Living in an Exponential Age

0:00Hi, I'm Raoul Pal, CEO and co-founder of Real Vision. Alongside brilliant minds like Edward Snowden, Benedict Evans and Balaji, I'll be on stage exploring the extraordinary potential of AI and the profound change it represents, not just for financial markets, but also for the world as we know it. With over 5 ,000 attendees and over 150 side events, Singapore will become a vibrant hub for a full week from the 3rd to the 9th of June. Visit superai.com to register and join me with 20 % off tickets with the code REALVISION. Link in the description. Thanks.

0:47What happens when computers start having original thoughts? Hi, everyone. Welcome to the Real Vision Dally Briefing. For the last show of the year, we wanted to do something special. So we're unlocking part of the December Exponentialist conversation between Raoul Pal and David Matten. This monthly sit-down is part of the Exponentialist Research Service from Raoul and David that covers the fastest period of technological change we've ever witnessed in human history. In this 2024 outlook, they dive into the trends and companies that are likely to have the most profound disruption on society and the investment landscape.

1:24Buckle up. RAOUL PAL, Welcome back to The Exponentialist, everybody. David and I get together once a month just to pick through the big theme and where we are, because it's happening at lightning speed, go through some of the stuff we're learning as we go, and give you some pointers of where we're all going. David, good to see you, my friend. DAVID BUCKLEY, Good to see you, Raoul. How are you? I'm good. It's been another crazy busy month in the exponential age. It's like, I've never seen anything like this. I know we say this all the time. It's a bit boring to people, but it's true. I mean, it's the eternal refrain.

2:04Strangely enough, it's the one thing that never changes as you approach the exponential age that you just feel constantly like, oh my God, it's getting faster and faster. This is my favourite time of year because it's a chance to look back across everything, look forward to the year ahead. And yet, when I do that, it really feels that as much as we've had a shocking year in terms of progress and advance, 2024 really is going to be wild. And it's so much of what my tangible feeling right now is that so much of what we've been talking about these past two, three years is really going to bite in 2024, is really going to feel real to people in a way it hasn't so far.

2:53RAOUL PAL Yeah, I mean, if you just look back over 2023, we've just started launching some of the AI. It started the summer of 22 with the launches of some of the stability, the stable diffusion images, stuff like that. And then it just went warp factor 10 as GPT just went from GPT-3 to GPT-3.5, GPT-4 to GPT-4 turbo nutty whatever to digest images, create images, interact, and then building on top of Google and Facebook and Microsoft, all of their products, and then Facebook coming into the race. It's like it's bananas. And then the open source models coming into the space as well. That's just one year.

3:54RAOUL PAL Yeah, one year. You look back across the year, and it's just over 12 months since ChatGPT launched. that was November 22. But that feels like ancient history. I mean, looking back over the year and thinking, wow, that's essentially part of my review of the year, just about. It feels insane. That feels like another age now. We've lived so many lifetimes since ChatGPT came along. And we're now just so much deeper into all of it. And what's happening with generative AI has gone insane. But what's happening with what I tend to call real-world AI is also insane. Now, when you look at the way full self-drive is developing at Tesla, the Optimus robot now, we saw the clips of it over the weekend, and how advanced it's become and how close it will be to walking into factories.

4:52Yeah, it just feels like we're on the cusp of monumental change, and it feels that way, or I'm convinced of that, even in a way I wasn't two, three years ago. It just feels like we're so much deeper in now. Yeah, I'm also struggling, because I think this through a lot, as you do, and I think I can see sort of directionally where 2024 goes, maybe into 2025, because the foundations of all of this stuff is being laid. But by the time I look out to 2030, it's like a black hole, because the advancements are going to be so large, and the impacts will be everywhere with this foundational layer driving it, the compute, the energy, and the AI.

5:44And then everything that gets built on top of it is going to literally change every single aspect of our lives. So what is your main thoughts for 2024? How do people get their heads around it? RAOUL PAL Yeah, I think you're right in terms of looking out to 2030. The scale and the complexity of this interconnected mega system we're building just defies comprehension. and we have to act accordingly. It's extremely difficult to draw forecasts out meaningfully that far. I mean, we can be directionally, you know, we can be intelligent about where we're heading directionally, but what the world will look like in 2030, I totally agree.

6:31It's just, it's extremely difficult to think about. In terms of next year, my overriding thought, like I say, is that 2024 is the year the exponential age really manifests and feels real in a way it never has before. And I say that because you see the seeds of that being sown right now and you see the signals of that around us right now. We can talk through some of them, but across space, so many of the technologies we talk about, space, robotics, AI, blockchains, you know, which I know we'll get into, you feel this becoming real or tipping even into the mainstream in a way it's never done before.

7:15So 2024 for me is a pivotal year when it comes to the exponential age and just a such an exciting year for our community. Because look, if you've, you know, we've just launched it, dare I say it, we launched it at exactly the right time. And if you've joined, you've joined at such an exciting moment to begin this journey with us. and we're at the foundation layer right now you know the the the first couple of essays really built some big models some foundational understanding and we're just going to take off next year into this incredible journey and let's be honest we're going to be like sprinting to keep up with what's happening around us while trying to continue to build these big models that deepen our understanding of what's going on yeah i think so so let's talk about some of the key thing.

8:05So let's talk about AI, because that's kind of the big daddy, the thing that's accelerating. Well, one of the key things is accelerating. We saw some interesting stories. One was that ChatGPT4 had dumbed itself down over Christmas, and they found there's a seasonality that it had learned from humans that it becomes less productive at certain periods of time and more productive, but over others, which is fascinating. Yeah, I mean, yes, I mean, that at least is the leading theory that, you know, deeply embedded in the training data is the idea that you just don't work quite as hard over Christmas and you take a break.

8:46And it seems to have picked up on that. And that's just another manifestation of something we talk about often, and we often talk about as a kind of fault, you know, that AI, these AI models reproduce bias or prejudice in their training data. But we can also, we need to be alert to that and do things to mitigate it often. But there are other ways that it's going to be a really powerful facet of these models. I mean, taking the serious underlying point here, their ability to sort of absorb human culture, if you like, in that way, and human preferences and reproduce them, will also be useful to us sometimes.

9:25This is a really weird one, but no one's come up with any other good reason why it seems to, and it really does seem to be kind of taking the month off. Yeah, because it was starting to get lazy, and so I don't want to do that task. Yeah, it was like, it's like, yeah, no, can you just go and read the documents yourself? Just things it's never said before. And then, you know, I mean, that just puts me in mind of, the way these models also are, and you see ChatGPT changing over the year in terms of what it's willing to do and the guardrails placed around it and some of the politics around that. I think a huge issue in the exponential age and a huge issue for next year that will feel much more real to people is the way machine intelligence and politics and social norms and society cannot be thought of independently any longer.

10:22So much of the way we govern our societies and the way we shape our norms and the way we live collectively is going to become an argument about how we represent that inside machine intelligence and the values we give machine intelligence. Hey everyone, we're going to take a quick pause and hear a word from our partners. We'll be right back.

10:47Have you ever wanted to trade Bitcoin but haven't dared try? With Plus500 Futures, you can trade crypto without the hassle of opening a wallet. With just a few clicks, you can register and start practicing with their free and unlimited demo. See a trading opportunity? You'll be able to trade it in just two clicks. Feel ready? You can move to real money with as little as$100 once your account is approved. And the great thing is that in addition to crypto, Plus500 gives you access to a wide range of instruments. S &P 500, NASDAQ, gas, and much more. Explore equity indices, energy, metals, forex, and beyond.

11:20With a simple and intuitive platform, you can trade anytime, anywhere. Experience the fast, accessible futures trading you've been waiting for with Plus500. With over 20 years of experience, Plus500 is your gateway to the markets. Visit us.plus500.com to learn more. Trading in futures involves the risk of loss and is not suitable for everyone. Not all applicants will qualify. Plus500. It's trading with a plus.

11:47RAOUL PAL, There's another school of thought that I'm a proponent of, is that we treat chat GPT as if it's just got human traits, and it's just basically modeled on humans. But I think it's different. I think it's like Confucius coming down from the mountain, speaking Mandarin, and we're asking questions in English, and we're trying to judge how smart this thing is. We don't speak the same language. But what was clear from the, and you and I were swapping Twitter messages about this, is that the DeepMind AlphaGo, this thing learned how to do things in ways that humans don't. And then there was the other thing, the science paper that had come out.

12:31RAOUL PAL Yeah, this is huge. RAOUL PAL This is again, completely original thought. RAOUL PAL Yeah. RAOUL PAL And this is what they're thinking with AGI is, obviously, it'll solve some of the main issues of physics and the understanding of the world around us and molecular biology and biogenetics and all of this stuff, because it actually doesn't think like humans. I used to hear a lot of this stochastic parrot stuff. It is clearly not that. Even in the interview with Jeff Bezos, which we'll come back to, he's like the same thing. It's like, we don't know how the hell these models work. I've heard that from Anthropic.

13:09I've heard it from Sam Altman. I've heard it from Lex Fridman. I've heard it from everybody. It's like, we actually have no idea. Yeah. And I think that's right. And this fund search result that you're talking about from DeepMind just last week, and I wrote about it in the newsletter, it's absolutely huge. I mean, I don't think it's got quite as much attention as it deserves, as much attention as it's had. It probably deserves more. But the kind of hugeness of it is somewhat buried because this is the first time. So this fun search, and it's called that because it searches through mathematical functions, right?

13:44Not because it's fun, though it is fun for people like us to talk about. This is the first time truly that a large language model has created entirely new information to solve the longstanding science or mathematical problem. I agree that AlphaGo Go made moves no human would ever have made. And that has fundamentally reshaped Go and is an incredible example of this synthesis of machine human intelligence. But nevertheless, the possibility of making that move was apparent to us. It's part of a structured domain where you can make several zillion moves. Which one are you going to choose? This is even a class apart.

14:30You know, there's not sort of a menu of options when it comes to solving this longstanding geometrical problem, the cap set problem. It's not like there are several zillion options and which one are you going to choose? There's no options. It's just a blank page. Solve it. And it did. That's never happened before. And that essentially appears to be proof that a large language model can create authentically new knowledge, not just like remix what's in its training data in really interesting ways and come up with stuff we never would have, but create something totally new. And this is what I mean.

15:06I mean, that puts us on the cusp then for an incredible 2024. How many longstanding problems are going to be solved in 2024 in that way? Because mathematicians have been banging their head against the cap set problem for about 50 years. People have built their career on trying to solve this problem, essentially. It's just one huge result in an incredible year that points exactly in the direction you're saying, you know, We don't understand. And the deep mind engineer leading that work is literally like, we don't know how exactly it's done this. So yeah. RAOUL PAL I mean, also, I'm getting a sense, and I'm starting to see some chatter about it.

15:52I mentioned it a while ago. I'm getting a sense that OpenAI are using AI to build AI and to build the business. So it's saying this is what you need to launch now. This is how we build it. Now, whether they're using, there is a school of thought that they're using ChatGPT5 to train ChatGPT4. And there's also a theory they're using ChatGPT4 to train ChatGPT5. So you don't have the human element in the middle of it, because these models are faster at training themselves than we can be. And I think that's one of the reasons why OpenAI is so fast at advancing and rolling out new products is because I think they're using AI to do it, which is, again, mind-blowing.

16:41RAOUL PAL Yeah. And I really wouldn't be surprised if that's right. And look at the deal they've just signed with the publishing giant Axel Springer to get access to all Axel Springer content. and they're going to use that in training data. And they're also going to serve summarized versions of Axel Springer content. And for people out there, that means Business Insider, Politico, whole range of massive publications will be summarized inside ChatGPT now. Officially, it's a done deal. You see how the big players are going to be able to accelerate via access to really, really valuable proprietary data sets.

17:21And yeah, it wouldn't surprise me if they are also using you know, there's using AI in the way you say. And there's definitely, yeah, there's, I mean, definitely, it really would appear that there's things going on inside OpenAI that we don't know about. That whole fiasco, which you have to count among your sort of tech story highlights of 2023, that whole fiasco was bizarre. And we don't really know the full story. But there's strange, strange and interesting things going on inside OpenAI. And yeah, I mean, eyes wide open again for the next 12 months, because it's going to be wild, I think. Yeah, the other thing that I've been following is the rise of open source because of the power that OpenAI has and others have.

18:12And obviously, Meta have gone down the open source models with Lama, and that seems to be going well. There's this new one that's come out, Mixtral, which has come out of France and seems to be super fast and very light in how it operates. But the one that really interested me was, I just saw a couple of days ago, something called Dolphin built on Mixtral, which is an unguarded AI, because I'm getting frustrated now with how guarded OpenAI ChatGPT is. You can't ask it any question without it giving you a disclaimer or shutting off the information to you. When it first came out, you could ask it amazing medical questions, and it would do diagnosis to all this stuff, and they've shut all of that down.

19:02So it's very interesting to see the rise of these completely open, completely uncensored models. And because they're open source, they're difficult for governments to regulate, which I think is another battleground. Yeah, well, exactly that. And that's exactly what I'm talking about when I talk about, you know, sort of in 2024 and beyond, AI is politics and politics is AI. You know, I think, yes, I feel exactly the same way about ChatGPT, you know, just relentlessly sort of it feels like a sort of scolding school teacher. You know, you ask it a question, it gives you a long disclaimer and tells you off about this and that.

19:45I think that sort of open source AI for the people movement is really, really interesting and that we're going to see a strong push to empower people with essentially their own models that they can tune as they wish that. Yeah, that are essentially unguarded, that they can use as they wish. And that fills me with some sense of hope and optimism that, you know, this incredible machine intelligence revolution can be, can be, if we let it, a ground up sort of empowering moment rather than a kind of top down imposition on people. Because I'm just, I'm endlessly fascinated by the history of the personal computer and the history of the internet.

20:26I mean, like you, I think it's fair to say I'm in many ways a historian, though I spend my time thinking about the future. That only makes sense in the context of understanding the past. That's something fun. I mean, we can talk more about that one day because that's an interesting story of its own. The personal computer and the Internet were intended to be. There was a dream of huge personal empowerment through those technologies. And in some ways that's come true. In some ways it hasn't. And I think with machine intelligence, we're going to see the same thing. We're going to see that battle between personal empowerment and sort of top down.

21:05Hey, everyone. We're going to take another quick break and hear a word from our partners. We'll be right back. Alongside brilliant minds like Edward Snowden, Benedict Evans and Balaji, I'll be on stage exploring the extraordinary potential of AI and the profound change it represents, not just for financial markets, but also for the world as we know it. With over 5 ,000 attendees and over 150 side events, Singapore will become a vibrant hub for a full week from the 3rd to the 9th of June. Visit superai.com to register and join me with 20 % off tickets with the code REALVISION. Link in the description.

21:44My big thing, maybe it's 2024, is Apple will embed in the M3 chip a large language model. So I'm talking to you on an Apple Mac, and I've got my iPhone next to me. And all of that, of everything I do, this conversation, every email I open, every website I browse, every document I store on my computer or in the cloud that I access from my computer can be referenced via my own AI, which is private. And it will help me navigate my life. And we're just starting to see the rise of AI embedded on phones. So they're not on the internet, not like my chat GPT on my phone. The models are actually on your phone.

22:28So you don't need the internet there. But I think that's a tip for me is that's where Apple is going. And that will change the game yet again, because nobody else has the hardware to do it. Now, Google will do some in Pixel in the phone. But it's just not the same complete suite of everything, all the music, every photograph you've got, every single element of your life. you can have an assistant for. Yeah. And again, I think that will make this feel so much more tangible to billions of people because Apple have a position in how many ever billion pockets around the world, right? People put their hand in their pocket and they pull out an Apple device, billions of them.

23:10And exactly as you say, they've been working on these edge AI chips. They've been putting AI chips in devices for a few years. They have incredible AI capability on device that is essentially not really being deeply used right now. And the obvious move that they must make at some point is to put a self-contained large language model on device and do what most people would call just revolutionising Siri. And this taps back to, again, one of my longstanding obsessions, virtual companions. You're going to have an AI conversational entity, a companion on your device that you talk to endlessly that knows everything about you that knows all the content you love you know all your phone exactly as you say your photographs where you need to be but goes beyond transaction and functionality is really a sort of counselor to you and a sort of philosopher and guide and when that drops on hundreds of millions of training on every single thing that you ever say on every zoom call every email you write every tweets that you write everything that you read it knows you better than you know yourself exactly and you see meta you know if we want to talk sort of in in a kind of um inside industry talk for a minute you see meta playing for the same space with this range of ai characters that were launched at meta connect a few months ago but like i like we i think we've said before i just find those a bit strange because they're based on celebrities and you don't really want to talk to you know i've been i think i've RAOUL PAL, I've alluded a few times that I've been playing around with, and it's now publicly launched, not mine, but Delphi.ai, I believe it is.

24:51They have been building personality-based bots where Balaji has uploaded all of the content, plus content of stuff that formed his opinions, books, and other stuff. So then you can have a conversation. And I've been testing. I mean, I can have a conversation with myself verbally in my own voice. And it's not long before it goes to video. And so it changes the creator economy, the expert economy, and all of these things. Because you could have Huberman as your personal wellness guide. You can have Raoul as your personal finance guide. You can have, you know, that's what's coming. And that's 2024 writ large.

25:39You'll see a huge rise of these things. Yeah, I think that's right. And I think it will really feel real to people when Apple make their move. And it also puts me in mind of one of my real highlights of 2023, which was that Harvard and Stanford paper where they created AI agents inside large language models, essentially simulated personas. So you had 25 different simulated people and they threw them into this virtual town called Smallville and gave them the aim of organizing a birthday party and just pressed go on that and just watched all this emergent social behavior. So, you know, people started talking to one another, buying balloons for the party, planning invitations, asking each other out on dates.

26:27We're going to see AI agents take huge steps forward in 2024. And again, it's just so deeply exciting, because what you have there is a little simulated society. And when we can scale that up, and we have the compute to scale that up, and you can simulate tens or hundreds of thousands of people inside these environments, that's going to blow apart consumer research. RAOUL PAL I saw on Twitter yesterday, there's a 20-year-old kid fucking around with a bunch of AI agents. He built an agent for portfolio management, an agent for reading corporate statements, an AI for put together all the component parts of what you need to become a financial analyst.

27:12And these AIs work together. These agents work together. So one goes and reads the thing. It gives the information to the other one. The other one asks it questions. Hey, can you go back and check this? It then builds out. By the end of it, there's a writing AI that puts the whole lot together as a research report. And that's using these different agents to do different tasks. And again, people aren't really aware of this stuff. But that's where a lot of the applications layer will come from, is the building of agents on top of each other. I think that's right. And I think exactly as you say, people don't, you know, AI agents is sort of nerdy enough for it to be way under the radar for most people.

27:50But I think in 2024, again, that will really bite in ways that people find really tangible and surprising and it's like an oh fuck moment. And look, OpenAI surely see that 2024 is the year that they have to make serious commercial inroads and they have to really start to commercialize this technology and really demonstrate to industry that it can be profoundly useful. And so I wonder if we'll see them really pushing and experimenting when it comes to agents, because it's those kinds of AI agents that are going to allow large corporations and other organizations essentially to automate a lot of tasks.

28:32And then that pushes us into, of course, really interesting and difficult social territory, because you're talking about replacement of human labor. I think that that's another thing really on my mind that will bite next year. The big players need to show this technology is truly impactful, not just exciting. they are going to make their play for that next year. And we're going to get into really tangible job losses, human labor replacement territory that will make this feel very real for people and uncomfortable. No one loves that, but it's coming. And before we move on to that whole topic, the other way I want to wrap up AI for next year is this US election is going to be a shocking moment.

29:19You will not be able to believe anything you read, anything you see. You don't know who's an AI, who's not an AI, what image is real, what quote is real, what video is real, what audio clip is real, nothing. And this is something I've talked about for a while, but we're about to start that cycle. And by the summer of next year, it's going to get crazy. The one thing I did stumble across, and it's not a recommendation, but something I'm doing some work on, is WorldCoin, which everybody hated because it's, oh, my god, it's evil. They scan your iris. But I actually met the guys at WorldCoin here in the Cayman Islands.