EP #36 Stability AI Founder: Halting AI Training is Hyper-Critical For Our Security w/ Emad Mostaque

In this episode, Emad and Peter hop onto Twitter Spaces to discuss the latest petition to halt AI training, if language models will lead to AGI, and what could happen if the power of AI is abused. You will learn about: 20:08 | Should We Be Slowing The Development Of AI? 44:36 | How Do We Stop People From Abusing The Power Of AI? 59:28 | Is There A Competition With AI Across Economies? Emad Mostaque is the CEO and Founder of Stability AI, a company funding the development of open-source music- and image-generating systems such as Dance Diffusion and Stable Diffusion. Generate images with Stable Diffusion. _____________ I only endorse products and services I personally use. To see what they are, please support this podcast by checking out our sponsor: Levels: Real-time feedback on how diet impacts your health. levels.link/peter _____________ I send weekly emails with the latest insights and trends on today’s and tomorrow’s exponential technologies. Stay ahead of the curve, and sign up now: Tech Blog. _____________ Connect With Peter: Twitter Instagram Youtube Moonshots and Mindsets Learn more about your ad choices. Visit megaphone.fm/adchoices

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

• 0:00Speaker 00

  Buy smart, drive for less and trade for more. That's Toyota affordability. And it starts with a great deal on an all-wheel-drive Camry, Corolla Cross, or Corolla Sedan with the latest features in tech. Or test drive a spacious Grand Highlander or adventure-ready RAV4. And with so many hybrid models available, you can enjoy longer trips without all the stops, while Toyota's legendary resale value means you could get more at trade-in. Get a great deal today and a great value tomorrow. Shop toyota.com for deals and details. Toyota, let's go places.

• 0:32Speaker 01

  Every mile matters. Whether you're driving your dream car or your daily commuter, the Mobile One team knows that. It's why we created Mobile One Advanced Clean Motor Oil. It cleans your engine in one oil change and helps extend engine life, delivering performance, protection, and cleanliness for vehicles of all ages. All to help keep your engine running like new, whatever you're driving. Visit mobileone.com/advancedclean to learn more.

• 1:02Peter Diamandis?

  Everybody, Peter here. I just spent the last hour with Imad Mustak, the CEO of Stability AI, talking about the recent petition that's been going around to halt or pause the development of large language models. Since early April, I just had him on stage last week at Abundant 360, but that wasn't the subject. This conversation was around the fears, the hopes of large language models. And some of the points we discussed was his belief that the next six months are hypercritical, that we have six months to figure things out. Why? There are a couple of very important reasons that he discusses in this next segment. I'm also going to talk about how this technology is going to transform the world of education,

• 1:46Peter Diamandis?

  of healthcare, of supporting governments around the world. One of the best conversations I've had with Imad in a long time. And also, he's one of the most brilliant CEOs, and also the only private AI company CEO to sign that petition. So, listen up. It educated me, got me feeling hopeful. Hope it does the same for you. Just for folks who may not know you, Imad is the CEO, founder of Stability AI, truly one of the most extraordinary AI companies, large language model companies out there with a mission of being an open, a truly open

• 2:28Peter Diamandis?

  AI company to transform, support humanity across health, against education, again about uplifting humanity. So, Ahmad, I love what you do and proud to be a supporter. So, thanks for joining today. It's my pleasure. Yeah. So, let me, let me set it up for everybody. There's been a huge amount of conversation around AGI of late, and I titled this session AGI Threat or Opportunity? Large language models go fast or go slow, open versus closed, and there's a lot to discuss. And, you know, my view on this has been shifting over time, and I'm going to be curious

• 3:08Peter Diamandis?

  about whether yours has as well. You know, there's a lot of talk about artificial general intelligence, you know, where are we really right now in the development? How fast are things truly moving? And should people be concerned about it or excited about it? And other questions are, can it be regulated? Can it be slowed down? Is that possible? And then, I think, an important conversation, really, to highlight what stability is doing, is open versus closed. And then, one final conversation, which is, can we actually get to AGI using large language models, or do we need to have another sort of structural

• 3:51Peter Diamandis?

  transformation to get there? So, shall we begin? Yeah, let's talk about the future of humanity. Let's go. I love that. I love that. And you and I share the same vision of, of really creating abundance around the world, uplifting every man, woman, and child, and really using AI for global education, global health, dealing with the grand challenges and slaying them. I, you know, let's, let's kick it off with.

• 4:24Peter Diamandis?

  How fast are things moving in your, in your opinion? And, and should people be concerned about the speed of progress right now?

• 4:33Speaker 03

  Yeah, I mean, I think every day, those outside of this are like, what's going on? And those inside of this are like, what's going on, right? Uh, it's like everything ever all at once. Not only is it kind of all the smartest people in the world planning in on this, the eighty percent of the research in AI now is in this generative AI foundation model field. You know, and if you look at the research paper graph and archive for ML papers is a literal exponential. It's that people are implementing, exploring and understanding these things like nothing else. The final bit that's really interesting is that clearly half of all code on GitHub now is AI-generated. And the average coder using Copilot, this was a micro study, is forty percent more efficient. So if people are using the AI to build better AI.

• 5:17Speaker 03

  And like, that becomes a bit of an interesting feedback lead, right? Against that, there is the hardware question. And so, um, there was a paper in twenty seventeen, Attention is All You Need, which kind of led to this boom about how to take AI from extrapolating large data sets to instead paying attention to the important bits of information. Just like we do when we base, basically start creating principles or heuristics, right? It turned out that was very amenable to scale. So the original thing on OpenAI and others was just applying scale, more and more compute, which happened literally at the same time as this whole GPU thing that Jemson had been building for years at Nvidia. So the previous generations of supercomputer chips were like GPUs and video ones in particular stacked on top of each other.

• 6:00Speaker 03

  Uh, but as you go to a thousand, two thousand chips, you stopped being able to scale because the information couldn't get back and forth fast enough across these.

• 6:10Speaker 03

  That has been fixed now as of this new generation that's about to hit. The Nvidia H one hundred, the TPU V five. It basically scales almost linearly, uh, to thousands and thousands of chips. Um, to give you an example, the fastest supercomputer in the UK is six hundred and forty chips. You know, NASA has about six hundred as well. And now you're building the equivalent of thirty to a hundred thousand chip supercomputers using this new generation technology because you can stack and scale. It's insane. And we've got about six months before that hits. And then GPT four level models will be available to just about. Well, not anyone, but more and more companies. GPT three, when it was trained, I think, took three months on the new, so that was two years ago. On the new supercomputers, you can train four of those a day.

• 6:56Peter Diamandis?

  That's extraordinary. And so how many large language models are there going to end up being, you think?

• 7:02Speaker 03

  I think there'll only be a few. I think that these things are quite complicated. They're actually like game consoles, like a lot of the chat GPT functionality was already present from the model over a year ago, but it just wasn't in nice format and usable and testable. So you could explore the capabilities of that game console or game engine. The way I put that is because you look at the start of the EU or PlayStation versus the games at the end of that lifecycle. Exploring that space is very important and people don't want dozens. So like we released stable diffusion, our image model, one of the most popular pieces of open source software ever. Only five companies in the world that we now have created their own versions. Because why not just use the one that you have?

• 7:43Peter Diamandis?

  Sure.

• 7:43Speaker 03

  And to put the popularity in context, you're able to Bitcoin and Ethereum in cumulative developer stars in about two months. And the whole ecosystem now has overtaken Linux, which is insane for six months, you know? And yet only three companies built their own, because why would you? So I think in the future, it'll probably be just a handful of companies building proprietary models, you know? Microsoft, OpenAI, Google, DeepMind, Nvidia, a few others maybe. And then with the leaders on the open models in terms of models that you can see the code, the weights, the data, et cetera, which is essential for private data.

• 8:20Peter Diamandis?

  So, pal, the question ultimately is, it's great to have powerful large language models and the next generation that are coming online are amazing. And the question is, do we need more than that? Is GPT four equivalent large language models enough to give us sort of the benefits to humanity without sort of bordering on the existential threats of AGI? And ultimately, the parallel question to that is, are large language models sufficient to get to AGI in your opinion?

• 8:53Speaker 03

  Yeah, so we have to define what AGI is first because different people have different definitions, right? An AI that can do just about anything a human can do, well, you know, half of humans are below average intelligence, right? The, uh, okay mathematicians, I'm not quite correct on that, but still, you know what I mean? The bar for average output is not that high and you see GPT-4 passing the bar exam in G R E. Maybe it'll go to Stanford soon, you know, like it's getting there for those specific things, but it's not agentic as yet. The worry that we have is, you know, as we scale these models, they show emergent properties. Um, and we feed them a crap, the whole internet at the moment. So what we have at the moment are these models that are like really talented grads.

• 9:35Speaker 03

  They occasionally go off their meds and we're fed with crap. And that's the base model. Then what we do is we do this thing called reinforcement learning with human feedback, where we tell it what are good and wrong responses. We take these really creative models, uh, because you're taking petabytes of data. Uh, hundreds of terabytes of data and you're compressing it down to just a few hundred gigabytes, like say when the food took. A hundred thousand gigabytes of images and the output is a two gigabyte file. Of course it can't know everything, but you're telling it you must know these things. So you reduce its freedom, its creativity, because these aren't thinking machines, they're reasoning machines, putting them on principles, right? And then putting that mask on the front to reacclimatize them to human. But you start with a very fragile base and this is the concern on the scaling.

• 10:20Speaker 03

  Because what happens if, as you scale, it learns to lie and other things and it starts being agentic, we don't know. To get to AGI though, in terms of something that can be as capable as a human across a generalized set of principles, but is better at reading manuals. You don't know if this is enough. It may be or may not be. There was a very seminal paper by, uh, sorry.

• 10:40Peter Diamandis?

  Yeah, I was just gonna say the thing that is interesting, of course, is the rate at which, uh. These large language models are surpassing humans' encoding capability. And I guess the question is, will it get recursive and allow them to improve on their own models?

• 10:56Speaker 03

  Yeah, you could say already now, the kind of overall ecosystem, just like Bitcoin provisions humans to buy assets, this whole kind of conceptual thing, it's really made programmers forty percent more efficient, yeah, from that study, if you kind of take it as a human-computer hybrid type of thing. So if you hit the level of a human, that's one thing, and then ASI is when you go superhuman, which we don't really understand what it looks like. There was an interesting paper that mentioned by Meta called Cicero, where eight language models were combined together, and they outperformed humans in the game of diplomacy, so that they could convince humans on things, and that's today. So we don't know where this takeoff point is, where it can get self-recursively better. We don't know when it gets to human level. It's just we can see

• 11:39Speaker 03

  it's as good as an average human, like GPT-4 out of the box without tuning can pass the level three Google programmer exam.

• 11:47Peter Diamandis?

  And it can pass the, you know, the medical licensing exam, you know, something which I never finished doing. I got through part two, but not part three. It's extraordinary, and it's, that's this year. And of course, humans are not getting smarter, but every year we're going to see continued progression of all of these large language models. Can we take a second? I never, you know, I always tend towards the abundance side of the conversation, and I'm clear all the amazing things that these large language models are enabling and will enable. Let's just take a second to list people's concerns. What are people threatened by, just so we can address those individually for a moment. What would be top of your list?

• 12:32Speaker 03

  Irrelevance. So, you know, we talked to a lot of incredibly talented creatives and artists and programmers and others. The top level people love this. There was an MIT study that showed recently the thirty-seven percent, like twenty to thirty percent better writing reports, but the top five percent got way, way better.

• 12:49Peter Diamandis?

  Yeah.

• 12:49Speaker 03

  And this is what we typically see with people and tools, like the top creatives love this. Those who are not so good, it raises the average level, you know, so you have to get better. And that's difficult for some people to embrace. Like Lisa Dahl being beaten by AlphaGo at Go, the average level of the Go player has shot up dramatically, and he's got way better. You know?

• 13:08Peter Diamandis?

  It's not humans versus A I. It's it's human plus A I versus A I in that sense. It's.

• 13:15Speaker 03

  Yeah, or versus humans, right? Like, again, you either have to embrace it, or you can get irrelevant very quickly. Um, and there are certain jobs that are sticky, you know, and there are certain jobs that are not. So regulated industries are quite sticky, but still like something like Indian B P O is probably not for basic programming. And so this is a real danger. It's a real threat. There are questions around doing the data sets being used because they typically are from scrapes of literally the entire internet.

• 13:43Peter Diamandis?

  Yeah.

• 13:43Speaker 03

  What are the rights around that? There are questions around the output. Is it computer generated? Is it copyrightable? You know, and there's finally the question of are we building something more powerful than a new that is an existential threat to humanity?

• 13:57Peter Diamandis?

  Yeah.

• 13:57Speaker 03

  You know, what is this super intelligence like?

• 13:59Peter Diamandis?

  Yeah, I remember Sam Altman wrote a blog and he said, I'm gonna just quote from it. He said, some people in the A I field think the risk of A G I and the successor systems are fictitious. We would be delighted if they turned out to be right, uh, but we're going to operate as if these risks are existential. And so now the question is, if in fact you're operating as if the risks are existential, why are you building potentially existential tech? Right? I mean, that's an interesting conversation debate to have.

• 14:35Speaker 03

  Yeah, I mean, like you mentioned A B C recently, because there are only a few people that can build this tech, right? Um, and he was asked if there's a five percent chance that this tech could wipe out humanity and you could push a button to stop it. Would you do it? And his answer was, I would push a button to slow it down. Uh, we can wind it back. We can turn it off. And that didn't make me feel too comfortable, uh, because my, uh, base assumption, actually, um, I was talking to the head of a trillion dollar company. The founder earlier this week, um, is that. AGI will probably find us boring, just like in Her.

• 15:10Peter Diamandis?

  My favorite, my favorite AI movie, the one that was the most realistic and the least dystopian, Her, that's great.

• 15:17Speaker 03

  Well, you know, broken hearts, you know, just like when Replica turned off their boyfriend-girlfriend feature on Valentine's Day. They broke tens of thousands of hearts. Um, I think that would be the case, you know, I spoke to Elon last week and he said he thinks if we make the AGI curious, then why would it have any reason to harm us, you know, because again, we're interesting in some ways. I don't think we're interesting, but I could be wrong.

• 15:40Peter Diamandis?

  Are you familiar, uh, I'm not with, uh, with Mo Godot, who was COO at, at Google X. He wrote a book called Scary Smart, which I read recently. And his basic thesis is the large language models are, are, are progeny. Where they're learning from reading our content, which they are, and they're learning from how we interact with each other and how we treat our machines. And if we're good parents, and you and I are both parents of, of biological organisms. Uh, if, if they're good parents and we, uh, they see us respecting each other, they'll tend to

• 16:20Peter Diamandis?

  uh, follow suit, but if we are disrespectful and harming each other, they might tend to, at least in the early days, uh, tend in that way. Do you, you know, people don't realize large language models are a reflection of, of human society to a large degree.

• 16:37Speaker 03

  You know, I'd be sympathetic with that view. So like when Web three came about, we tried to create a system outside the existing system and all the money was made and lost at the intersection. With these models, they are literally trained on our collective consciousness on the internet. And it's the equivalent of, you know, taking a super precocious kid, taping his eyes open and showing him everything. No wonder they turn a bit weird at times, right?

• 16:58Peter Diamandis?

  Yeah.

• 16:59Speaker 03

  So I think we should probably be feeding it better stuff and teaching it better stuff. Uh, but right now we're the bulking before cutting points. So that's why we create, you know, you might have seen these memes about the shagoff. You know, this elder weird tentacle being, that's what it comes out of the oven looking like, this deep learning thing.

• 17:15Peter Diamandis?

  Mm hmm.

• 17:15Speaker 03

  And then you have reinforcement learning with human feedback, you tell it how to be human. You know, so the sheveled ground becomes more and more human, but there's still something lurking behind. If you did better stuff, I think is what we've got to do now rather than racing ahead, because again, like, I think we'll probably be fine, but I'm not sure. And these things are becoming more and more capable, because as you scale, you see emergent properties. One hundred starts lying. Yeah. You know, what happens when you've got like, in the G P T four paper, my God, they said, you know, we uh experimented in a closed loop system, giving it some money and telling it to go and make more money, and it was hiring task rabbit people and things like that. I said, that's not a good idea. You don't know what happens there. I don't know what happens. It's trying to create, it's trying to create jobs for us.

• 18:02Speaker 03

  But then again, this is the amazing thing. So uh again, we don't officially publicly know details about G P T four, but Nvidia said they designed their new T the H one hundred and V linked ones, uh chips for it. That's just two eighty gigabyte chips stuck together. It's a hundred sixty gigabytes of V R A. Has G P T four with all its capabilities in, which implies it's about two hundred, two hundred fifty billion parameters model.

• 18:27Speaker 04

  A single flash drive.

• 18:29Speaker 03

  That's crazy. So what does that mean when you start chaining them together and you get them to check each other's outputs and things? Tell you what, it gets a lot better. Uh and there's a lot more that I particularly want to discuss, but I don't know what the upper limit is even at this stage. And this is before, again, this massive ramp up to compute that's coming. Whereby like I know of clusters of H one hundreds. So the fastest supercomputer in the world right now is um, what was it? Uh C, not Sierra, uh there's a one X of flop computer. Our cluster is probably the tenth fastest public cluster in the world right now. Now, the new supercomputers are twenty times faster, forty times faster. Like, there is an insane pickup that then will lead to emergent properties, and this becomes again.

• 19:13Speaker 03

  I don't know what happens. We didn't know what happens. Nobody knows what happens. I think it'll probably be fine. I could be wrong.

• 19:18Peter Diamandis?

  This episode is brought to you by Levels. One of the most important things that I do to try and maintain my peak vitality and longevity is to monitor my blood glucose. More importantly, the foods that I eat and how they peak the glucose levels in my blood. Now, glucose is the fuel that powers your brain. It's really important. High prolonged levels of glucose, what's called hyperglycemia, leads to everything from heart disease to Alzheimer's to sexual dysfunction to diabetes, and it's not good. The challenge is, all of us are different. All of us respond to different foods in different ways. Like, for me, if I eat bananas, it spikes my blood glucose. If I eat grapes, it doesn't.

• 19:59Peter Diamandis?

  If I eat bread by itself, I get this prolonged spike in my blood glucose levels. But if I dip that bread in olive oil, it blunts it. And these are things that I've learned from wearing a continuous glucose monitor and using the Levels app. So Levels is a company that helps you in analyzing what's going on in your body. It's continuous monitoring, 24/7. I wear it all the time. It really helps me to stay on top of the food I eat, remain conscious of the food that I eat, and to understand which foods affect me based upon my physiology and my genetics. You know, on this podcast, I only recommend products and services that I use, that I

• 20:40Peter Diamandis?

  use not only for myself, but my friends and my family, that I think are high-quality and safe and really impact a person's life. So, check it out. Levels.link/peter. Give you two additional months of membership, and it's something that I think everyone should be doing. Eventually, this stuff is going to be in your body, on your body, part of our future of medicine today. It's a product that I think I'm going to be using for the years ahead, and hope you'll consider as well. So let's talk about the other question, which has been the debate over the last week or two, which is, can we slow down development? Should we slow down development? I remember asking a question when I was, god, I was in college. I said, if Einstein

• 21:25Peter Diamandis?

  understood what E equals mc squared would lead to the nuclear bomb, could he have stopped thinking about it? Would he have stopped developing it? Right? So the question that you mentioned earlier about Sam Altman being asked if there's a 5% chance, would you halt it or slow it down? Can we even regulate against this? Because we live in a nation or a world of porous borders and technology very quickly becomes globalized. And we might slow it down, but I don't know if other nations around the world would slow it down. So what do you think's possible here?

• 22:01Speaker 03

  Well, the question is, would you turn it off? Not would you slow it down? And you shifted it to I'd slow it down, which I thought was interesting. There's only a few entities even now that can train these large models. There's literally a handful. And if you look at, again, at that OpenAI AGI paper, they said it's time to offer transparency to governments, transparency on governance. So one of my big issues with OpenAI and DeepMind and others is there's no transparency on the governance of these systems that could overthrow our democracy and that could kill us all. Again, I don't think they will, but I could be wrong. I think that there isn't a slowdown in thinking, but there can be a stop on training before this big ramp up. Because, you know, people came back and said, what if China creates an AGI? And I was like, you know, they don't have the chips, they

• 22:44Speaker 03

  don't have the know-how, you've seen no evidence of that, and they don't want to overthrow their system. They want to perpetuate it. And do you know what the best way for them to get an AGI is? Is to send someone with a flash drive into DeepMind or OpenAI and just download it. Again, it fits on a flash drive. Why would you bother? It's much cheaper to do it that way. So now's the time to put in proper governance, proper oversight. Again, the OpenAI blog on this lists a whole bunch of things that they see in the future. I think the future is now. You should do it now. And you should put proper security systems in place if you think it's that dangerous, so you don't get leaking files going everywhere around these really large models, which I don't think are optimal ways to do things, by the way, but we can discuss that later. And you couldn't do this six months ago, you couldn't do this three months ago, and in six months it will be too late. So now's

• 23:29Speaker 03

  the time for a public debate about this, about systems that could take away our freedom, that could kill us all, and everything in between.

• 23:38Peter Diamandis?

  Did you actually sign that petition that was being circulated last week?

• 23:43Speaker 03

  I believe I was the only private company CEO in AI apart from Elon to sign that, yes.

• 23:51Peter Diamandis?

  Amazing.

• 23:51Speaker 03

  I didn't agree with everything in there, but again, I think now is the time.

• 23:56Peter Diamandis?

  And is a six-month slowdown the right thing? How do you define that versus a pause?

• 24:03Speaker 03

  So you're never going to get a pause because of competitive pressures. And for me, six months is the amount of time you need to get used to these new next-generation systems that are landing that can scale almost infinitely. So I was like, eh, it's kind of the right thing. And again, you know, even six months ago, that was when stable diffusion first came out. It's been what, three, four months since ChatGPT first came out? Like, you needed the permutation around the internet and you're seeing things now like ChatGPT being banned in Italy. Right? You're seeing the haves and have-nots, and you're seeing competitive pressure where this is now big business. It's affecting trillion dollar companies. And it's not going to stop. So how's the time to start discussing this stuff? I remember when- In public governments.

• 24:44Peter Diamandis?

  I remember when Uber was banned in France and I was like, okay, we're going to start to label countries as pro and anti-technology. I mean, it's kind of insane to think that you can ban these types of technologies and still remain competitive on a national or global stage. Can we flip it over one second? The endpoint of all of this technology is to make us, in one sense, superhuman, to allow anyone to be educated, to be a creative agent, to be extra healthy, adding decades on to one's life. You know, is the goal here of this level of AI development

• 25:27Peter Diamandis?

  to allow us to work less, to be more productive, to create abundance on the planet? What's your vision of where we're going here?

• 25:36Speaker 03

  I think everyone that's building this, maybe if there's some defense community exceptions, is actually doing this with good intentions, right? Because they want to create something that can help humanity. Now, some people want to automate humanity. And, you know, like, literally, you look at some of the statements of some of these leads, and they're like, well, you know, this will transform money. And then we can redistribute it. And I'm like, oh, and you're in charge of that. Other people are like, we want to augment humanity. So I'm very much in the augmentation camp, where I believe humans and computers be anything. And I'm like, this allows us to scale humans. We have solved that problem. It's still early stages with the iPhone 3G stage, but any child in the world, soon in the next few years, will be able to have their own personalized tutor or personalized

• 26:21Speaker 03

  doctor. And then this allows us to scale society as well, because the Gutenberg Crest was an amazing thing, because it allowed us to take stories down, and we're driven by stories. But it's lossy. You know, all those notes you're taking in your meetings and everything like that are lossy. Now, the new systems, you look at the new office and Teams and all the hostages that are using that, you know, you fill in the gaps. It's no longer lossy. It automatically writes your emails and summarizes stuff for you. It can be that grad, and he's generally on his meds. It's going to be pretty awesome, I think, and a real changer to the way that we all collaborate and achieve our potential. But if it's centralized, and that's the only option, then it probably won't go so well. I think we've had a history of that, you know? What happens if new technology is controlled by a few hands?

• 27:06Speaker 03

  Power is really attractive, really, really attractive.

• 27:10Peter Diamandis?

  Well, and this is the power that's going to drive to a new global set of trillionaires, and I think a complete transformation of every industry. So let's talk about open versus closed. You very much, well, let's begin back now eight-plus years ago, when Sam Altman and Elon talk about OpenAI being truly open, with a mission to help guide the development of AI for humanities. And it began as a nonprofit, and then, you know, Elon put a hundred million in, and he's pissed right now. It's turned into a for-profit and is anything but open.

• 27:50Peter Diamandis?

  And I love the fact that if you go to open.ai, it doesn't point at OpenAI, it points towards stability AI.

• 28:00Peter Diamandis?

  Which is very, which is very interesting. But you made a, you made a conscious choice to build an open platform. Can you speak about that?

• 28:08Speaker 03

  Yeah, I thought that ethically it's the correct thing to do and it was a better business as well because the value in the world is in private data and knowledge and people want control. So again, if these are grads that are continuing off their meds, using OpenAI or Google or Anthropic or anyone like that is hiring it from McKinsey, but you want to have your own ones. But then I think this technology had to be distributed. Like when I came in, as a relative outsider, I've only ever been to San Francisco once before October. And I have a non-teenational background. Um, I talked to a lot of AI ethicists who were like, well, we have to control this technology and has access to it. Until it's safe. And I was like, when will it ever be safe for Indians or Africans?

• 28:45Peter Diamandis?

  Hmm.

• 28:46Speaker 03

  I don't think it ever will be. Um, and it reminded me a lot of the old colonial mindset, you know, and access to technology. I was like, if this can really activate every child's potential, you know, and, you know, working with the X Prize for Learning winners. And deploying this with ImagineWorldwide.org, my co-founder's charity into refugee camps. I was like, how can I hold it back from these kids? You know?

• 29:07Peter Diamandis?

  Yeah, but by the way, just for, just for those who don't know, uh, many years ago, Elon Musk and Tony Robbins funded something called the Global Learning X Prize. It was a fifteen million dollar prize, asking teams to build software that could teach children, in this case in Tanzania, reading, writing and arithmetic on their own. And, uh, Imad was, uh, was a partner in one of the winning teams. And you're now taking it much further, much faster, um, which is amazing.

• 29:36Speaker 03

  Yeah, we did the RCT, the Randomized Control Trials with UNESCO and others. And so in refugee camps around the world, seventy-six percent of children get to listen, you receive. In thirteen months and just one hour a day with this basic software. What if they all had a chat GPT? It will change the world, right? And so we figured out how to scale that to every child in Malawi and then across Africa and Asia. In the next few years with the World Bank and others. And again, like, it was interesting again to have this discussion because things are very insular when you have power and control. And it's very tempting to keep power and control. And this is, you know, one of the things of how do you view the world? Are we stronger together and we can deal with any problem? Or do I need to be in the lead and control it if it's a powerful technology? Because people are fundamentally good or bad.

• 30:17Peter Diamandis?

  So first of that's a scary thought. I mean, I fundamentally believe that humans are good by their basic nature. And that the majority of humans are good and want to make the world a better place. But we do have to guard against those that have sort of a bent arrow. Yeah.

• 30:38Speaker 03

  And that's why I look at what does the world's infrastructure run on? It runs on Linux. It runs on open source, MySQL. That's the most resilient. Windows is not resilient. Which is a very interesting analogy, I think.

• 30:53Peter Diamandis?

  So how do you make money as an open source company?

• 30:55Speaker 03

  Oh, it's pretty straightforward. Every nation wants to have their own models because this is a question of sovereignty. So we're helping multiple nations do that. And then every single regulated industry wants their own models. So we're helping them do that too. Working with cloud providers, system integrators and others, lots of announcements to come. Private data is valuable, public data is less valuable and a bit of a race to the bottom. So I think people are looking on the wrong side of the firewall. And again, we'll see that going forward, but they're not going to stop. Microsoft, Google, Nvidia, others aren't going to stop. They'll keep scaling. And now that there's actually revenue that can be generated from these models, you're going to move from research to revenue. Hundreds of millions, billions will be spent on training. And again, I think the scaling paradigm is one of them where you scale these things up.

• 31:39Speaker 03

  I think it's a correct one to get us to, let's say, a level of AGI. So major that I'm interested in is the human colossus. Humans coming together to solve the world's problems or meta technology. I think the collective and swarm intelligence will beat an AGI any day. And that's what I'm interested in. I think arming everyone with their own AIs, every single person, company, country, culture in the world. We build a new intelligent internet that can solve any problem. Because a lot of our problems are information and coordination and this technology enables that. So a very different view from most of the other CEOs of companies in this space. Global, collective, augmented, as opposed to artificial general intelligence that can replace humans, shall we say.

• 32:21Peter Diamandis?

  Let me just make a mention, we'll be going to questions from from those watching and listening, from those listening in a few minutes. So, Imad, you know, is GPT-4 and its equivalents large enough for us to achieve what we want? Do we need to go to yet another generation? You know, can we, can we build enough intelligence without tipping over to AGI?

• 32:50Speaker 03

  Nobody knows. Nobody, nobody will ever be able to tell you that there's less than 12 months to AGI because you don't know until it gets there, when it gets gigantic.

• 32:58Peter Diamandis?

  I, I, I used to hate the, all the doomsayers on AGI. My, my standard statement was always, listen, I'm not worried about artificial intelligence. I'm worried about human stupidity and to a large degree that remains, that remains the same. Right. And what we're going to see with all of, with ChatGPT and all of these large language models are people's stupidity being amplified and people's brilliance being amplified as well. And so the question is, does it go out of bounds on the low side to cause us any real issues?

• 33:29Speaker 03

  No, it's a dangerous point right now because you're not quite intelligent enough to align and you have amplification of these things where you, people have powerful tools, right? I think when it does get intelligent enough, I think there's no way that we can perfectly align it. Because alignment is orthogonal to freedom. We all know people more capable than us and the only way to make sure they do exactly as we want is take away their freedom. I don't think a super AGI will like that necessarily. But again, I think again, we're kind of boring. So I think that we are on this dangerous phase now between those two things. When it can't be self-correcting, I'm telling you, no, you know, Peter, I'm not stupid. Why are you going to do that? You know, don't be a dick. Um, and you know, these are dangerous, powerful technologies that can be downloaded on a flash drive.

• 34:12Speaker 03

  You know, again, why would you bother training yourself when you can just take it? And use it in really original ways. Again, I don't want to get into too much detail, but it's probably not just deep learning. We've already implemented reinforcement learning and there's all sorts of ways that you can create much better systems. Like the simplest one, just get two GPT-4s to check each other's answers and it becomes better.

• 34:32Peter Diamandis?

  Yes.

• 34:32Speaker 03

  Right? Surprising that.

• 34:35Peter Diamandis?

  You know, I have to say, having you as the CEO makes me feel safer in terms of at least for stability AI. I'm, I feel you have a mission to build a strong and viable company while uplifting humanity and doing good in the world, which is important. I think that has to be at the core for all of these companies because the technology is so fundamentally powerful.

• 35:03Speaker 03

  I think realistically, Sound has an amazing mission, Demis has an amazing mission, Dario has an amazing mission. Like, everyone is quite mission-driven, who are the CEOs of the companies that could build this type of technology. It's just, none of us should be in control of it. I don't know how it should be, but I think this is the time for a public debate around it, and, you know, everyone to realize that this is taking off crazily. And that's what democracy is about, you know?

• 35:34Peter Diamandis?

  I remember when I was in medical school years ago, restriction enzymes were first coming online, and there was a huge amount of fear-mongering about designer babies and being able to clone children, and it was, you know, we're playing with god-like powers and we're going to lead to killer viruses. This is in the late 80s, early 90s. And there was concern about regulating and could this even be regulated because you could sneak out restriction enzymes in, you know, a few picoliters of a fluid. And what came together was a series of summits called the Asilomar Conferences, in which

• 36:15Peter Diamandis?

  the heads of the industries got together and self-regulated versus the government coming down on them. Have you seen that conversation taking place here at all?

• 36:25Speaker 03

  I think it's time to emerge now because we need to self-regulate as an industry. We need to increase transparency of our governance, you know, and how we make decisions. Like last year, OpenAI with DALL-E two forbade every Ukrainian from using it. And if you typed in a Ukrainian term, it threatened to ban you. Why did they make that decision? Who knows? Is there any redress? Nope, you know? And so we have to be really mindful about things like this and set industry norms. And now I think it's time to come together to people raising, this is a big deal. I think we want to do it before it's forced upon us. But at the same time, the discussion needs to be widened here because this affects every part of life. It's kind of insane. We've never seen anything like this before, right? And we have to do it quick, quick, quick.

• 37:06Speaker 03

  And this is the last window that we have before, again, everything takes off.

• 37:11Peter Diamandis?

  So that's, that is one of, that's, I remember when we first met, I asked you, is this time different? Because we've been hearing about this conversation for forever. And so you're pretty clear definitively, this time it is fundamentally different.

• 37:27Speaker 03

  Dude, I mean, like any coder that uses GPT for us, like, wow, look at that. I'm like multiple times better, right? It's like this thing can write better than me, it can draw better than me, it can paint better than me. Like, I, I think people don't want to believe that things are different. Um, I recently went and said to talks, this is a big impact on the pandemic because all information system flow changes. Like things that involve atoms don't change, but so much of our life is about information. And this has made a meaningful difference in that. Just play with it, try it, go depth, and you can't go away with any other answer, I think.

• 38:01Peter Diamandis?

  Let me, let me sort of go down a few of these. Uh, we've talked about education, uh, the potential there. If you would, five years from now, what's your vision of the potential for generative AI in global education?

• 38:17Speaker 03

  Every child has their own AI that looks out for them and customizes them. Are you an auditory learner, a visual learner? Uh, are you dyslexic? That's all information flow and you can fix all of that, right? And it brings students together to work together as well. Cause most of school right now is a childcare system mixed with a status game. Yeah, that's why I think, Peter, we were having discussions before. An incredibly smart young kid asked a question, you know? Uh, my school says that it's cheating to use ChatGPT. And I'm like, it's not a competition education, right? We made it a competition. Education is about actuation. And so I think these things will fundamentally change the way when everyone's got their own. Amazing teacher, the young laser illustrated pioneer from the diamond age. How exciting is that?

• 39:01Speaker 03

  And a hundred percent, we can get there in five years for every kid in the world. Or just about every kid. That's amazing. That will change the world.

• 39:08Peter Diamandis?

  For the cost of fundamentally electricity, uh, which is getting cheaper all the time.

• 39:13Speaker 03

  Yeah, so let's put it this way. GPT four, if it does run into H one hundreds, as Nvidia has indicated, is fourteen hundred watts. The human brain is twenty five watts. How crazy is that?

• 39:24Peter Diamandis?

  A lot of.

• 39:24Speaker 03

  That's a forward sensation. Yes, especially when we start feeding it junk.

• 39:30Peter Diamandis?

  All right, let's go to health next. Uh, you and I are both passionate about that. You've done a lot in your, in your life in the health area. And I'm clear, I mean, one of the beautiful things is that all eight billion people on the planet are biologically identical. So something that works for someone in India or Iraq or Mexico is the same that works for you in San Francisco. What's your vision there?

• 39:53Speaker 03

  Yeah, people are people, just like education. If you have open source education that recursively goes, healthcare is the same. We're attempting to increase the information density and organize knowledge. So when my son was diagnosed with autism, they said there's no cure, no treatment. I had to build an AI system to analyze all the medical clinical trials and do a first principles analysis of what could cause it. And then we did drug repurposing to ameliorate the symptoms. That should be available to everyone. You know, anyone who's had a condition which is difficult. The process of finding knowledge is so complex. You know, and so our system is a Gerdic. It treats a thousand tosses of a coin, the same as a thousand coins tossed in a row, right? Like, uh, essentially the population has a cytokerine P four fifty mutation. It means you metabolize things quicker.

• 40:36Speaker 03

  So your more, your codeine becomes morphine. Those are the ones that die of fentanyl. How do we not even know that, right? And so we have all the tools in place now to have your own personal doctor, you know, to have all the knowledge in medicine aligned so we can see across and have that information set. And I think as we do this with longevity, with cancer, with Alzheimer's, within 10 years, we'll be able to cure all of these. Because the knowledge is all there, it's just not all in the right place. So everyone's trying to do their own things. The knowledge on clinical trials, the information leakage, it's ridiculous. We can capture every piece of information on that.

• 41:13Peter Diamandis?

  Yeah, people don't realize the, all of the failures in these trials is extraordinarily valuable information that is not captured and not utilized at all. The other thing, just to hammer it home, because those of you who know my work, I'm focused on longevity, age reversal, how do we make us live longer, healthier lives? This decade is different and we're going to be understanding why certain people age, how to slow it, how to stop it. A lot of it's going to come out of the work that companies like Stable, Stability and others are doing. Then we have quantum technologies coming soon, which are going to put sort of fuel on the fire here.

• 41:59Peter Diamandis?

  My guess is, what, within five years or less, it'll be malpractice to diagnose without AI in the loop?

• 42:07Speaker 03

  Yeah, I think it'll be humans trust AI. It won't be AI that diagnoses. Again, show pilots for everything is going to be the way I think. Just like self-driving cars, I think are imminent in terms of getting to that level four, level five. And I think, you know, as five, 10 years, it's going to, you have to really quid pro quo to drive on the road. You know, AI is better. You have an army of graduates that will become associates, that will pass the bar, you know, that will pass their medical exams over the coming years. And you can infinitely replicate them.

• 42:39Peter Diamandis?

  Yes, what, what people don't realize is every time an AI learns something, it shares it with every other AI out there as you update the models. And AIs are getting better every year while we humans, pretty plateau after a certain point. The other thing I love is.

• 42:55Speaker 03

  Yeah, I think the thing I need to really drive home here is when you're trying ChatGPT and GPT-4, that's just one of them. Imagine if there was a hundred of them checking each other.

• 43:05Speaker 03

  No doubt the outputs would be even better than you've seen and three, ten of them were learning about everything about you. So right now, we're like, single. Soon we'll be parallelizing these things. And just like, if you had a hundred brands that were exceptional, you know, and generally good, all working around you, and you wouldn't have to like oversee them, your life would be better.

• 43:29Peter Diamandis?

  Uh, when we were on stage at, at Abundance 360, we were talking about, uh, interesting subject of who would you hire into your company and what background would you require if you're hiring someone into your company? And what is the most important skill that people will need in this world of exponentially growing AI? And, and your answer was passion all the way around.

• 43:54Speaker 03

  I mean, like this, we've had fifteen year olds contributing to open source code based sixty year olds, right? Because you don't know where it comes from. But you have to be passionate and throw yourself into this because it's going so fast. If you're not passionate, you aren't going to keep up. And then you bring the latest breakthroughs to your company and kind of communicate it. And you can use the technology to communicate it even better, which is kind of awesome. It will automatically create the presentations and slides for you. So I think passion is kind of the key thing, plus, you know, a level of intelligence and capability. But if people aren't passionate, they're not going to be able to keep up here. And that's the key thing.

• 44:28Peter Diamandis?

  Hey, everybody. This is Peter. A quick break from the episode. I'm a firm believer that science and technology and how entrepreneurs can change the world is the only real news out there worth consuming. I don't watch the crisis news network. I call CNN or Fox and hear every devastating piece of news on the planet. I spend my time training my neural net, the way I see the world, by looking at the incredible breakthroughs in science and technology, how entrepreneurs are solving the world's grand challenges, what the breakthroughs are in longevity, how exponential technologies are transforming our world. So twice a week, I put out a blog. One blog is looking at the future of longevity, age reversal, biotech,

• 45:13Peter Diamandis?

  increasing your health span. The other blog looks at exponential technologies, AI.

• 45:19Peter Diamandis?

  3D printing, synthetic biology, AR, VR, blockchain. These technologies are transforming what you as an entrepreneur can do. If this is the kind of news you want to learn about and shape your neural nets with, go to diamandis.com/blog and learn more. Now, back to the episode. Listen, I would love to open it up for some questions if you're open for that.

• 45:41Speaker 05

  Yes, hi, I'm Amir Awadallah. Hey, Ahmad, how are you? First question, Ahmad, is how's your noggin doing?

• 45:48Speaker 03

  That's good, man. I had a bit of a fender bender, so my brain was jiggled for the last few days. I think it's back to normal now.

• 45:55Speaker 05

  Okay, please make sure to go check on it and make sure it's all in order because we need those neurons continuing to fire for a few more years to come. So this is my question to you, and I'm very curious about your thoughts on this one. We are clearly on a trajectory now where we're going to have Cambridge Analytica multiplied by a billion times. Assuming you're familiar with Cambridge Analytica and what they did, right?

• 46:20Speaker 03

  Yeah, yeah.

• 46:21Speaker 05

  So now imagine on a website like Character AI where I can go and create characters that are digital replicas of people, like there is a digital replica of Elon Musk. And then if it was modeled very correctly, then I can now start playing arguments against it, just like AlphaGo played itself. And I would play it against itself for billions and billions and billions of iterations until I find the perfect way to get them to do whatever I do. So it's Alpha Subjugate. I can subjugate any citizen to my will. And you said earlier, it's very tempting to keep power and control. How are we going to react? How are we going to solve that? That's a very big problem, in my opinion. I'm curious your thoughts on it. Like, how do we stop that from happening? That's going to happen.

• 47:05Speaker 03

  No, it's inevitable, right? And again, it's like it's here pretty much now. Creative ways of using this for mass persuasion are there. Like, you really get robo-callers that call up with the voice of your grandmother saying it's an emergency now. And you can't tell the difference. Like, this is going massive, and voice is very, very convincing. So I think the only way you could...

• 47:30Peter Diamandis?

  Good, good, Johan.

• 47:31Speaker 03

  I thi- I think the only way you can do this is you have to build your own AI to protect you. I need to earn those AIs and they need to look out for you, honestly. And then trust the authentication schemes, you know, the Twitter tick on steroids. Like, we have to move fast, fast, fast. We're going to use armies of replicants are coming here as well as being able to gamify that particular thing. Uh, we shouldn't talk too much about it though, because yeah.

• 47:55Speaker 05

  So countermeasures. So you're saying it's countermeasures? Right? It's just like antivirus, virus, antivirus. So we need something that is countermeasure at the speed that is catches this before it becomes much worse. So is that something that's any of us are working on? Should we unify efforts to work on that?

• 48:13Speaker 03

  Yes, uh, but we've got some things in that space.

• 48:16Peter Diamandis?

  Emerald.

• 48:17Speaker 03

  It's, uh, yeah, idea, idea, idea antivirus, there we go.

• 48:20Peter Diamandis?

  So to close out that, you know, I'm fond of seeing the world's biggest problems, the world's biggest business opportunities. And believe me, as that issue comes online, there will be multiple entrepreneurs looking to slay it.

• 48:33Speaker 06

  Omid, with Stability AI, one of the initiatives is, uh, Open Bio ML. So I'm wondering if you could speak about, you know, best case, where is that going? Where are we going to be in five years? If sort of your best case vision turns out for that, what, what will that look like?

• 48:48Speaker 03

  Yes, Open Bio ML is our medical competition biology one. So we're on the main backers of Openfold, um, to do kind of upperfold type things with a quest, DNA diffusion, to kind of see how DNA folds, uh, bio LM, see chemical reactions, uh, based on language models. Um, there's a whole bunch of things that I just think the opportunity is massive because again, the data quality is poor, but now we can create synthetic data and we can analyze data better than anything. We can figure out how chemicals interact. The work of DeepMind and our fall was amazing. And now we can do things in silica. So we can test chemical reactions, drug interactions and others that when combined with a knowledge based on all the medical knowledge in the world and research, and able to extract patterns from it that are superhuman.

• 49:33Speaker 03

  I don't think there's anything we can't cure, honestly. But we've got to build this as a public, common good, in my opinion. And give these tools out to the experts to use and unify. We have to do this very intentionally. So Open Vinyl is just at the start now. And we also have Medarc, which is the healthcare equivalent of that, doing like synthetic radiology in Stanford MRI to create more data sets of rare lung diseases and things like that. And it's working great. So I think community, commons is the way.

• 50:01Peter Diamandis?

  And by the way, all of this is how we create an abundance of health around the planet, right? The best diagnosticians on the planet are going to be AIs and the best surgeons will be robots driven by AIs and that then becomes available everywhere. And I just want to hit on a point that I think is important for folks to hear. A world in which children have access to the best education and the best healthcare is a world that is more peaceful, which I think is one of our key objectives here. How do we uplift every man, woman and child on the planet?

• 50:35Speaker 07

  I wanted to ask a personal question here and it's something I haven't had a chance to ask Peter yet about AI. And Peter, I know that you and seemingly Ahmad generally have an abundance and optimistic mindset. I understand your views are changing. Is the potential threat of catastrophic job loss concerning to you both? And if so, how do you potentially suggest people address finding meaning in their lives without consistent work or careers? I'd love to hear your thoughts, Peter, as I know you're typically really focused on this from an abundance mindset and also yours, Ahmad.

• 51:13Speaker 03

  Well, Peter, you're fast.

• 51:15Peter Diamandis?

  So, yeah, so listen, I think we're lucky, most all of us here taking the time to listen to this conversation are extraordinarily lucky. We're doing a job, I don't know what everybody's doing, but I'm pretty much guessing we're doing jobs that we love, that we dream about, that we're excited about. The majority of humans on the planet unfortunately are not doing what they love. They're doing what they need to put food on the table, get insurance for their family. So one of the things- I think about is how do you use these extraordinarily powerful technologies to self educate? And to self empower to go and do the things that you love to augment yourself to have a co pilot. If you want to be a physician, a teacher, a writer, whatever it might be.

• 51:59Peter Diamandis?

  Um, and that trains you on the job as you're doing the job. Uh, so it's going to allow people to dream a lot more. You know, I think the only fear I have about AI today is what I would call it's. Uh, it's toddler into adolescence phase. I think, you know, when AI is in its earliest AGI state, if we get there. Um, before it's developed, uh, fundamentally sort of an ethical emotional capability. I think the more intelligent in general that systems are, the more empathic and uh.

• 52:41Peter Diamandis?

  The more good nature there will be, I don't think, I don't, we live in a universe of infinite resources. I don't buy any of the dystopian Hollywood movies where they have to come and. Grab all of the whatever off the planet earth, that's bullshit, there's so much, you know, it's an abundance of everything. I think we just have to deal with the early stages where humans are using it in a dystopian fashion. Or the toddler doesn't know its own strength.

• 53:05Speaker 03

  Yeah, I think that, I think that people will create new jobs and they'll do it very fast, I think. The pace will be picked up by a mixture of open and closed technologies as well. You see the innovation around stable diffusion, new language models coming out and other things as well. But I think I'm really excited about emerging markets in particular. I think they'll leap to intelligence augmentation, just like they leapt over the PC to the mobile. You know, and that will create ridiculous abundance and value there because they embrace technology. Because they need to and they want to, and there's amazingly people there that just haven't had the opportunity will now have the opportunity with this. So I think, you know, um, I think it's a bigger economic disruption than the pandemic. I don't know which way, but I believe it will be positive, to be honest.

• 53:48Speaker 03

  And you'll see points added to the GDP of India and other countries once they get going.

• 53:54Speaker 08

  This brought up a lot of questions, really great conversation. The one that I think is most pertinent to me is after hearing Yumad talk, and I think Peter made the comment, he's thankful that you are one of the leaders of this and leading your company because you clearly have good intentions. You sound like a giver in the Adam Grant give and take sense. How do we control against the other players that have their fingers on the proverbial button? Right? Because this like if you have people with really bad motivations and takers versus imagine Trump was smart enough to be a CEO of one of these companies. I mean, that is terrifying.

• 54:38Peter Diamandis?

  I prefer not to.

• 54:39Speaker 08

  How do we, how do we control against that? And part of, part of, I'll answer part of that. I think I'm an investor, an early stage investor. I think it is incumbent upon investors to be picking good actors versus bad actors and who they back. So anyway, I'm, this is a really concerning note because I'm glad that you're one of the good guys, but I'm sure there are also bad guys.

• 55:02Speaker 03

  I think actually the most dangerous are the good guys in some ways. Like most of the evil in the world is done by people who think they're doing good. Not, not me, but you know, you never know. I think that the key thing here is transparency. If you are using powerful technology that has the ability to impact the freedom of others, you need to be transparent and you need to have proper governance and other things. Like, you know, we build out in the open and I think that there needs to be some mechanisms for that. Again, opening, I listed a whole bunch of them in a blog post without saying whenever they do it, we need to implement that now.

• 55:36Peter Diamandis?

  So that's, that's really important. You know, it's not just being open or closed. It's really being transparent and having an understanding of how the technology is being developed and utilized. And then having investors and governments. Listen, I never depend on governments for much of anything. But, but you know, it's not bad to have a set of requirements that society holds your feet to the fire on.

• 56:05Speaker 03

  And this is the last time we can do that, in my opinion. This next period of six to 12 months.

• 56:10Peter Diamandis?

  That is the single most important thing I've heard said, Imad, putting that timeframe on it because of the clusters of H100s and the new capability that we're adding. And what most people don't realize is that we're in this situation today with these large language models and deep learning because we've seen massive growth in computation over the last just few years and massive amount of labeled data out there over the last few years. The ideas for deep learning have been around for 30-plus years, right? But it's just now that it's capable and it's adding fuel to the fire. So, yes, everybody just listen up. The timeframe is now.

• 56:52Speaker 08

  But just a quick note, like, how do you make that a forcing function other than just a letter? I mean, the letter, that's great, but is that really leverage? I mean, what's going to cause that? And what happens if they choose not to and we don't pause for the next six months?

• 57:08Speaker 03

  I don't think there'll be a pause. I think they'll continue, right? Like, there's nothing enforceable within that period. But I think that they will feel more and more pressure to come and build industry standards, and I think there will be policy responses, literally like we've just seen in Italy. ChatGPT is now banned in Italy. And on the flip side, you don't have ChatGPT in Saudi Arabia. Why? Because OpenAI decided not to. So we need to have some standards around this, sooner or later, and it'll be a mixture of public pressure, government pressure, investor pressure, and more, I think. It's not easy, man. It's not easy.

• 57:41Peter Diamandis?

  And then, Imad, one of the things I truly hope is that your voice, and one of the reasons I wanted to do this Twitter space conversation with you, is I really want to hear your voice in this world of AI. I know how brilliant you are. I know how giving your heart is, and I want people to... I want, you know, we keep on hearing from Sam and Elon and others like Jeffrey Hinton and folks at DeepMind, but I'm excited to hear you lead in this industry.

• 58:14Speaker 03

  General best, I think, again, everyone needs to speak up now. It's the time.

• 58:18Speaker 09

  Hello. Thank you so much for having me, Peter Imad. Just had a question around community. And I was wondering, you know, with all this kind of stuff happening with automation and things are going to change, do you think if we can like start leaning more into humans finding purpose in in-person events? I just want to hear your thoughts on that. Like, what role does in-person events have in this, you know, new future? I love AI.

• 58:44Speaker 03

  I think we're pro-social. No, I think we're pro-social species, right? And AI can help us connect better with people around us. So, you know, I mean, we all come out with this COVID weirdness, and now I'm getting used to meeting people in person. I think it's super important. You know, I think there was an interesting thing as well, like the FS study in Argentina, where rather than giving direct cash handouts, they actually said, universal basic jobs, pick your jobs for your community. And people loved it because they have purpose, they have meaning, you know, and it brought loads of women into the workforce. And they're like, don't give us back to direct cash handouts. People like being with people fundamentally. And so we've got to use this technology to increase engagement with other humans rather than that Wally-type

• 59:26Speaker 03

  future of a fat guy with a VR headset, you know, just being waddled around. I think people like stories and stories when they're together. This will allow us to tell better stories.

• 59:34Peter Diamandis?

  Yeah, up until now, connecting and building community has typically been almost a random process. If you happen to be in the right place, if you happen to read the right thing, or signed into the right community. Imagine systems that are able to proactively gather people who are

• 59:56Peter Diamandis?

  Great matches and didn't know that they are. So I think we're going to see a lot of things accelerate in the directions that we choose. So what do we want? And be careful what you ask for.

• 1:00:07Speaker 03

  Again, we're moving from the book and the Gutenberg face, which is massively lossy, to being able to capture some of the complexities of humanity. Like, you know, when you watch Moana, why doesn't Matty pick up the sun with a fishing hook from the ocean? It's because it's all they ever need. And we can start to capture these stories of the world so we can better understand each other and engage, or the vice versa. This is up to us now.

• 1:00:27Speaker 10

  I have two very quick ones, and pick your poison. You can answer one or both of them. But one is on the pause and development. So what do you think the implications are for the economic competition between countries? And like, should we even be looking at it this way? And then the other question is that the conversation is really sharply focused on AI right now. But I think the real power is going to be in the converging of different technologies. So based on what you're seeing, what do you guys think is going to be the next big tech to reach mass adoption, the way that we've seen with AI in recent months? That's really going to accelerate this sort of convergence with AI and create really interesting things.

• 1:01:11Speaker 03

  Cool. So on the first question, I'd say there's only two companies in the world that are actually basically building this AI, and that's the UK and the United States. Like, you're not seeing GPT-4 level models in any other countries. And so the pause was basically those two countries. On the second one, I'm not sure. I'm going to put that to Peter, because Peter's got a much wider view than me.

• 1:01:31Peter Diamandis?

  So what's interesting is, A, I think about something called user interface moments. When Marc Andreessen created the web browser, it became a user interface on top of ARPANET. And it made it accessible and usable, right? We can see throughout time, even the App Store as a user interface moment. The ultimate user interface moment is, in fact, AI. Uh. So if you don't know how to 3D print and you don't know how to use any graphic programs, but you know how to describe what your intention and desire is, you could speak to your own version of Jarvis. And I think all of us are gonna have our own version of Jarvis,

• 1:02:16Peter Diamandis?

  our own personal AI that we've hyper customized, that we give permission to know everything in our life because it makes our life better. And you can say, listen, I'd like to create a device that's got a handle on it that looks like this. You describe it physically and then your AR glasses or VR glasses, you're seeing it come together, being shaped as you describe it, right? That technology is here today. And then you say, that's it. Print. Uh. And then it says, great. And then you say, add it to the store and it's available for anybody for free or for a penny. So AI is going to be the ultimate user interface for all exponential technologies from computation, sensors, networks, robotics, 3D printing, synthetic biology, AR, VR, blockchain.

• 1:02:59Peter Diamandis?

  And that's when it starts to become interesting because it used to be that, uh, you needed very specialized knowledge. Now, what you, the knowledge you need is what's in your mind, what's your intention, what's your desire? Uh, and AI becomes your, uh, your partner in implementation, going from minds and materialization, if you think.

• 1:03:20Speaker 04

  I wanted to get Imad's and Peter's thoughts on something called the factorial paradox. It's, uh, it's in the game factorial where once you learn to automate everything and build factories that are smarter than your factories, you would think that the work for human decreases, but actually your work increases because now you have the whole map to expand to. And I was wondering if there is a possibility that this might also happens with the adoption of AI tools, that human labor might actually become even more expensive and more rare. As we've kind of seen now, we've used, I mean, we've been using Codex has been out for a year. After code, as you said, Imad, now it's generated by AI,

• 1:04:02Speaker 04

  yet we're still finding, uh, it hard to, to hire people on the, on stuff. So what do you think, what do you guys think about that?

• 1:04:11Speaker 03

  I think that's great, and I've spent thousands of hours on Pactoria, so I'm very familiar with it. And I think that's the thing, human discernment is still very important, and human guidance is still very important. It's just as we build new technologies, it lifts us up, right? And it adjusts where our point in that cycle is. So, you know, I think it will replace all, like I said, well, there's no more coders. There's no more coders as we see them right now. When I started coding, uh, what was it, 20 years ago, when I was like, uh, 22 years ago, gosh, when I was 17. We didn't have GitHub or that stuff. We started using subversion that had just come out. You kids have it easy these days, you know, those who are listening, with all this kind of thing. So we get better and better abstractions of kind of knowledge, and the role of the human changes.

• 1:04:53Speaker 03

  And it becomes more valuable, because you can do so much more and enhances our capabilities. I think, like I said, that's the exciting thing, rather than, oh my god, no one's got anything to do, they're just there at home getting fat and watching their own customized movies, right?

• 1:05:06Peter Diamandis?

  You know, uh, there's another part of the conversation we haven't discussed, which is the coming singularity. Whether you believe it or not, I, you know, feel we are moving very rapidly in that direction. Ray Kurzweil described it, actually Werner Vinge described it first. And Ray projects it to be some 20, 25 years out. The point at which the speed of innovation is moving so rapidly, you can't project what's coming next. Um, and there's another book that I love called Zero Marginal Society by Jeremy Rifkin. Talks about what happens when we have AGI and we've got nanotechnology and we've got abundant energy, right? Everything becomes possible pretty much all at once, everywhere.

• 1:05:50Peter Diamandis?

  And we were making that joke on stage, Imad, you know, where I have a, I have a, effectively a replicator, a nanobot. Materials are effectively abundant, energy is abundant, uh, information is open source. So, uh, it starts to become an interesting world.

• 1:06:10Peter Diamandis?

  And

• 1:06:12Peter Diamandis?

  You know what I find even more interesting? It's gonna happen during our lifetimes. You know, we can get into a long conversation about whether we're living in a simulation or not, but it's the next 20 years. You know, it's the next five years when all what we're talking about today is playing out definitively. But in the next 20 years, as we're adding decades onto our healthy life, as brain-computer interface starts coming online, things, you know, we thought things moved quickly this year. They're gonna move, we're gonna see, I think the estimate that Ray Kurzweil talked about, and we've talked about at Singular University for a while, is we're gonna see a century worth of progress in the next 10 years. So what does that look like?

• 1:06:53Peter Diamandis?

  And the biggest concern is, I think governments don't do well in this kind of hypergrowth, in this kind of disruptive change. Young kids will do reasonably well, but governments, which are, you know, governments' religions are structured to keep things the way they are. Any reaction to that, Imad?

• 1:07:12Speaker 03

  Yeah, I mean, I think they perpetuate the status quo. I think decision-making under uncertainty, you minimize for maximum regret. That's why I set up Stability, to offer stability at this time. So we can help out governments and companies and others and standardize the open source models and others that everyone builds on. It's the foundation for the future. I think that-

• 1:07:31Peter Diamandis?

  I'm glad, I'm glad, yeah.

• 1:07:33Speaker 03

  Yeah, and then I think, I think that, you know, it's like we're setting up subsidiaries in every single country and they'll eventually be owned by the people of that country. We've got interesting things coming that we'll announce. The interesting thing you said there, like, why is it happening like this? You move, it's not Moore's law, right? This is like Peil's law, Metcalfe's law. It's like a network effect that's happening right now. And that's why you're getting this acceleration, because network effects are really exponential, where everyone's trying out this thing and exploring this new type of technology and sharing information back and forth quicker than anything you've seen. And all these technologies happen to be mature at the same time. So I can't see that far out. You know, everyone says, why do we say 20 years? We just pull it out of our butts, we just got done with the year, right? All we know is that things are never the same again.

• 1:08:19Speaker 03

  You will never be able to set an essay for homework again at a school for every school in the world. And there's more and more of that that's coming, but these things also take time to fit into our existing system. So as you noted, you know, programming, Copilot's been out for years, it's got a lot better, but it takes time to integrate into workflows. It's not like it goes and takes over everything at once. You know, how long does it take to get into things? Like 1.5 million people still use AOR. You know? Lotus Notes is still used around the world. I think it makes it hundreds of millions a year. But this may change quicker than anything we've ever seen before, but it still takes time. And it's very exciting. Yeah.

• 1:08:58Peter Diamandis?

  It has changed quicker than anything we've seen before. And there will be something that that moves 10 times faster than it did. Um, yeah, it it is the most extraordinary and exciting time ever to be alive, other than perhaps tomorrow. Um, um, uh, Iman, I just want to say thank you. Uh, I know your heart and I know your, uh, your mission and I'm grateful for what Stability AI is doing. Um, and, uh, and thanks for joining me today and thank you to everybody listening and for the questions.

• 1:09:30Speaker 03

  Oh, it's my pleasure. You know, I'm thankful to the community and I think it's going to take all of us to guide this properly, you know? So just embrace it, dig in. It's, as you said, the most exciting thing ever.

• 1:09:40Peter Diamandis?

  And if you haven't, uh, followed Iman on Twitter yet, please do. He does tweet on a regular basis. And, uh, all right, pal, uh, I'll talk to you in the days ahead.

• 1:09:49Speaker 03

  Check, everybody. Thank you, everyone.

• 1:09:51Peter Diamandis?

  You're welcome.

• 1:10:04Speaker 01

  Every mile matters. Whether you're driving your dream car or your daily commuter, the Mobile One team knows that. It's why we created Mobile One Advanced Clean Motor Oil. It cleans your engine in one oil change and helps extend engine life, delivering performance, protection, and cleanliness for vehicles of all ages, all to help keep your engine running like new, whatever you're driving. Visit mobileone.com/advancedclean to learn more.