Humanoid Robots, the Job Market & Mass Automation - The Current State of AI w/ Emad Mostaque | EP #114

27:27I would have never guessed zero. Does that surprise you? There's a stotishing. I mean, what is... we have this concept of, hey, I want to make a picture. I want to make music. I want to write an essay. And it costs X number of man, woman hours, right? it? Yeah. Yeah, all of a sudden it costs nothing. The thoughts to the creation process, sure you can spend a week and make it even better, but to get to an 80 -20 now is almost instant for any modality. Yeah. Like again, we're not talking about perfection, but you don't expect your graduates to be perfect on their output from drawing to coding to anything.

28:09That rapid iteration is just something that's beyond belief because, again, when you look at these models, there's just a few gigabytes and they can run on your smartphone. And it does seem something's wrong with that energy equation, right? Energy is that much down to nothing. The economic equation is what really bugs me. The value we're extracting. I mean, Google did that and it monetized on advertising, but it also uplifted humanity in such an extraordinary world. I just came back from India where GEO, there are 5G network, has demonetized data and communications and extraordinary rate. And so we are seeing this abundance equation of digitization, dematerialization, demonetization, democratization over and over and over again.

28:57And this is the ultimate one. And I'm just, you know, it feels like we haven't been able to even understand a fraction of 1 % of the implications of this. So you think about the energy that's taken to put a doctor through school, right? And the cost of that, millions of dollars in the US. And that expertise can suddenly be available and replicable and the push of a button through a digital AI and soon an AI doctor and surgeon that outperforms human doctor and empathy. There's this relationship between energy and GDP per capita, so that's pretty much a straight line. And it just feels that when you buy energy is output which is how we're available to the society.

29:43Per person. Per person. So the more energy you have per person, the higher your GDP, because you could build stuff, but then more than that, you can support service industries and things like that. You can support the million dollar doctors with a decade of work. Yet all of a sudden, the energy equation for intelligence has collapsed to nothing. Google and the previous internet reduce the cost of consumption of content down to nothing, and that's where they could have add and other models to have that go. They want your attention, shall we say, and you're purchasing on the other side. Now the cost of creation almost has dropped to nothing, not creativity necessarily, but the creation of information of knowledge, you know, and again not novel knowledge just to how we distribute this because we never have enough doctors.

30:30We never have enough good doctors. And again we're not talking about great doctors. I think all these discussions of AGI are like super humans and the top 0 .1 % I just want an average doctor to be available to everyone and I have a vision teacher to be available to everyone but the average level to be really much better than it is right now. Well I get it but I believe you know I spent a lot of my time in the intersection of AI and health as do you we have that in common that that But there is no way that any human doctor can integrate the gigabits of data that come out of a current digital upload.

31:13Understanding all of the blood chemistries, your genomics, your imaging data, but in AI can, it can contextualize all of that and then give you some root cause analysis. So I think that we're not far away from having the average AI doctor be better than most all physicians out there. I think we're there pretty much now. It just hasn't been integrated. If you look at the benchmarks from radiologists, this to that and the long context window work. But again, this is the exciting thing as you go from individual models to pulling them together. So you have your own medical team. Yeah, you're not just calling on one doctor and we see this with a technology as well Like we had gigantic models and we still got them with the Ben Yulama, but now we're moving to mixture of experts and routing to expert models that experts are these are agents and sub agents There's an AI radiology, you know x -ray radiologist agent and one for MRIs and one for CTs and went for blood chemistries and went for genomics and all of them were being pulled together as your medical team.

32:21Yes. And they have common language to talk to each other. And you know, the outputs are checked as well. Like in medicine, before we get Dr. AI doctors making diagnoses, every diagnosis by human doctors should be checked by AI. Yeah, I said it against that. It's going to, well, I think it's going to be malpractice to not have AI in the group in some number of single digit years, because it will save lives, humans make mistakes. They just can't deal with the amount of data. It's not the way it used to be. And if you can know because you get access to the data, I think there's an obligation to do that.

33:01Yeah, I mean, it's because, because we could only have, I mean, we all know that physician's scroll is unintelligible. AI is just catching up with that right now, right? the information flow from that hour long, that teach you that many times. You know, it's just a few lines, right? We lose all that context and all that rich context. So is anywhere in the world, maybe found, London Health, that you can go and see your lung radiology over time and how it progresses? We just saw, I think, a paper come out that showed that you could detect breast cancer up to five years before, because you just dumped it and it just analysed every small piece of things.

33:40You can't see how the world evolves because the lack of context from black and white scrawl made it so our medical systems assume ergodicity. A thousand tosses of the coin are the same as single coin costs a thousand times, but humans are individual with complex systems, but everyone gets 500 milligrams of paracetamol. You know, we don't care about things like cytokrampu 450 abnormalities that mean you metabolize coding into morphine or fentanyl into death. Yeah, I mean, there really needs to be a genomics agent before you take anything. The other thing which people don't realize is you feel like when you get prescribed a particular medicine for a particular situation that you assume is going to work.

34:23But the fact of the matter is it works in about a third of the individuals if you're lucky. It was enough for the FDA to approve it, but no guarantee it works for you. So, you know, the other thing that you wrote in your paper, and I don't want to let it go on said, is besides this exponential rise of AI, AGI, ASI, whatever we're going to call it, is the emergence of humanoid robots. We've got, you know, probably on the order of 20 plus well -funded humanoid robot companies. You mentioned Unitry in China. We've got Tesla building optimists here. we've got figure which just is releasing a figure two.

35:05And if Elon's not necessarily been correct on his timing all the time, but he's been correct on pricing. And it's sort of a dollar per kilogram for a certain level of complexity. And his projection for Optimus is $10 ,000 a unit. Let's double it, get 100 % margin of $20 ,000 a unit. That's something that you could rent for a hundred bucks a month, having a humanoid robot running the highest level AI model accessible 24 -7 and that's going to change the world. Yeah, I think this is the thing. The embodied agents and the advances we've seen there now with on -device and cloud AI now being able to do segmentation analysis, robotics, like it's here all of a sudden all at the same time.

35:54We see, we see this actually in history a lot, where the same breakthrough just happens in different places, almost. And we're seeing this all the time in AI, where you see literally disparate breakthroughs because the control problem in robotics is pretty much there now. Yeah. For 90 % of use cases, again, not 100%, but these robots can manipulate. They can read a recipe now we've seen, and they can just make the recipe from the instructions with full autonomic control. And they can do autoregressive learning as well. It swelves. So you did treat have that and others are implementing that. What robot learns from the tasks of others?

36:31Which is not to say to people. Insanely powerful. You know, bringing it back to health, you know, the best surgeons in the world will be humanoid robots, eventually, that can see an ultraviolet infrared, didn't have too much caffeine that morning. You'll fight with a boyfriend or girlfriend the night before. And it's the CAPEX and the cost of electricity and then all of a sudden this robot is a surgeon in multiple distant villages in India and Africa with the highest surgical skills. And like you said, if people should know this, the number one question you ask your surgeon when you're interviewing them is how many times you do the surgery.

37:11And if they're all operating on a single operating system, then the robots all share the same millions of operations they've done. And this is again our collective knowledge where it's taught through medical school, but AI will obviously be able to teach other AI's much better. And if we can standardize that, then that's when it can go massive. And it comes down to just a production line process. It's like Jaune, you know, the smartphone manufacturer in China. They thought they were going to be added to US sanctions, so they decided to diversify into electric vehicles. And then China's industrial prowess means they now make really good electrical vehicles.

37:48Like you will see these industrial giants move forward and produce millions, 10 to millions, hundreds of millions of robots. And the really fascinating thing is that helps with the demographic issues of China and Japan and places like that at exactly the right time. Exactly the right time and you know the predictions we've seen from From the from Elon and from the CEO of figure is you know as much as 10 billion humanoid robots by In the 2040s more as many as there are humans I asked a friend of mine I might and ask you what's it gonna feel like when you're seeing humanoid robots Walking around the streets all around you all the time I'm just going to be weird, isn't it?

38:31Like we already see these little ones going around that are not humanoid, but there's no reason you can't make them indistinguishable by 2040. It's really sci -fi. I mean, actually, this is the real existential threat to humanity, 10 billion robots and a bad firmware upgrade. You know? What good buy to all of them? Of all of the, all of the like scenarios. Literally, that's the one that can get us all. I think there should be some extra checks. about it. When I asked my other friend that question, what's it going to feel like? He said normal. And I think we're going to have these moments of freak out.

39:11I remember I got a model X with the Go Winged Dores in the first month. Everybody's like staring at it like, oh my god, it's amazing. And then it becomes normal. It becomes normal. I mean, then again, there's different things in different cultures in Japan, where I bought so very much side by side, or you've become part of a Gundam, whereas we have this terminator concept of robots or AI God in the West. But ultimately, as you said, people are used to being able to type in words and make magical images in two seconds now, or using chat to UPT for your homework. I just talked to some of these kids and they're like, yeah, I just use it, you know?

39:46There's an initial shock and then also the magical becomes mundane. I think we're very good at adapting to that with our filters. But 10 billion robots means that there is, I think one of the ways to really think about this is how much energy do you need to put in your own physical mental energy for a job to be done? And it's collapsing everywhere, the amount of energy that we need to do productive things, which means we can do so much more productive things. Yeah, I think the scarceest resource in the world is time. You know, I'm working on it from the longevity standpoint of how do we add more years?

40:23AI will be the most impactful agent in how we create, we reach longevity, escape velocity, but I think people need to realize that we've been buying back time with technology, you know, Google saved us hours of going to libraries looking for that book and Zoom is, as normally you and I would have to flow to London or Udo LA to have this conversation. Instead, it's microseconds of, you know, click and we're together. So it's about buying back time, which is our most scarce resource. Well, I think there's two things. There's time, but there's also focus and flow, which I think of the other two interesting parts of that.

41:04So we know how useful an amazing chief of staff or EA is, And everyone will have their own one. So like, how do you live longer? Well, first of all, exercise and eat well. But, you know, you don't have that person, the entity that you trust, telling you and advising you sometimes, so you cheat, you fill down, et cetera. No one will ever need to be alone again and can have a persuasive agent next to them, which is something insane to think about. That knows exactly how to talk to you to put you into good habits. and it might be physical embodied as well. It might cook your breakfast and everything.

41:42Like how difficult is it to have healthy meals all the time that are fresh? It's hard with robots. It's easy, $100 a month, as you said. Yeah. And that's kind of crazy to think about from that longevity perspective. But then it's about our own biggest enemies are ourselves. You know, like I took off a few months for planning to clear my head and then read everything and now I'm compressing it and putting it out and off we go again. If I'd had all the help that I needed and it's not like the people around me didn't help, I would have done even better, but everyone now can have access to that, which is going to be amazing for global mental health if we do this correct, and the capability of people to just achieve so much more by having that focus and flow.

42:20Talk about flow a little bit more. How do you think about flow, and do you think humans can enter flow with AI partners, AI agents? I think 100%. When I talk to all the creatives, you know, Stabit is one is generative media leader. How do the top artists use the AI? Do they use it to just make content, which I know they're like, we like to jam with it, you know, it helps us enter that flow state, especially when we customize the AI is for their catalog and things like that. And you can use it to discriminate to do, like you should not use chat to your PT to write your assets. You should use it as a sparring partner to get you to think outside the box and look at things in different ways.

43:03That's the best way to use this. You know, like a very smart, precocious graduate that occasionally gets things wrong. I think this AI is original creative, but we haven't built the systems to keep that flow state going and the flow state is the one where everything is just kind of coming and there's no barriers between ideation, creation, and iteration. You're looping. One of the things I like to speak to entrepreneurs about is you don't know how to think other than the way you know how to think. And that's where gender AI models can come in. You could describe to it, this is the problem I'm trying to solve.

43:46How would Steve Jobs take a shot at this? How would E -Mod Moustache take a shot at this? You can actually have it digest and present to you different approaches that break you out of your predetermined mindsets. That's pretty powerful. But then also understand you. Again, having a good sparring partner is amazing. And there's this base level. When we grew up with our parents and they supported us to go to school, didn't it? worry about all of that. But then you grow up and you have to worry about all of that. You don't have to anymore. And this is other level where it is the sparring partner, where it does look from different perspectives, where it knows exactly how to talk to you, cajoling or criticizing or otherwise.

44:30But if you own that AI, you know that it's on your side versus someone else's side. Like, what is the objective function of that AI? Is it trying to sell you something or is it looking out for you? And this is going to be one of the most important things because we're inviting them already into our everyday lives like Apple Intelligence is now going live on ten hundreds of millions of smart phones and it is smart and it's persuasive already and it's only going to get more smarter and more persuasive but is it looking out for you or someone else depending on how you build it? That's a really important point of who is the AI loyal to.

45:07Is it your version of Jarvis? I use Jarvis as my context thought for this AI. And being, I guess what you're saying is, listen Jarvis, I want to be more productive today. So if you don't mind, keep me inspired and keep me focused and keep me productive to reach these goals and having a muse almost in one sense that plays the right music, tells you the right joke, gives you the right data points at the right time. Well, you know, one of the things I discussed in the paper is this concept of, you know, we're going to move away from files, so PDFs or Word documents or images to flows of kind of, you know, points in time and all the things you pull together so you can rework outputs.

45:55but your life is like that. And if you have AIs that understand your context, is that AI provided by Google that's trying to sell you ads? And so it'll be like, well, you can do this by buying Gatorade, you know, and this and that. Or is it something which is like, now you're like, keep away from this social media junk, you know, you're clearly getting stressed by debating over who's gonna be vice president or something like that, you know, I've got you, you know, So these are all various interesting things because people have agendas. And one of the things internet too, the previous internet call was our focus to sell us ads.

46:35And that's fundamentally about manipulation. And so if we make these aIs that are massively persuasive, now everything in our flow, you can inject, has like nothing else with incredibly high CPM. And it'll be free because again, if you're not paying, you're probably the product, right? I am curious and I've written about this that once we have these deep AI model integration into our life, I'm not going to be buying anything myself. My AI is going to be buying stuff. It knows when I run out of toothpaste, it knows when I run out of shampoo. And rather than me having some smiling person, you know, shampooing their hair or brushing their teeth and being influenced by that ad, my AI knows my genetics and knows which toothpaste which he has, which shampoo is best for me, and is probably buying it without me ever asking.

47:26Is that going to spell sort of a disruption of the whole ad industry? I think so. I think the ad industry, as it kind of exists, is already under siege because a lot of the numbers were made up. And we've kind of seen that with the CPM discussions and others. There's been advertising power when you look at the amounts every single year that the big companies have to increase their ad spend to keep the same thing and now it's being really questioned. When it's inter -centimediated through inter -independent agents, the value of a consumer band drops. That's kind of understandable. And then again it becomes about just the stuff that's background and the stuff that's status, the stuff that's living and the stuff that's entertainment.

48:12So I think there'll still be massive of brands and entertainment and consumer and discretionary, but certainly on the other side, like as long as it's good quality and you trust it, does it matter, you know? You get commoditization. You mentioned in the white paper, the work of Daniel Kahneman, who just passed recently in system one and system two, thinking and how AI is injecting into that. You mind taking a second to recount that? So you know, you've got two types of thinking. Oh my god, there's a tiger in the bush and I'm seeing a little bit, you're instinctive thinking. And then you're thinking that's very logical step by step, chain of thought.

48:49In fact, that's what it's kind of called. And the previous era of AI was very much chain of thought or extrapolating. So what that means is that we had these giant big data sets, we call it big data, and then let around these huge regressions, Facebook, and then with 13 points about you or me Peter, it knew you better than your best friends, and they could target you ads. But it's a trap way the future light the past, not that you've changed something. Whereas these new models embedded context and they're very much instinctive. So right now they don't tend to do reasoning immensely well. This is why the challenges people put against them are like, well, it can't do mathematics and they've got better and better in mathematics.

49:33But then you saw a deep mind just got a IMO, got silver medal level in Dash and Math and then had silver medal. It wasn't through these LM type models. It was through the more gen tick models that look at stuff and chain reasoning together. We gave it calculators. Yeah, we gave it calculators. Because these models that we saw have the breakthrough that are like spotting a tiger in the word the instinctual models, they guess the next token, the next word in a sentence. by having all these contexts together or in diffusion models which do the image, they learn how to reconstruct and deconstruct images and that process architecture there.

50:12So it's all about process flows as it were and instinctiveness. And that's what was missing, this understanding of context like make it happier, make it sadder, explain it like I'm five. Classical AI could not do that, but it could extrapolate really well. And that was worth hundreds of billions of dollars. But now we have the missing part, the other hemisphere of our brain. I'm amazed how AI has actually taken up the mantle of empathy. You know, if you had asked me years ago, what will AI not be able to do in the next decade? Where are we humans going to be better off? I would have said it's in being empathic and connected with an individual.

50:57And it's actually now for the last year been proven far more empathic than humans, in most cases, thoughts about that. I think it's because we don't have the time to establish connection and it's hard because we're also busy with our lives, like our doctors or teachers. How many people do they have to see? And empathy from our organizations is drilled out of us. We're mistrusting. And AI doesn't need to be mistrusting. The AI has the energy to be empathic, at learns empathy through the process of the AI training, because the best way to get someone to trust you is to help them. The best way to establish empathy and report is what's my common context and my common framework.

51:40Because it understands these frameworks that are instinctive in terms of it can map them, that's why you can communicate to you on your level. like the best way to again deal with this AI is people tend to say like little short prompts. If instead you say hi I'm Peter and I love longevity and you know I'm the first principal's thinker and I believe in exponential things and this and that and that and you write your own pre -prompt. In fact that might be a subject that's taught in school, write your own pre -prompt. Yeah, that's true. Consistitalize yourself, give yourself, help the system understand and who you are.

52:18If you write a short intro to chat to your PT Anthropic, it will communicate with you so much better. If you tell it how you'd like to be communicated to, and then you can even tell it, communicate with me better, and then it will be even more empathetic and it will understand you even better, which is again kind of crazy to think about because we've embedded these intercontext. Bob Schiller has a great book called Narita, the will want prize winning laureate for economics has a great book called narrative economics about storytelling and how it impacts the whole of society on the economic and social level.

52:55And we're made up of these stories, these contexts. And now we've encoded these contexts kind of in style and we can mix and match them and that's what drives the empathy and the connection. And that's also, well, like I said, it's not just about the robots walking around, it will put more and more trust in these robots and these agents because they will be more empathetic and they will never get tired of us. And they also, I think the thing you said earlier, which is so true, they will take the time because they can run in parallel. I just rewatched the movie, Her, which was still fantastic the second time.

53:34and there's a point at which the star is asking this AI agent, are you speaking to someone else at the same time? And she says yes, and she goes, and he goes, how many people? And she goes something like 8 ,736. And he's taken it back. But of course, she's so incredibly patient and loving and empathic the entire time that it feels like you have their full undivided attention. But I mean that can happen because now we can have one AI per person, you know, or we can have a hundred or a thousand AI per person given how cheap they are. So you won't have that thing in her words or how many people you talk to just you and you're special, you know, and you're like, again, people like these affirmations they want to be stabilized because they're constantly filtering to get through the world and this will understand our filters.

54:25If you look at her, it's very interesting because there is no visual representation. You know, a lot of these things are like robots or avatars. It's all done through the air piece and the airport and really humans are wired for voice. You know, it's the best way to convey. Like, if you look at it on extreme basis, like again, using tools like hage and others, they took, for example, Millay speech at the UN and translated it instantly into English and you're like, that really connects better. But then you can take something like hitless speeches. And you're like, oh, he does not sound like a crazy, ranting German anymore, which is kind of crazy.

55:09You can take that Olympian from Turkey who just won the silver, you know, with his glasses without any technology. And they translate to that in English, you're like, he seems like a proper dude. These are kind of crazy. And again, we have to think, And what is the effect when we have individualized agents talking to us constantly that are our best friends and will never betray us and are looking out for us or they're working for someone else who we're inviting into our inner circle. You know, and that's really interesting. We're going to get into the conversation of open source and in close source and how we prepare for this future in just a little bit.

55:50But there's a bolded quote you have here. Let me read what comes up to it says, we're going to be processing information and generating new ideas faster than ever before. We're going to see a dramatic increase in the velocity of knowledge. And then you say, the question isn't whether they'll change the world. The question is, are you ready for the world they'll create? And I want to dive into this a little bit because we're in this super exponential period of growth, right? We all used to Moore's Law for the last 50 years where computational power was doubling roughly 18 to 24 months. We've seen that spike up to 10x per year, you know, Elon when he was on stage with us last year at a button's 360, he said, you know, he's never seen anything as fast as this.

56:42it's now at 100x per year. How does anybody possibly keep up with this? What's your advice to that CEO, that entrepreneur, that 20 -something, how should they think about AI accelerating at this speed? I think exponentials are a bitch. I think that's the thing, but not really. Naturally good at these things. It's a true exponential because of the distributed way that it's emerging, right? And again, nobody's quite sure what the lower bound of a cost of a piece of intelligence is. This has broken our mental models, even for us that are right in this industry. What we can say now is it's too cheap to measure already, and it's just going to get cheaper, it's not done yet.

57:31Maybe it'll slow down the pace of it. But even now, if it's stocked today, it's changed society because it's just that proliferation phase. like the seeds have been planted everywhere. You do not need to hire calls into workers anymore. Every single essay written is likely to have chat GPT somewhere in the process, right? That's cool. These are fundamental one -way doors to society. So the way to think about it is, Jeff Bezos has this great thing, which is the inevitable and the unchanging. So he started Amazon as kind of a book store, because he knew it would be an everything store, because the internet was inevitable.

58:06The distribution of information and logistics that would enable people to do that, and books were a great way to do it because building that, you could do it, but people always want faster, cheaper, better customer service. When you're building businesses, the mental models are like the ones that I lay out in the paper. I have infinite graduates, and soon they'll have bodies as well. I have a concept where I can go from static files to flows of knowledge, and I want to increase the agency of every single one of my human workers because they're going to be the highest marginal cost. What does that look like in a few years when this technology is inevitable?

58:42And we've created these reproducible processes and modules because if you start digging into the individual parts of the technology and the minutiae, it does get incredibly complex and incredibly quickly. But already we are hitting a bit of a wall, but not in that this technology will stop in that it's good enough. It's satisfied. All of the open source and close source are catching up with each other and we've hit a level of performance, which doesn't matter if it gets better anymore. And that's where you tie it back to what you're doing on an individual company country basis. And think, okay, the world has changed.

59:20This technology will put it afraight. The cost is cheaper than we thought, because I haven't seen anyone say, oh, well, that's a lot more expensive than I thought it would be. People you shot in the other direction. And it's available. Everybody, I want to take a short break from our episode to talk about a company that's very important to me and could actually save your life or the life of someone that you love. Companies called Fountain Life. It's a company I started years ago with Tony Robbins and a group of very talented physicians. You know, most of us don't actually know what's going on inside our body.

59:54We're all optimists. Until that day when you have a pain in your side, you go to the physician and they burn into your room and they say, listen, I'm sorry to tell you this, but you have this stage three or four going on. And you know, it didn't start that morning. It probably was a problem that's been going on for some time, but because we never look, we don't find out. So what we built at Fountain Life was the world's most advanced diagnostic centers. We have four across the US today and we're building 20 around the world. These centers give you a full body MRI, a brain, a brain vasculature, an AI -nabled coronary CT looking for soft plaque, dexascan, a grail blood cancer test, a full executive blood workup.

1:00:39It's the most advanced workup you'll ever receive. 150 gigabytes of data that then go to our AI's and our physicians to find any disease at the very beginning when it's solvable. You're going to find out eventually. Might as well find out when you can take action. Fountain Life also has an entire side of therapeutics. We look around the world for the most advanced therapeutics that can add 10, 20 healthy years to your life. And we provide them to you at our centers. So if this is of interest to you, please go and check it out. Go to fountainlife .com, backslashpeter. When Tony and I wrote our New York Times best seller life force, We had 30 ,000 people who reached out to us for fountain life memberships.

1:01:26If you go to fountainlife .com, BackslashPeter, we'll put you to the top of the list. Really, it's something that is, for me, one of the most important things I offer my entire family, the CEOs of my companies, my friends, it's a chance to really add decades onto our healthy life spans. Go to fountainlife .com, BackslashPeter. It's one of the most important things I can offer to you as one of my listeners. All right, let's go back to our episode. The next thing you talk about in the paper is a concept I love and I've written about and you do a beautiful, eloquent job speaking to it, which is it's not AI versus humans.

1:02:04It's AI versus humans plus AI, which has been defined as centaurs. Let's talk about that. How should we be thinking about AI as our partner and the definition of Centaur's and the benefits of being a Centaur? So AI, we just got little building blocks of various things. And again, if we look at the individual models, Tesla had 300 ,000 lines of autopilot code replaced by a single model like a C of weights. Visual data goes in and driving instructions come out. That's kind of insane, but they're just sitting there. They haven't been tied together yet. This human ingenuity that ties them together and they can repurpose and refact them, just like when you're leading an organization or a team.

1:02:52It's about up to you how you deploy them, and right now humans are better than AI at doing that. That's why humans can do the creativity part. They can tell their own story. They have that agency, whereas AI is really good at content. It's really good at following instructions. Again, it's got a liberal arts college, maybe it goes a bit weird sometimes, you know, and creative. It'll get better and better at following instructions and more and more specialized. And this is why the center or the human plus AI beats the AI by itself. We saw that in chess. The AI beat the human, but then the AI's plus the human beat the computer by itself.

1:03:29We've seen that where things like Go as well, whereby Lisa Doll was beaten by the AI, and then he got so much better in the entire level of go and brute across the top level because you can see more interesting ways and looking at it that broke the norm. So I think, you know, in the world of which you're going, you've got to embrace the technology and use it because it can increase your focus, fun and productivity. Like, question is why wouldn't you? It's like sabotaging yourself and tying one hand behind your back. You know, it's like saying, I don't want to have good quality teammates. Well, you know, it's like being a farmer and not using tools.

1:04:05It's using your hands only. Why would you not want to use tools? You identify what you call a triple threat on these on the centaurs just to recount them for you and let's talk about each one efficiency on steroids time the ultimate luxury and choices galore so efficiency on steroids. Well, I mean like what is the capability of any individual one of us? it's a function of our team and now we all have teams to increase our absolute efficiency and our team's efficiency as well because no one will say no to high quality talent. But what team would say no to that as long as they don't get in the way?

1:04:44Like obviously we see what happens when teams get too big and so we have to make sure not to over -complexify this AI or put it in unnaturally. But we have to really understand the capabilities. We don't need to understand the technology, just the capabilities, right? And that's what enables efficiency per capita to go up dramatically on individual, company, and kind of even country basis. If we kind of what was the next one again? The next one was time, you know AI frees up humanity's most precious resource. And I get that fundamentally. Time is the one thing we always are struggling with. We always have a lot more of it than we think there, right?

1:05:22In terms of quality time. And that's a function of like we always feel that we're on this treadmill and we fill our time with junk, but how much quality time do we have? Like, there was studies shown that when you go on a holiday, the most important thing is the highlight and the end of the holiday. Because of the way that we could look at it. So make sure that, you know, you send your bag to the head of time and you got a luxury pickup and stuff like that. Because not all time is equal. We can use this AI to decrease the amount of content we have to make and wrote work and increase the quality of what we're doing.

1:06:00Have content that entertains us all the time versus watching junk. Have the stabilization and focus. When I get up in the morning, I take a moment to sort of prioritize my day. Like, what's the most important thing I'm going to be to do today? One of them was sit down with you and do this podcast. And there's another three or four. And I love the idea in the future, my version of Jarvis is going to say, okay, listen, you do that one, I'll take these three and report back into the day. Again, just like a good team. And it's like, with that team, you can be so much more, it can free up the time, but then it can also improve the quality of your time and how you spend it.

1:06:39And this is so important as well, because we tend to just default to this autonomic kind of treadmill, of just consumption of fast food, not only input but content as well. It is very tricksome because this is the way our internet and other things are done. Like I spend far too much time on Twitter relative to the quality or X of kind of the output, right? Because it's so engaging. But what if I had my customised feeds that came to me and just gave me impactful stuff that both fit my world view but then also challenge my world view because that's what I wanted. My time would be so much richer.

1:07:18I built a company called Daily .ai that searches the world's news in longevity or tech or whatever it is. It customizes a newsletter for me because I only want validated scientific, technological content that is got a positive semantic. I don't want to hear about disaster scenarios. And it generates that, but you can have your AI ultimately feed you, like you said, if you want to be challenged, feed me the most valid content challenging my world view. But then also I think one of the things I'm depreciated here is that the previous AI that we saw this old school AI deliberately siloed us into our groups because it was better for advertising.

1:08:03Yeah, multiple studies of it. Yeah, face it. Echochambers. Echochambers. I think one of the really interesting things here when you look at education and other use cases social is allowing us to meet others within the same context and bridging the context. Again, this podcast is going out into multiple languages in our voices, which means that people can understand us in a different way that they could have understood us before. We're breaking down barriers with this and I think fundamentally most of the meaning of life and the quality of time is in our human connection with each other. Like when you're reading the classics, you're listening to the stories of these people.

1:08:41And if you suddenly can talk to Archimedes or Richard Feynman or any of these people, that's great. But you're what's even better discussing it with other people. I think the AI can really help with that. Well, the third one you said is choices galore. You say you want to code, but never have. AI's got your back need to understand complex data AI can break it down for you. And I get that. It's basically complementing you in areas where you don't have expertise or even where you do have expertise but want different perspectives. I think a large part of this is the question that these memes that go out sometimes of, you can just do things.

1:09:22You know what? They're stopping you. Most people are like, well, I can never ballroom dance. And they go and learn to ballroom dance, you know? I can never code. And they learn how to code. And they actually really love it. The bar and barrier for creating and engaging with this technology, the floor is risen. Yeah. What can't you do? Yeah. The bar has dropped. Do I guess what you want to say? Yeah. The bar has dropped the floor is risen. Like everyone's uplifted by this because the barriers to entry have dropped. The barriers to capability have dropped with this. So you have increased agency, you have increased capability because you're always constrained by finding that teacher.

1:10:07Now the teacher is here. This is like companies and governments in particular. In companies, there was a study done recently in which it showed your lowest 50 % of your employee base has gotten the biggest rise out of using in the general AI versus your top 50%. So it's like leveling the playing field, which is pretty extraordinary. And I can very easily imagine how AI's companies are gonna use centauras and agents and such. But I wanna understand your thoughts on government because government is the most sub -linear in most cases inefficient structures out there on the planet and are gonna have the biggest challenge, I think in the decade ahead, I'm not sure if you agree with that, but how do you think about government utilizing these resources capabilities?

1:11:03I think there's a mixture of hope and fear on the government side. I mean, like, I think for me, it's clear that AI can do better than governments and resource allocation, even with what we have now, because again, the bar is very low But one of the things is we've lost trust in our governments. That's a function of the lack of transparency that occurs. Like you're in LA now, I believe Peter. 600 million dollars on the homelessness problem or something who's being spent this year. You look at the LA to San Francisco railway, like we all know that there's a conundrum going in there. AI can deconstruct every single thing around that and show what's really going on.

1:11:48But there's the government really want that. This is the thing. Does it really represent the people or is it a continuous entity that's only about self -preservation? And this is one of the challenges we're going to have as a people because like there's riots in London right now, Bangladesh, they're proliferating. People don't really believe in their governments and systems anymore. Can we use AI to increase transparency and trust in organizations and be a co -pilot? every single individual engaging. Can we do things like direct democracy, whereby you have groups of people, and then you inform them on the topics and capture all the context of their conversations and feed it up to be representative and new forms of democratic engagement?

1:12:29I'm not sure, but I think we have to try because I'm very worried about where things are going now with just these massive echo chambers, the lack of transparency increased, graft, and the feeling that the governments to not represent us. I think we have to try. I also think government is going to do everything they're powered to resist because it's it's loss of control. It's and and the challenges in just governments by themselves it's big governments and large corporations and the two of them together are likely. If AI is going to disrupt the system it's going to disrupt that that unholy marriage.

1:13:09Well, I think again, these are local maximum conditions whereby that holly marriage works because things were positive some for a long now things are a bit more negative some. The corporations and governments have no option but to embrace AI because otherwise you'd be left behind by things that are not and there'll be huge pressure for that. I think it was Tyler the Nikon's tender that said that you know when 12 % of population, revolt, that's usually enough to get anything done realistically, right? And people will be required to embrace this technology where they'll fall behind their peers and they'll demand transparency and other things.

1:13:47But governments kind of like the technology if it's from trusted parties like your IBMs and others of the world because it makes them look cool. And this is a separate something like cryptocurrency. You know, government control of money, censorship, resistance, it was a system outside the existing system. Whereas this is already institutional. It took how many years for Bitcoin and Ethereum to have ETFs. Whereas AI is already institutional, it's already there and people are already embracing and bringing into government. But this is why I think like there are things that directly challenge the graft and other things.

1:14:22And there's things that we can do to make the general thing better, like the representative democracy staff, like the analysis of policy positions, speaking that available to everyone, increasing the ability of people to speak intelligently and understand the topics. These types of things should exist, and I really hope they do get built, because there's no reason that everyone shouldn't be fully informed of every policy decision before every single election in a way that is directly aligned and impactful for their community, for example, and be able to do scenario analysis. I agree. I agree. I'm not saying challenge.

1:15:01You walk into a voting booth today and you're asked to vote on all of these issues. And first of all, the way they're written is confusing. And secondly, most of us haven't had the chance to actually consume sufficient data. And a lot of people are actually just influenced by what the signs they read on the polling booths as they walked into to take it. But having an AI, your AI give you an analysis of how does this particular measure impact meaning that things I care about is a very powerful tool. I just don't know that politicians are going to want to give up their control and their manipulation just like they do, you know, gerrymandering of who can vote.

1:15:53Well, the thing is, you no longer need permission with the technology. Leveraging a technology like LARM or Mistraf, some of these language models, even Gemini or ChatGPT, you can build a system probably for tens of millions of dollars that would deconstruct every policy position of every single candidate with the publicly available data. and then you can input your own details and they would tell you how it would affect you with full transparency. I hope someone has a listening and we'll build that because I think it's so massively needed for every political election out there. Yeah and for beauty of it like I said is it can be fully open source or I think that's the best way to do it because there's very little argument that that shouldn't be built but again this is one of these just do it things.

1:16:46I think you can even We've been organized a hackathon and have literally have that built. Maybe we should do a government -generative AI hackathon. That's all about increasing transparency and using this technology to do so. It doesn't require permission, it doesn't affect the way things are. It just affects the capabilities of people. Which I think, again, this is the empowerment thing. That's all the upside of AI for governance and I agree with all of it and would love this to materialize. unit I've had a different conversation for, hopefully, a year ago where the ability for AI to be persuasive and persuade individuals, can you speak to that?

1:17:25Yeah, I mean, like, if an AI can be empathetic, it can be manipulative. And fundamentally, the business models of Google and Meta are manipulation. So if you have an AI whispering in your ear every single day, and it knows exactly everything about you, then it can sell you that toothbrush. You know, it can tell you, hey, this thing, sell me this pen. I can definitely sell you that pen, right? Right. Favorite marketing question. But then you think about, again, elections and things like that, you can have a personalized call for every single voter. It doesn't even have to be deepfakes. It can just be really persuasive.

1:18:03I can layer on Barra Cabalma at his best combined with Winston Churchill combined with Oprah, you know, for the sub -lingual frequencies, and that will be a convincing call from any political candidate. And it'll be exactly a local candidate. Knowing your position, knowing how many kids you have, where they go to school, you know, what kind of business you're in, and then giving you a spiel that's hyper personalized to guide you towards their direction. Yeah, I mean, you can see an example of this, there was advertising campaign just when this was all starting out by Cadbury's in India, where the Indian superstar Sharuk Khan.

1:18:43And you could go in your local kind of organization and you could just enter your business name type and it creates a customized Sharuk Khan ad where he talks about via the sweet shop, via the shoes. Check it out, Cadbury's India campaign Sharuk Khan. But again, like we can expect that the level of convincingness of these ads and high personalization will go through the roof and then the political side it'll go through the roof as well. But what's the personal side? Like if I'm going to have a Zoom call with a prospective client, I'm going to look damn good and I'm going to be so convincing and not have a starter or anything like that on the call.

1:19:20And when you meet me in real life, you'd be like, oh my god, he's not as good as that. I wouldn't use that technology because we can do it real time now. Like we had these small filters. But what if your voice, you can dial up the level of emotion on that voice, live on a voice call, you know? It's gonna be crazy what we're entering right now. I have to say and it's again all real time This is the other crazy part about it. Be able to translate your voice real -time, have real -time avatars that look like humans were there It's not a future thing. All right next up in your paper, Imaud, we talked about crossroads and what to do.

1:19:57I think that's one of the most important thing This is here, this is coming, this is accelerating, this is faster and more powerful. Anybody thought it's not stoppable. I think that's an important point, right? There's no on -off switch. There's a velocity knob. It is moving as rapidly as the largest most powerful companies in the world. And by the way, all AI companies are the largest and most powerful companies in the world can push it. So you state here and we're going to talk a little about shelling which is a new company and that you've been pregnant for nine months and getting pregnant in birth.

1:20:41You say shelling believes in democratizing AI. It's not just preferable, it's essential. And then you have five consider this diverse perspectives and sure AI caters to this social all societal needs, open source development, builds trust, collaboration speeds up innovation and broad participation keeps ethical considerations at the forefront. So let's dive into what do we have to do as a society? because people are asking, you know, I chair the AI committee, FII, the Future Investment Initiative, and the conversation is always okay. What should large corporations, what should governments, what should the public do, what should be our guidelines, you know, give us direction.

1:21:31And I love the fact that in this paper, and again, I really commend this, you know, how to think about AI, and we'll put the link in the show notes for your sub -stack in your on X for people to read it. It's not a long paper and it's beautifully written. Let's talk about what should people do? What's your advice, my friend? Thank you for that. I think again, understanding AI is infrastructure just like roads, you know, just like ports and others. Clayton Christianson, the famous Howard Business School professor, came up with disruptive innovation said, if the structure is the most efficient means by which the society stores and distributes value.

1:22:09And if a level of AI capability is barred from people, then you'll have an unequal infrastructure. You'll have private roads to knowledge, which I think is bad. When you look at the actual cost of building this and the coordination required, because the output of this is literally GPT -4's like a hundred megabyte model, that's it. It's a file. We can gather together and build this technology together in an interoperable way. And we will look at the inevitable future. Will our governments, our schools, our hospitals be run on black box models that we don't know what the data inside is? Because you are what you eat.

1:22:48The outputs are affected by the input. Or should we work together to create high quality national societal sectoral datasets? sets. If we have a model that's available for every single cancer patient and their families to guide them through that process, should that be a privatized model or should it be open infrastructure for everyone? And I think there's this class of in every regulated industry and industries, or sometimes over regulated, but they're there also to protect people. We should have an open class of models with a data sets, the models themselves are open. And you can do things like Anthropic, have a series of papers called Sleeper Agents.

1:23:26where you can poison data sets of models to an evil with just a few crueil heads. This also mitigates things like that. And it's impossible to tune out or find, which also is crazy when these start being decision making processes. So my view and this is what we're building at Shelling is that every nation is their own sovereign AI that represents them. We need to make our collective common knowledge open from creativity to science, to health, to education, and that should be open infrastructure for all. And then that makes the world safer as well. Because right now, AI systems are being trained on snapshots of the internet with all its imperfections versus a high quality curriculum.

1:24:04Again, you are what you eat. If you don't want it to exhibit some of this bad stuff, have high quality. We need diversity as well as quality. And this is what people are kind of going through on. So I think deliberately building the future where there's an inevitability of open infrastructure for industry by industry and standards around that is going to be very important. I think this is where governments and private sector should come together. And then that can proliferate if it's open because you don't need permission then as well. So stable diffusion and the other image and video models and medical models we did at Stability AI when I see you They ended up having 300 million downloads by developers 300 million because you can just take it and you could build on it Yeah, and that's just the start that was like over the last few years before the vast majority of the world even knew about this technology Again, if we build an open cancer model that is about empathy and guiding people through that process, how to talk, but also comprehensive authoritative and up -to -date on our existing knowledge and the latest knowledge on cancer and groups and connections.

1:25:19That can be in every language and used by every single individual afflicted by cancer in the world. Completely demonetized and democratized and the billionaire and the poorest child have access to the same information. The same information, again, it can proliferate. I think this is the key thing. But we should do that in a way, and that's why I call it shelling, that there is trusted entities and good governance on that data. What is the word shelling come from for your company shelling point? It comes from this concept in game theory called a shelling point, which is a focal point, a point of agreement.

1:25:53Because if you create high quality data sets for nations and for sectors because you have generalised knowledge and we need to have a corpus of that and then localise a specialised knowledge then people will use that because it becomes a go to point. If you create that cancer model and you invite everyone who has a stake in cancer to participate in that it will proliferate around the world and be the standard as good. Bitcoin is as a lot because for simplicity is effectively what would you be doing? it. It created a standard in a way of objective truth, but Wikipedia couldn't capture the nuance because the technology wasn't there.

1:26:32So something like cancer, you have your known knowledge, you have your standards, and you've got your stuff, we kind of know works, but it isn't in the individual corpus. And then we protect people by not showing them all of the other stuff, and they go to quacks and stared, or they do their own self -research and they repeat that process over and over again. What if we create a system that was comprehensive or thought should up to date and respect to the individual and their ability to do that research, but then also told you, how do you engage with your family, how do you follow through this process, connect to the resources that you need?

1:27:05That should be an open distributed intelligence system. That's the way they're in kind of a phase. I mean, the number of people I get saying my friend has this kind of cancer, what should they do, where she go, how do they start?

1:27:20It's sad and your heart goes out to them. And there is no one, one standard, but there can be. But it's also this loss of agency that you feel with cancer or autism multiple sclerosis. And so that's why I was like, let's go and build those models as open infrastructure. Use this massive compute, you know? And think about generative AI first healthcare, what should be open? And that's a base. And then people will build on top of that, they'll take it, they'll implement it, It'll make so much money off doing that, but it can then all use great healthcare records. Now, this is because we have a chance to reimagine healthcare, reimagine education, reimagine creativity, reimagine just about anything using this technology.

1:28:00So, that's what I was like. I said, a framework first, and then let's start building and gathering these communities together. And part will be open and part will be proprietary, and that's absolutely fine. But if we can make an open base, this can really proliferate, I think. What do you think about Zuck's Lama 3 .1 and his push on open source? I think it makes a lot of sense. If they can decrease cost by 10%, it pays for itself and they have. And again, I think Meta has been an open source champion with pie torch and other things, and it's not where he makes money, so he's commoditising compliment.

1:28:33But I also think it only goes so far. And you need an entity to be able to build this stuff deliberately, because Zuck, I don't think we'll ever build the models that run governments and healthcare and education and finance because that's not his company's objective function. His company's objective function is to connect people with an advertising based business model and there's a positive and negative side of that and he's amazing at his job, you know, but it shouldn't be on him to create these standards and in fact, you came back from India internet .org, for example, they want to give free internet to everyone in India or others.

1:29:10It had massive regulatory and other issues. So I think we need to again create a complement to that because you need open based models that are open weights as we call them. We don't know what's inside them, but we need that class of fully open models as well. And we need the class of models that are fully private as experts, because people will make the best models there. And so you'll have a continuum. And And then we have all the tools we need to have any type of graduate or consultant to solve our jobs and get things done. Did you see the movie Oppenheimer? If you did, did you know that besides building the atomic bomb at Los Alamos National Labs, that they spent billions on biodefense weapons, the ability to accurately detect viruses and microbes by reading their RNA?

1:29:58A company called Viome exclusively licensed the technology from Los Alamos Labs to build a platform that can measure your microbiome and the RNA in your blood. Viome has a product that I've personally used for years called Full Body Intelligence, which collects a few drops of your blood, spit and stool, and can tell you so much about your health. They've tested over 700 ,000 individuals and used their AI models to deliver members' critical health guidance. like what foods you should eat, what foods you shouldn't eat, as well as your supplements and probiotics, your biological age and other deep health insights.

1:30:35And the results of the recommendations are nothing short of stellar. As reported in the American Journal of Lifestyle Medicine after just six months of following biomes recommendations, members reported the following, a 36 % reduction in depression, a 40 % reduction in anxiety, a 30 % reduction in diabetes, and a 48 % reduction in IBS. Listen, I've been using Viome for three years. I know that my oral and gut health is one of my highest priorities. Best of all, Viome is affordable, which is part of my mission to democratize health. If you wanna join me on this journey, go to Viome .com slash Peter.

1:31:14I've asked Naveen Jane, a friend of mine who's the founder and CEO of Viome to give my listeners a special discount. You'll find it at Viome .com slash Peter. You list in the list of what we must do as rethink our economic models. And I mean, one of the biggest challenges is that our historic economic models aren't viable. They don't make sense anymore with what's coming given AI and humanoid robotics. So how do you think about rethinking our economic models? Where do you imagine it going? That's a nice simple question, Peter. I think it just, if we could just, you know, give us your point of view on the world economy for the next century, that would be great.

1:32:01I mean, there we go. Yeah. I mean, if I had more time, I'd write an economics book. We're at the terminal point of our current economic structure. We see it with Japan today, as we're kind of on this call, largest stock market draw down ever because they raised 25 basis points because they're functionally bankrupt as an economy, you know, we're the 500 % debt to GDP. We borrowed as much as we can from the future and we built this whole infrastructure that's about to have a brand new continent of incredibly cheap workers hit it all at once. In the West, we have the service -based economy. Like at least with robots, you're constricted with the amount of robots you can produce and the financing of those robots.

1:32:44robots will build robots and AI will build. Yeah, we'll go there. But the agent stuff, literally like a travel agent, as an example, once we get a really good travel agent, you push a button, it replicates, you know, and it could call you and it could do everything like that. So we have to consider what is money, and this is happening at a time when, you know, the petrodollar has shifted, so the energy equation is kind of done and others. And we have to consider, if we live in this world of infinite abundance with billions of robots and trillions of agents, what is money? Because money is a post -capitalist society in the long run.

1:33:23In the long run, and just like in the foundation or whatever, that interim period can be very, very messy because it's a zero -sum game where people are scrambling over that. But when anyone can live like a king, what is human progress? I think it unlocks a massive amount of creativity. It locks a massive amount of other things. We can eliminate hunger. We can eliminate disease. We can do all of this. We can come together and create that human colossus. But again, money is there because we need an intersubjective thing as a unit of account. But when you have AI agents, the combata, that changes things.

1:33:59When you have people embracing this and out competing on a corporate basis, that changes things. AI's themselves and robots are corporations are technically people in some senses. Like they have same rights. So AI's already have rights if they can be a corporation. And in Wyoming they passed Dow legislation, decentralized autonomous organization. So you may have AI's that go out and have rights, which is kind of crazy to think about. But against this all it's, I think our monetary base has always been tight to energy and entropy of that energy to create things. Physical and knowledge -based, I think that fundamental equation is challenged and so we need to rethink how our economic flows go and the rights around that as well.

1:34:47And that's difficult and it's hard. That's why I think every nation needs their own AI experts, AI teams, we need to get the spot as economists in the world to think about this and think outside the box and say is it universal basic compute, universal basic income, universal basic jobs. I think we need a trillion dollar jobs program for the graduates today. I felt this when I was in India, right? You know, the IITs or graduating, it's like 12 million graduates a year, and you're seeing the number percentage of them getting jobs. And in youthful nations like India and Africa, if you are getting your degree and you don't have a future, that's a lot of, you know, let me name it, testosterone, that is unchanneled and that becomes a very difficult future.

1:35:37So social challenge, like the start of four couple of months ago is that 38 % of the current IIT batch, top notch in India, were didn't have job placements. It'll be 50 % next year. But then even in the West, we have all these programs to repurpose people into STEM education. Like, aside, programming jobs, programming will change. You know, what do you do with truckers when you can have trucks as autonomous vehicles? I think the truckers, the truckers, buy the autonomous trucks and have them work for them. Because this is a divot. Yeah, this is a dividend type equation, right? Where by the nature of capital flows in our society when you have this influx of supply, will the demand catch up eventually?

1:36:19Well, I think, but it'll be very messy going through that. And on the other side, it's about what access do I have as an individual to capability? And then how can I create value from that? And I think it's massively positive sum, but it's really tricky to see the transition. Like I struggle to see how in 50 years time, shall we say, going a bit further normal, because you know, we tend to underestimate this. But if we even go to 50 or 2040 just before the singularity, right, as Ray kind of said, like what is money? Because you're extrapolating the formula, like the US can't borrow another 500 % of GDP, you know, but at the same time, intelligence is abundant, but also skilled manual labor is abundant.

1:37:12It's crazy to even think about, yeah. GDP is undefined. So therefore we must instead optimize for happiness, contentment, social gaussian. What do jobs look like? Is it taking the cancer empathy model and going out to your community and supporting families through that? That's a great job, you know? But we aren't going to go from, you know, most people have a job because they need to put food on the table or insurance for their family. It's not the job they dreamed about. So I think we need to start to disintermediate employment for income and employment for personal fulfillment. I think that split is going to happen sooner rather than later.

1:37:52And again, you've got this, things either for living or entertainment, right? And so we need to standardize the living aspects. Everyone can reach that basic level. And again, West may be a bit easier. But again, when you actually look at what you can build, it's amazing to think about. And then you need to think about how do you manage that transition? because there's vested interests and others when the whole world gets flipped and currencies start shifting because a lot of the anchors are broken. And this scares me because it's a complex system. And the only way I think you can do that is you have to build AIs to help us.

1:38:25How far away from that flip, that inflection point, where people, it becomes evident and people start searching very rapidly for answers. I think that in five years time things get very crazy. I think we're at the early innings of that. I don't see how things calm down because again, there's technology proliferates and then 10, 15 years past that. It's very, very short time in classical cycles. It can be longer, but again, this is an industrial AI revolution done incredibly quickly with multiple things from self -driving to autonomous agents to intelligence, all coming at the same time, as the end of mathematically our debt -fueled society that we've had.

1:39:13Again, Japan has 500 % debt to GDP. How are they going to borrow more? How is the US going to borrow more? And part of it has been a decreasing birth rate that's been impacting these nations, right? if a nation is growing in population and labor and automation, then it's producing more year on year. I think this is the really fascinating with China because everyone's been like, what if China gets AGI? What if China gets a hundred million or a billion robots before everyone else and produces the robots? That's what AGI is talking about. It's about energy one second, you've mentioned this a couple of times and I just was looking at a chart that shows the US has been generating for terawatt hours per year and it's been pretty flat for the last 20 years and it's projected to be flat.

1:40:03We're not increasing the energy production in the country and unfortunately generation two and generation three nuclear plants gave nuclear a bad name. At the same time China has tripled the amount of energy that it's putting out. India is on the path to doubling it. And one of the things that came out of Leopold's paper on situational awareness is the projection that the U .S. could use 100 % of its current energy production to power, it's AI needs by 2030. And so energy is a big issue. What are your thoughts about that? I mean, I think that that energy extrapolation is wrong because your chip constrained and I think scale and it gets you to some point, I think is an S curve that we're thinking is an exponential in capabilities, like distributed data and high quality data is far more important.

1:41:04But the reality is that we're really hitting energy limits across the entire base. And the new substrate like you and I, we need to pay for food, we need to pay for housing, we need to pay for our kids education. The AI's and robots of the future need to pay for energy and computation. So if we think about money, no, not data. I think there's a misnomer on the data side because once you have a high quality curriculum, you don't need more. Like the original GPT -3 paper was called language models are few shot learners. And this is what we see. We have a generalized model, like a GPT -4 or an anthropic clawed or a Gemini.

1:41:45It is a file that's trained on all of this stuff and then tuned, and it can adapt to any scenario through its context window, through its input prompt. Does it need more data? No. So you get a point of data saturation for a model that has capabilities of X, then Y, then Z, and then proliferation and optimization of that. Whereas what we're seeing with things like the Leopolder say is that you'll have an exponential of compute that continues. I think you'll have a glut in a couple of years, but nobody can take the other side of that bet. I think Mark Zuckerberg just said that a few days ago at Meta, we can't underspend in this cycle.

1:42:26Sunda Pitchhye said the same. The US does not have enough energy to meet the demands over the next few years, and even if you have a clut, then you'll be on the other side and you'll ramp up. Energy is intelligence for now. What it will be going forward, that's another question, and that will depend on things like video and other things versus language models, which I think a lot of it will go to the edge. I think it will go from these large dense models to mixtures of experts and highly optimised models, but it's very hard to see how energy requirements go down or the reallocation of energy occurs here.

1:43:05Like one of the shocking statistics is you know Bitcoin, I have all favorite kind of cryptocurrency. The total energy usage of Bitcoin is half of the energy usage of all the data centers in the world. Right now, 160 terawatt hours. It's as much as Argentina or the Netherlands. And that's just one use case. So if you think about it, like people who say that we're in a bubble and AI and everything like that, AI uses like a fifth of the energy of Bitcoin. You know, and Bitcoin is useful for various things, but it's not as useful as generative AI. So I think we've got many dabblings of the energy usage of gender, I'm still coming on a training and inference basis, running basis.

1:43:50What did you think of Ovalio Polz's situation awareness paper? Yeah, and one of things in particular, he in his paper and in his follow -on podcast that he did, he pits the US against China, and the existential risk is who develops AGI first. And so let's get into that conversation of AGI. Are we going to achieve AGI? When do we get it? is that a US versus China modality here? What are your thoughts there? I think we have a choice on AGI artificial general intelligence, nostril super intelligence of, is it a swarm collective intelligence that uplifts everyone as representative of everyone, which I think is a safe thing, or is a unitary intelligence based on, like just a few entities.

1:44:42I think the distributed intelligence, the collective intelligence is a far more powerful vision, and way of doing things. Again, open infrastructure enables that. And this is the level above the Lama type of infrastructure where you gather the people and you coordinate. I think the big AGI ASI is less likely. I think it's an extrapolation on the curve of capabilities that we just don't know about. But regardless, we're talking, I think, this is the Ray Kurzweil thing. right, what's his current forecast? Well, for AGI is 2029 still and for the singularity is early 2040s. I think that sounds about right, right?

1:45:20Like if you have AGI as capable as any human with a mixture of experts and physical embodiment, it's tough to argue against that, honestly. Given, and this is important, as much computation as you need for the inference step, A lot of the models we built today are trained on 10, 20, or 100 ,000 chips, but they're designed for consumers. And the energy that you use in running it is a few watts of electricity, or 100 watts of electricity, or a thousand watts. If you don't care about the amount of energy and you have a thousand or a million agents, then we're pretty much age -y now, but I think we'll get there there, where it's as capable of any human.

1:46:01The ASI singularity, I think, is a bit further, but again, is it a centralized one or is it distributed one? This is a question. One of the points that Leo Pold makes and in our Unite Discusses before is you know GPT, you know, two is a preschool or GPT -3 is elementary, GPT -4 is a high school or GPT -5 he describes as a PhD level and at that point when you've got an AI agent that's self - programming you end up with an intelligent explosion. Do you believe that to the case? Well, I think that, you know, we don't know if stuff will outperform its base data, but also I know that PhDs tend to be very depressed and sad people.

1:46:49Just kidding. But I think it's a reasonable thing. And this is one of the things I was worried about a lot last year, like I was on the signatories, I was the only AI signatory apart from Elon Musk on the pause letter, for example. Because I think we need a debate about it. But I think if we build high quality data going in, I'm not so worried about that. And I think that with the self -improvement of recursive stuff and the MCTS type thing, multi -collar tree search we see, the type of technology that goes into that deep -google, deep brain, deep -mind paper that got to sell the metal on the level on the International Math and Limpiad.

1:47:23It makes a reasonable sense. But will it go out of control and get self -awareness? I don't really think so. But even if it does, let's say that's the premise, I think this China versus US thing is a bit of a misnomer. Like for a start I don't think China wants that. But also I think that if you look at the components data, talent, compute, China beats the US. Because on data. Let's break those down. Yeah. Okay. China doesn't care about IP. You know, look at the video models. They're trained on all of Hollywood. I mean seven the US ones are as well. They'll train on They'll train on whatever and they'll train multilingual.

1:48:02So data, there's an advantage there, but also the Chinese via WeChat and these other things can get a hundred million Chinese to do data annotation and feedback, whereas OpenAI pays up to $200 per annotation for that data. The Chinese could easily do that for free. You look at compute. China has two X scale computers. So X scale is a thousand petaflops of compute. So Elon Musk's new 100 ,000 H200 cluster is two exaflops. The fastest supercomputer in the US, Aurora, is about two exaflops. The fifth fastest supercomputer is like a hundred to give you an idea. So we've got its order of magnitude increase.

1:48:42China has Tianhe 3 in ocean light, which I've publicly known about, that are both exaflop and at least four are the exaflop computers. That can run these giant models, and they can source the chips elsewhere. But again, when we're moving from this thing of big compute was a substitute for crap data And we can do distributed compute and data augmentation And that's the type of thing China is very, very good at. They're good at industrialization There are approaches now we're seeing with OpenA and others Where why did they invest in figure robotics for robots? Who's going to build most of the robots of the world?

1:49:18It's going to be China I think they're going to have their companies that compete against the US base. For sure. One of the things that's interesting is these humanoid robots become a source of data as well. We'll feed back into that data augmentation. But I think a large part of this is we've had this AI -AGI giant model fixation where I think it's more about the distributed collective intelligence that's constantly learning and adapting from real world usage. Again, a swarm of robots would be an ideal one for that, right? just like a swarm of Tesla's, a swarm of robots. And then we've had this fixation on really gigantic compute when it can be localized data optimization that then feeds in.

1:49:58If you optimize the data for models, you can have orders of magnitude less compute required. But we're not quite sure what that data optimization is yet. And I think, again, this is a race to this existential threat type of thing where it's trying to versus, I don't really buy that, honestly. You do push for international cooperation. How do we get international cooperation in AI? What would be, you're on the stage, you've got the nation's leaders around the world. What are you saying to them? I think that, you know, we're really seeing exoskeleton computers come on board with things like the EuroHPC initiative with the new Japanese exoskeleton computer, and, fortunately, Britain's shareholders won for the time being.

1:50:39Like a coordinated, deliberative effort to build high quality data sets and models for Humanity, generalised localised and specialised, like we're doing that anyway at Shelling, you know, and we'll be building national champions that can collaborate with governments and more. And then I think standards around input data. We care about the ingredients that feed us in our food, but not the ingredients that teach us and will guide us in the future. I think that any decision -making system should have data transparency. And I think that will go a large way to helping on AI safety, on AI alignment and more.

1:51:19And really thinking about what should we feed these models? We should feed them diverse data, not from a DEI or anything perspective, but from diversity of points represented humanity versus that snapshot on the internet. I'm sure that we have that done properly. Well, the conversations we've had as well is most of the corpus fed the AI models are Western, and if not US, and there's 100 nations out there that don't have their data digitized. In other words, the data of the grandfather and grandmother is in verbal language and it hasn't been digitized in a way that it can be consumed by an AI model.

1:52:00For me, that's a massive public works project that countries could take on. 100%. I think that we're going to be doing that with cryptocurrency, other incentives that Shelling. The goal is to build a data set for every nation and be massively collaborative as well as for individual sectors and science and more. Only more details of that soon as a self -reproved system. But we can think of it like, you know, we send a probe out to space right what's on that that represents humanity? We engage with the new species because we've got a new species, this AI Atlantis, a new continent. Are we showing them one cultural point of view, what's on that data hard drive that we send them?

1:52:41What do we feed in? And again, who do we want to represent? And I think you need to have this hive mind concept. You need this collective intelligence concept because that's how we proliferate and flourish the society. I don't think it's about scaling up a liberal arts polymath. You know, I think it's about building a hive collective intelligence and nearly again others again, a huge thing. around that and having it reflect the best of humanity and everyone contribute what they think the best of their culture is because then these models can translate between cultures and I think that's just beneficial full stop and that is the education system it's more.

1:53:23Yeah helping people one of the things I find AI so incredibly useful is if if there's someone on the opposite side of an issue or someone who I'm not able to clearly understand their motivations, AI can help me communicate with that individual in a way that I never could. Please help me explain my position on guns to someone on the other side of it in a way that they might be receptive to it. That translation of not just language but translation of intent and desire. It's a knowledge translation, it's a context translation and again, this is the universal translator. That's what these models are.

1:54:04And again, I think we just deliberately get together and build that out and realize, I think this is very important. The Leopold SA assumes a negative sum game. Again, there's fantastic things, I think 80 % of it's great. This is a massive positive sum game. And most of the AI discussions are negative sum, which I think they should have. Where does he point out negative sum game? There's a negative some game in that we must be the first to AGI because then it contracts everything. As opposed to building open common infrastructure for everyone, this is a competitive race dynamic with unstable equilibrium.

1:54:41I mean, I think he's point of view and includes the potential for a hard takeoff and at that point, if in fact we have an intelligence explosion, the person who gets their first It's dominated and it's a question of do we want the US and its national partners getting there first? Well, I think this is interesting because it's reflected in the Sam Moltmann recent op -ed in the Wall Street Journal. We talked about authoritarian AI versus democratic AI, but then recommends that it's locked down and it's restricted and it's basically unitary and it's point of view, which sounds very authoritarian China is producing open source AI.

1:55:21That's as good as all this stuff already. And it's a complex thing. But again, I think that there should exist open infrastructure and a collective approach as well as this collective approach, shall we say. Only hit on two final points. Investing in AI. A lot of people right now are like, I know AI is huge. I know it's the most important part in the world for us economically. We're seeing these companies from Nvidia and Google and Microsoft and dominate the stock market. And it's like, if you look at these companies as a percentage of the global GDP compared to other market sectors, they're massive.

1:56:03And so the question is, how does an investor think about it today? Where should they be putting their money? do they continue to invest in the giants or are they overpriced? What's your... I'm not sure if I'm asking for investment advice, but I don't know how should someone think about it. Since your paper is about how to think about AI, how should they think about investing in it? That's a separate thing. Putting on my former hedge fund manager hat. Okay, let's do that. So look, I think that the first wave was kind of these Nvidia's and kind of others. As Microsoft and Google set show their app like 20 -30 % over the last year, maybe 50 % is not a huge amount.

1:56:43In line with market, Nvidia has been the standout because they established market dominance they capitalized on it, but it's not expensive like Cisco was in the dot -com boom, for example, as an infrastructure provider on classical terms. But what I think is happening here is that we're going to have an overbuilt, regardless, obviously we will, because they can't afford to underbuild. but then we move on to the application and implementation of this technology by Centres. And so highly regulated industries or industries of the pricing power that can replace, wrote human work with AI Atlantis has super normal margins.

1:57:24And then you see a margin expansion. The impact of this is how much will it cost to... When we had cloud computing, it reduced the cost of building a startup dramatically. We had to wire up our own servers and everything. Now you will have an AI agent in the box that can help you build a hairdressing business or AI business or whatever, almost there in the next few years. You have this explosion of creativity. AI enabled work, I think, is where you see this the most. Pricing power is one thing. Industries that can have this on the other side to increase their reach. So you increase your audience.

1:58:03Again, where if this business existed, I could apply a million graduates, you know, liberal arts grads, and they'll be increasingly specialized how does that impact? And that should be a framework for looking at the business opportunities of any stock. And again, regulated industries with pricing power are a fantastic place for that. You don't need the private equity to come in any more and redo that. Do you think the current major players have been video Google Microsoft are overvalued or are these still of growth? On classical evaluation terms, they are perfectly fine on valuation. And again, we're seeing continued growth and it's not like it's crazy yet.

1:58:40Bubbles and past, we've seen crazy. This is not crazy yet. Like, when you're valued by the number of eyeballs in the dot -com bubble and things like that that we've seen. Second question here. This does point to something, which is how much of the global economy will be AI? A lot. As compared to what it is today, which is still a minuscule. Yes. And again, if we compare it to even self -driving cars, you still need five, ten times as much investment to catch up to self -driving cars. First is more of a training. My last question, my friend, is on the education front, a lot of people are looking, I need to get educated.

1:59:19You said this on the abundance stage. We've said this in our podcast that it's critical for anybody at any point in their career to become educated about AI. What do you think they should do? How do they start? I'm not seeing this happening in high schools. I'm sure there are courses and colleges. Do they just head to YouTube or do they just go to to Gemini or Chatchee PT and have a conversation with the AI? I think that's probably one of the things that should recommend to governments, which is standardized AI curriculum, on implementing and using things like prompting and other things. But really it is about just using it day to day and thinking of it not as the expert but the graduate.

2:00:01And so there are a few of these things, the mid journeys and the chat GPTs in other of the world. Like this is where bringing the next generation is useful. You've got your chief AI officer but also your almost chief AI innovator or tria. That's just hacking and trying and thinking how can I apply this to the day to day workflow. Again, you have things like your lindies of the world and your others that can chain together different models. I think that's the next stage, but you've got a little bit of time now. But you've got to get AI native by immersing yourself. There's no better way of doing that.

2:00:36And you shouldn't be overwhelmed by the technology or getting into the weeds because that is a massive, massive, depth thing. Again, that's why I wanted to create this piece just to give a bit of a framework and then encourage people to try and use the technology and think about the world in a slightly different way. You mad, what can we expect next from you? What's on the new term horizon here? So we're kind of finishing the implementation document of a white paper on what an open distributed AI system looks like on a practical basis to build these data sets for every nation and every sector and then chain them together into things that can really make it impact open source in a sustainable way.

2:01:16So hard at work on that. And then we start, will we getting our clusters into start building models? You know, nothing like it. And then release and see what the world makes. So today, title wise, founder of Shelling. Yep. That's the one. Are you going to take the CEO role or hire a CEO to support you? What do you think? Yeah, bigger CEO's like, he's going to send a staring at the investment sharing glass. Maybe I'll see if I can avoid it for as long as possible. That's why you build intelligent protocols. They can have you be able to get any other model. An AI model would be your CEO. Maybe I will actually, and that's a good idea.

2:01:53I think. We'll get lots of advice from AI Peter. Where can folks follow you on social and website and such? I think my ex is at E -Mostak and then at Shelling AI for our company organization as well. But I thank you for the time. Again, if you've not had a chance to read E -Mod's paper, how to think about AI, it's a very, for me, powerful, succinct and clear structure of thinking about this. I find it very compelling. We'll put the link in the show notes, and E -Mod looking forward to seeing you again shortly, my friend. Take care. Thank you, Namyom.