intuitions. I think it will be very powerful for sure. I think that what I think is most likely to happen. Is that there will be multiple such AIs being created roughly at the same time. I think that if the cluster is big enough, like if the cluster is literally continent sized, that thing could be really powerful indeed, right? If you literally have a continent sized cluster like those, those AIs can be very powerful. And I, like, all I can tell you is that if you're talking about extremely powerful AGIs,

• 1:04:51Speaker 00

  like truly dramatically powerful, then yeah, it would be nice if they could be restrained in some ways, or if there was some kind of an agreement or something. Because I think that if you're saying, hey, like, if, if you really, like, what, what is the, the concern of superintelligence? What is one way to explain the concern? If you imagine a system that is sufficiently powerful, like really sufficiently powerful, and you could say, okay, you need to do something sensible, like care for sentient life, let's say, in a very single minded way, we might not like the results. That's really what it is. And so maybe, by the way, the answer is that you do not build a single, you do not build an RL agent in the usual sense.

• 1:05:36Speaker 00

  And actually, I'll point, I'll point several things out. I think human beings are a semi RL agent. You know, we pursue a reward and then the emotions or whatever make us tire out of the reward, we pursue a different reward. The market is like, kind, it's like a very short sighted kind of agent. Evolution is the same. Evolution is very intelligent in some ways, but very dumb in other ways. The government has been designed to be a never ending fight between three parts, which has an effect. So I think things like this.

• 1:06:11Speaker 00

  Another thing that makes this discussion difficult is that we are talking about systems that don't exist, that we don't know how to build. Right? That's the other thing. And that's actually my belief. I think what people are doing right now will go some distance and then peter out. It will continue to improve, but it will also not be it. So the it, we don't know how to build. And I think that a lot, a lot hinges on understanding reliable generalization.

• 1:06:41Speaker 00

  Now I'll say another thing, which is like,

• 1:06:45Speaker 00

  you know, one of the things that you could say is what would that cause alignment to be difficult is that human value, that it's, it's um,

• 1:06:53Speaker 00

  your ability to learn human values is fragile, then your ability to optimize them is fragile, you will, you actually learn to optimize them. And then can't you say, are these not all instances of unreliable generalization?

• 1:07:06Speaker 00

  Why is it that human beings appear to generalize so much better? What if generalization was much better? What would happen in this case? What would be the effect? But those we can't, we can't, like, those questions are right now still answerable.

• 1:07:18Dwarkesh Patel?

  Um, how does one think about what AI going well looks like? Because I think you've scoped out how AI might evolve, will have these sort of continual learning agents. AI will be very powerful. Maybe there will be many different AIs. How do you think about lots of continent compute size intelligences going around? How dangerous is that? How do we make that less dangerous? And how do we do that in a way that

• 1:07:48Dwarkesh Patel?

  protects equilibrium where there might be misaligned AIs out there? And bad actors out there?

• 1:07:56Speaker 00

  So one reason why I liked the AI that cares for sentient life, you know, and we can debate on whether it's good or bad. But if the first N of these dramatic systems actually do care for, you know,

• 1:08:14Speaker 00

  love, humanity, or something, you know, care for sentient life, obviously, this also needs to be achieved. This needs to be achieved. So if this is achieved by the first N of those systems,

• 1:08:27Speaker 00

  then, then I can see it go well. At least for quite some time. And then there is the question of what happens in the long run? What happens in the long run? How do you achieve a long run equilibrium? And I think that there, there is an answer as well. And I don't like this answer. But it needs to be considered.

• 1:08:51Speaker 00

  In the long run, you might say, okay, so if you have a world where powerful AGIs exist, in the short term, you could say, okay, you have universal high income. You have universal high income. And we're all doing well. But we know that... What do the Buddhists say? Change is the only constant. And so things change. And there is some kind of government political structure thing. And it changes. Because these things have a shelf life. You know, some new government thing comes up and it functions. And then after some time, it stops functioning. That's something that you see happening all the time. And so I think that for the long run equilibrium,

• 1:09:32Speaker 00

  one approach, you could say, okay, so maybe every person will have an AI that will do their bidding. And that's good. And if that could be maintained indefinitely, that's true. But the downside with that is, okay, so then the AI goes and like earns, you know, earns money for the person and, you know, advocates for their needs in like the political sphere. And maybe then writes a little report saying, okay, here's what I've done. Here's the situation. And the person says, great, keep it up. But the person is no longer a participant. And then you can say that's a precarious place to be in. But, so I'm going to preface by saying,

• 1:10:13Speaker 00

  I don't like this solution, but it is a solution. And the solution is if people become part AI with some kind of Neuralink++. Because what will happen as a result is that now the AI understands something. And we understand it too. Because now they're understanding it's transmitted wholesale. So now if the AI is in some situation, now it's like you are involved in that situation yourself fully. And I think this is the answer to the equilibrium.

• 1:10:44Dwarkesh Patel?

  I wonder if the fact that emotions, which were developed millions, or in many cases billions of years ago, in a totally different environment, are still guiding our actions so strongly is an example of alignment success. To maybe spell out what I mean. The brainstem has these...

• 1:11:10Dwarkesh Patel?

  I don't know if it's more accurate to call it a value function or a reward function, but the brainstem has a directive where it's saying mate with somebody who's more successful. The cortex is the part that understands what does success mean in the modern context. But the brainstem is able to align the cortex and say however you recognize success to be, and I'm not smart enough to understand what that is, you're still going to pursue this directive.

• 1:11:31Speaker 00

  I think, I think there is, so I think there's a more general point. I think it's actually really mysterious how the brain encodes high level desires. Sorry, how evolution encodes high level desires. Like, it's pretty easy to understand how evolution would, would endow us with the desire for food that smells good, because smell is a chemical, and so just pursue that chemical. It's very easy to imagine such, and make evolution doing such a thing. But evolution also has endowed us with all these social desires. Like, we really care about being seen positively by society. We care about being in a good standing.

• 1:12:14Speaker 00

  We, like, all these social intuitions that we have, I feel strongly that they're baked in. And I don't know how evolution did it, because it's a high-level concept that's represented in the brain. Like, what people think, like, let's say you are like, you care about some social thing. It's not like a low-level signal like smell. It's not something that, for which there is a sensor. Like, the brain needs to do a lot of processing to piece together lots of bits of information, to understand what's going on socially, and somehow evolution said, that's what you should care about.

• 1:12:53Dwarkesh Patel?

  Yes.

• 1:12:54Speaker 00

  How did it do it? And it did it quickly, too.

• 1:12:56Dwarkesh Patel?

  Yeah.

• 1:12:57Speaker 00

  Because I think all these sophisticated social things that we care about, I think they evolved pretty recently.

• 1:13:03Dwarkesh Patel?

  Yeah.

• 1:13:04Speaker 00

  So evolution had an easy time hard-coding this high-level desire. And I maintain, or, you know, at least I'll say, I'm unaware of good hypothesis for how it's done. I had some ideas I was kicking around, but none of them, none of them.

• 1:13:22Speaker 00

  Are satisfying.

• 1:13:23Dwarkesh Patel?

  Yeah, and what's especially impressive is it was a desire that you learned in your lifetime. It kind of makes sense because your brain is intelligent. It makes sense why we're able to learn intelligent desires. But your point is that the desire is, maybe this is not your point, but one way to understand it is the desire is built into the genome and the genome is not intelligent, right? But it's able to, you're somehow able to describe this feature that requires, like, it's not even clear how you define that feature and you can get it into, you can build it into the genes.

• 1:13:53Speaker 00

  Yeah, essentially, or maybe I'll put it differently. If you think about the tools that are available to the genome, it says, okay, here's a recipe for building a brain. And you could say, here is a recipe for connecting the dopamine neurons to like the smell sensor.

• 1:14:08Dwarkesh Patel?

  Yeah.

• 1:14:09Speaker 00

  And if the smell is a certain kind of, you know, good smell, you want to eat that. I could imagine the genome doing that. I'm claiming that it is harder to imagine. It's harder to imagine the genome saying you should care about some complicated computation that your entire brain, that like a big chunk of your brain does. That's all I'm claiming. I can tell you like a speculation, I was wondering how it could be done. And let me offer a speculation and I'll explain why the speculation is probably false. So the speculation is, okay. So the brain, it's like, the brain has those regions, you know, the brain regions. We have our cortex, right?

• 1:14:49Dwarkesh Patel?

  Yeah.

• 1:14:49Speaker 00

  It has all those brain regions. And the cortex is uniform, but the brain regions and the neurons in the cortex, they kind of speak to their neighbors mostly. And that explains why you get brain regions. Because if you want to do some kind of speech processing, all the neurons that do speech need to talk to each other. And because neurons can only speak to their nearby neighbors, for the most part, it has to be a region. All the regions are mostly located in the same place from person to person. So maybe evolution hardcoded literally a location on the brain.

• 1:15:21Speaker 00

  So it says, oh, like when, when like, you know, the GPS of the brain, GPS coordinates, such and such, when that fires, that's what you should care about. Like, maybe that's what evolution did, because that would be within the toolkit of evolution.

• 1:15:33Dwarkesh Patel?

  Yeah, although there are examples where, for example, people who are born blind have that area of their cortex adopted by another sense. And I have no idea, but I'd be surprised if the desires or the reward functions, which require a visual signal, no longer worked. You know, people who have their different areas of their cortex co-opted. For example, if you no longer have vision, can you still feel the sense that I want people around me to like me and so forth? Which usually there's also visual cues for.

• 1:16:09Speaker 00

  So I actually fully agree with that. I, I think there's an even stronger counterargument to this theory, which is, like, if you think about people, so there are people who get half of their brain removed in childhood.

• 1:16:22Dwarkesh Patel?

  Yeah.

• 1:16:23Speaker 00

  And they still have all their brain regions, but they all somehow move to just one hemisphere, which suggests that the brain regions, the location is not fixed. And so that theory is not true. It would have been cool if it was true, but it's not. And so I think that's a mystery, but it's an interesting mystery. Like, the fact is somehow evolution was able to endow us to care about social stuff very, very reliably. And even people who have like all kinds of strange mental conditions and deficiencies and emotional problems tend to care about this also.

• 1:16:54Dwarkesh Patel?

  AI tools like defects, voice clones and agents have dramatically increased the sophistication of fraud and abuse. So it's more important than ever to actually understand the identity and intent of whoever or whatever is using your platform. That's exactly what Sardine helps you do. Sardine brings together thousands of device, behavior and identity signals to help you assess risk. Everything from how a user types or moves their mouse or holds their device, to whether they're hiding their true location behind a VPN, to whether they're injecting a fake Camera feed during KYC selfie checks. Sardine combines these signals with insights from their network of almost 4 billion devices.

• 1:17:38Dwarkesh Patel?

  Things like a user's history of fraud or their associations with other high-risk accounts. So you can spot bad actors before they do damage. This would literally be impossible if you only use data from your own application. Sardine doesn't stop a detection. They offer a suite of agents to streamline onboarding checks and automate investigations. So as fraudsters use AI to scale their attacks, you can use AI to scale your defenses. Go to sardine.ai/thawarkesh to learn more and download their guide on AI fraud detection.

• 1:18:13Dwarkesh Patel?

  What is SSI planning on doing differently? So presumably, your plan is to be one of the frontier companies when this time arrives. And then what is... Presumably, you started SSI because you're like, I think I have a way of approaching how to do this safely in a way that the other companies don't. What is that difference?

• 1:18:34Speaker 00

  So the way I would describe it as there are some ideas that I think are promising, and I want to investigate them and see if they are indeed promising or not. It's really that simple. It's an attempt. I think that if the ideas turn out to be correct, these ideas that we discussed around understanding generalization.

• 1:18:56Speaker 00

  If these ideas turn out to be correct,

• 1:19:00Speaker 00

  then I think we will have something worthy. We'll did turn out to be correct. We are doing research. We are squarely age of research company. We are making progress. We've actually made quite good progress over the past year, but we need to keep making more progress.

• 1:19:14Dwarkesh Patel?

  Yeah.

• 1:19:14Speaker 00

  More research. And that's how I see it. I see it as an attempt to be...

• 1:19:22Speaker 00

  An attempt to be a voice and a participant.

• 1:19:27Dwarkesh Patel?

  People have asked, your co-founder and previous CEO left to go to Meta recently. And people have asked, well, if there was a lot of breakthroughs being made, that seems like a thing that should have been unlikely. I wonder how you respond.

• 1:19:43Speaker 00

  Yeah, so for this, I will simply remind a few facts that may have been forgotten. And I think these facts, which provide the context, I think they explain the situation. So the context was that we were fundraising at a 32 billion valuation. And then Meta came in and offered to acquire us. And I said, no. But my former co-founder, like in some sense, said yes. And as a result, he also was able to enjoy from a lot of near-term liquidity. And he was the only person from SSI to join Meta.

• 1:20:25Dwarkesh Patel?

  It sounds like SSI's plan is to be a company that is at the frontier when you get to this

• 1:20:31Dwarkesh Patel?

  very important period in human history, where you have superhuman intelligence and you have these ideas about how to make superhuman intelligence go well. But other companies will be trying their own ideas. What distinguishes SSI's approach to making superintelligence go well?

• 1:20:47Speaker 00

  The main thing that distinguishes SSI is its technical approach. So we have a different technical approach that I think is worthy.

• 1:20:59Speaker 00

  And we are pursuing it. I maintain that in the end, there will be a convergence of strategies. So I think there will be a convergence of strategies where

• 1:21:10Speaker 00

  at some point, as AI becomes more powerful, it's going to become more or less clearer to everyone what the strategy should be. And it should be something like, yeah, you need to find some way to talk to each other. And you want your first actual, like real, super intelligent AI to be aligned and somehow be.

• 1:21:35Speaker 00

  You know, care for sentient life, care for people, democratic, one of those, some combination of thereof. And I think this is the condition that everyone should strive for. And that's what the society is striving for. And I think that this time, if not already, all the other companies will be realizing that they're striving towards the same thing. And we'll see. I think that the world will truly change as AI becomes more powerful.

• 1:22:03Dwarkesh Patel?

  Yeah.

• 1:22:04Speaker 00

  And I think a lot of these forecasts will, like, I think things will be really different and people will be acting really differently.

• 1:22:12Dwarkesh Patel?

  What, speaking of forecasts, what are your forecasts to this system you're describing, which can learn as well as a human and subsequently as a result becomes superhuman?

• 1:22:23Speaker 00

  I think like five to 20.

• 1:22:26Dwarkesh Patel?

  Five to 20 years?

• 1:22:27Speaker 00

  Mm hmm.

• 1:22:28Dwarkesh Patel?

  So I just want to unroll your, how you might see the world coming. It's like, we have a couple more years where these other companies are continuing the current approach and it stalls out. And stalls out here, meaning they earn no more than low hundreds of billions in revenue. Or how do you think about what stalling out means?

• 1:22:46Speaker 00

  Yeah. I think there, I think it could, I think it could stall out and I think stalling out will look like, it will all look very similar.

• 1:22:58Dwarkesh Patel?

  Yeah.

• 1:22:58Speaker 00

  Among all the different companies, something like this. I'm not sure because I think, I think, I think even with, I think even, I think even with stalling out, I think these companies could make a stupendous, stupendous revenue. Maybe not profits because they will be, it will be, they will need to work hard to differentiate each other from themselves. But revenue, definitely.

• 1:23:18Dwarkesh Patel?

  But there's something in your model implies that when the correct solution does emerge, there will be convergence between all the companies. And I'm curious why you think that's the case.

• 1:23:29Speaker 00

  Well, I was talking more about convergence on their larger strategies.

• 1:23:32Dwarkesh Patel?

  Hmm.

• 1:23:33Speaker 00

  I think eventual convergence on the technical approach is probably going to happen as well. But I, I was alluding to convergence to the larger strategies. So what, what, what exactly is the thing that should be done?

• 1:23:43Dwarkesh Patel?

  I, I just want to better understand how you see the future unrolling. So currently we have these different companies and you expect their approach to continue generating revenue.

• 1:23:50Speaker 00

  Yes.

• 1:23:51Dwarkesh Patel?

  But not get to this human-like learner.

• 1:23:52Speaker 00

  Yes.

• 1:23:54Dwarkesh Patel?

  So now we have these different forks of companies. We have you, we have thinking machines, there's a bunch of other labs.

• 1:23:59Speaker 00

  Yes.

• 1:23:59Dwarkesh Patel?

  And maybe one of them figures out the correct approach. But then the release of their product makes it clear to other people how to do this thing.

• 1:24:07Speaker 00

  I think it won't be clear how to do it thing, but it will be clear that something different is possible.

• 1:24:12Dwarkesh Patel?

  Right.

• 1:24:12Speaker 00

  And that is information. And I think people will then be trying to figure out how that's how that works. I do think though, that one of the things that I think, you know, not addressed here, not discussed is that

• 1:24:30Speaker 00

  with each increase in the AI's capabilities, I think there will be some kind of changes. But I don't know exactly which ones in how things are being done. And so like, I think it's going to be important, yet I can't spell out what that is exactly.

• 1:24:47Dwarkesh Patel?

  And how are the?

• 1:24:50Dwarkesh Patel?

  By default, you would expect the company that has, the model company that has that model to be getting all these gains because they have the model that is learning how to do all, has the skills and knowledge that it's building up in the world. What is the reason to think that the benefits of that would be widely distributed? And not just end up at whatever model company gets this continuous learning loop going first?

• 1:25:11Speaker 00

  Like, I think that empirically what happened, so here is what I think is going to happen. Number one, I think empirically, when

• 1:25:22Speaker 00

  let's look at, let's look at how things have gone so far with the AIs of the past. So one company produced in advance and the other company scrambled and produced some similar things after some amount of time. And they started to compete in the market and push their, push the prices down. And so I think from the market perspective, I think something similar will happen there as well. Even if someone, it's okay, we are talking about the good world, by the way, where? What's the good world? What's the good world?

• 1:26:01Speaker 00

  Where we have these powerful, human-like learners that are also like, and by the way, maybe there's another thing we haven't discussed on the on the spec of the superintelligent AI that I think is worth considering, is that you make it narrow. Can be useful and narrow at the same time. So you can have lots of narrow superintelligent AIs. But suppose you have many of them. And you have some, and you have some company that's producing a lot of profits from it. And then you have another company that comes in and starts to compete. And the way the competition is going to work is through specialization.

• 1:26:42Speaker 00

  I think what's going to happen is that the way competition, like competition loves specialization. And you see it in the market, you see it in evolution as well. So you're going to have lots of different niches. And you're going to have lots of different companies who are occupying different niches in in this kind of world where you might say, yeah, like one AI company is really quite a bit better at some area of really complicated economic activity. And a different company is better at another area. And the third company is really good at litigation.

• 1:27:15Dwarkesh Patel?

  But is this contradicted by what human-like learning implies? Is that like it can learn?

• 1:27:19Speaker 00

  It can, but, but you have accumulated learning. You have a big investment. You spent a lot of compute to become really, really, really good, really phenomenal at this thing. And someone else spent a huge amount of compute and a huge amount of experience to get really, really good at some other thing.

• 1:27:35Dwarkesh Patel?

  Right.

• 1:27:36Speaker 00

  You apply a lot of human learning to get there. But now like you, you are at this high point where someone else would say, look, like, I don't want to start learning what you've learned to go through this.

• 1:27:46Dwarkesh Patel?

  That would require many different companies to begin at the human-like continual learning agent at the same time. So that they can start their different research in different branches. But if one company You know, gets that agent first or gets that learner first.

• 1:28:05Dwarkesh Patel?

  It does then seem like, well, you know, they like, if you just think about every single job in the economy.

• 1:28:12Dwarkesh Patel?

  You just have instance learning each one seems tractable for.

• 1:28:17Speaker 00

  Yeah, that's, that's, that's a valid argument. My, my strong intuition is that it's not how it's gonna go.

• 1:28:24Speaker 00

  My strong intuition is that, yeah, like the argument says it will go this way.

• 1:28:27Dwarkesh Patel?

  Yeah.

• 1:28:28Speaker 00

  But my strong intuition is that it will not go this way. That this is the, you know, in, in theory, there is no difference between theory and practice and practice there is. And I think that's going to be one of those.

• 1:28:39Dwarkesh Patel?

  A lot of people's models of recursive self improvement literally explicitly state, we will have a million Ilias in a server. They're coming in with different ideas and this will lead to a super intelligence emerging very fast. Do you have some intuition about how parallelizable the thing you are doing is? How, how, what are the gains from making copies of Ilya?

• 1:29:00Speaker 00

  I don't know. I think.

• 1:29:04Speaker 00

  I think there'll definitely be, there'll be diminishing returns because you want, you want people who think differently rather than the same. I think that if they were literal copies of me, I'm not sure how much more incremental value you'd get. I think that.

• 1:29:20Speaker 00

  But people who think differently, that's what you want.

• 1:29:23Dwarkesh Patel?

  Why is it that it's been, if you look at different models, even released by totally different companies, trained on potentially non-overlapping data sets, it's actually crazy how similar LLMs are to each other.

• 1:29:35Speaker 00

  Maybe the data sets are not as non-overlapping as it seems.

• 1:29:39Dwarkesh Patel?

  But there's, there's some sense that it's like, even if an individual human might be less productive than the future AI, maybe there's something to the fact that human teams have more diversity than teams of AIs might have, but how do we elicit meaningful diversity among AIs? So I think just raising the temperature just results in gibberish. I think you want something more like different scientists have a different, different prejudices or different ideas. How do you get that kind of diversity among AI agents?

• 1:30:04Speaker 00

  So the reason there has been no diversity, I believe, is because of pre-training. All the pre-trained models are the same, pretty much, because they pre-trained on the same data. Now, RL and post-training is where some differentiation starts to emerge because different people come up with different RL training.

• 1:30:24Dwarkesh Patel?

  Yeah, and then I've heard you hint in the past about self-play as a way to either get data or match agents to other agents with equivalent intelligence to kickoff learning. How should we think about why there's no public proposals of this kind of thing working with LLMs?

• 1:30:46Speaker 00

  I would say there are two things to say. I would say that the reason why I thought self-play was interesting is because it offered a way to create models using compute only without data. Right? And if you think that data is the ultimate bottleneck, then using compute only is very interesting. So that's what makes it interesting. Now, the thing is

• 1:31:13Speaker 00

  that self-play, at least the way it was done in the past, when you have agents which somehow compete with each other, it's only good for developing a certain set of skills. It is too narrow. It's only good for like negotiation, conflict, certain social skills, strategizing, that kind of stuff. And so if you care about those skills, then self-play will be useful. Now, actually, I think that self-play did find a home, but just in a different form. In a different form. So things like debate, prove or verifier, you have some kind of an LLM as a judge which is also incentivized to find mistakes in your work.

• 1:31:58Speaker 00

  You could say this is not exactly self-play, but this is, you know, a related adversarial setup that people are doing, I believe. And really, self-play is an example of, is a special case of more general like

• 1:32:10Speaker 00

  competition between agents. Right? The response, the natural response to competition is to try to be different. And so if you were to put multiple agents and you tell them, you know, you all need to work on some problem, and you're an agent, and you're inspecting what everyone else is working, you're gonna say, well, if they're already taking this approach... It's not clear I should pursue it. I should pursue something differentiated. And so I think that something like this could also create an incentive for a diversity of approaches.

• 1:32:38Dwarkesh Patel?

  Yeah. Um. Final question.

• 1:32:43Dwarkesh Patel?

  What is research taste? You're obviously

• 1:32:47Dwarkesh Patel?

  the person in the world who is considered to have the best taste in doing research in AI. You were

• 1:32:57Dwarkesh Patel?

  the co-author on many of the biggest, the biggest things that have happened in the history of deep learning, from AlexNet to GPT-3 to so on. What is it that, how do you characterize how

• 1:33:09Dwarkesh Patel?

  you come up with these ideas?

• 1:33:11Speaker 00

  I can answer. So I can comment on this for myself. I think different people do it differently.

• 1:33:18Speaker 00

  But one thing that guides me personally is an aesthetic of how AI should be by thinking about how people are, but thinking correctly. Like, it's very easy to think about how people are incorrectly. But what does it mean to think about people correctly? So I'll give you some examples. The idea of the artificial neuron is directly inspired by the brain. And it's a great idea. Why? Because you say, sure, the brain has all these different organs, it has the folds, but the folds probably don't matter. Why do we think that the neurons matter? Because there is many of them. It kind of feels right, so you want the neuron.

• 1:34:00Dwarkesh Patel?

  Yeah.

• 1:34:01Speaker 00

  You want some kind of local learning rule that will change the connections. You want some local learning rule that will change the connections between the neurons.

• 1:34:09Speaker 00

  Right? It feels plausible that the brain does it. The idea of the distributed representation.

• 1:34:15Speaker 00

  The idea that the brain,

• 1:34:18Speaker 00

  you know, the brain responds to experience, a neuron that should learn from experience, not response. The brain learns from experience. The neural natural level of experience. And you kind of ask yourself, is something fundamental or not fundamental? How things should be?

• 1:34:31Dwarkesh Patel?

  Yeah.

• 1:34:32Speaker 00

  And I think that's been guiding me a fair bit. Kind of thinking from multiple angles and looking for almost beauty. Beauty, simplicity, ugliness, there's no room for ugliness. It's just beauty, simplicity, elegance, correct inspiration from the brain. And all of those things need to be present at the same time. And the more they are present, the more confident you can be in a top down belief.

• 1:34:56Speaker 00

  And then the top down belief is the thing that sustains you when the experiments contradict you. Because if you just trust the data all the time, well, sometimes you can be doing a correct thing, but there's a bug. But you don't know that there is a bug. How can you tell that there is a bug? How do you know if you should keep debugging or you conclude it's the wrong direction? Well, it's the top down. Well, how should you can say the things have to be this way?

• 1:35:18Dwarkesh Patel?

  Right.

• 1:35:18Speaker 00

  Something like this has to work. Therefore, we gotta keep going. That's the top down. And it's based on this like multifaceted beauty and inspiration by the brain.

• 1:35:29Dwarkesh Patel?

  All right. We'll leave it there.

• 1:35:31Speaker 00

  Thank you so much.

• 1:35:32Dwarkesh Patel?

  Thank you so much.

• 1:35:34Speaker 00

  All right. Appreciate it.

• 1:35:35Dwarkesh Patel?

  That was great.

• 1:35:36Speaker 00

  Yeah, I enjoyed it.

• 1:35:37Dwarkesh Patel?

  Yes, me too. Hey, everybody. I hope you enjoyed that episode. If you did, the most helpful thing you can do is just share it with other people who you think might enjoy it. It's also helpful if you leave a rating or a comment on whatever platform you're listening on. If you're interested in sponsoring the podcast, you can reach out at dwarkesh.com/advertise. Otherwise, I'll see you on the next one.