The Machines Are Taking Our Jobs - Thank God? Emad Mostaque’s Guide to the next 1000 Days

Today, my guest is Emad Mostaque, famously the founder of Stability AI and currently the founder of the Intelligent Internet and author of the provocative new book, The Last Economy: A Guide to the Age of Intelligent Economics. Emad has long been one of my favorite thinkers in the AI space. Very few people manage to grapple seriously and honestly with the world-changing nature of AI while also building something that matters in the here and now. But Emad has.

Since founding Stability AI in 2019, he’s demonstrated a deep understanding of AI technology trends, a keen eye for talent, the ability to inspire people with a positive vision for a wondrous future, and an appreciation for the stakes and risks, as evidenced by the fact that he signed, while still CEO of Stability, the famous 2023 pause letter. The fundamental problem that Emad addresses in the new book is that human society is built on the premise of scarcity. This makes sense. In humanity’s hunter-gatherer past, everyone had to contribute to the group’s survival, and freeloading simply couldn’t be tolerated. Still today, as Elon Musk puts it, if you don’t make stuff, there’s no stuff.

But what happens when an AI doctor, which doesn’t need to eat, can provide better frontline medical advice than a human doctor at 1/1,000th the cost? Such technology represents abundance for patients around the world, many of whom will enjoy better access to medical expertise than ever before. But taken to its logical conclusion, it implies poverty for human doctors. Similarly, what happens when all the cars can drive themselves and the millions of Americans who earn a living by driving are no longer needed for that purpose?

And zooming out, what happens when this pattern repeats itself across a majority of the economy, leaving displaced human workers with nowhere to go, all in less than a generation? Emad argues that there’s no escaping these questions. Even if AI capabilities stalled out today and we never got a truly powerful AGI, the AIs we already have, with proper implementation and integration into existing systems, are powerful enough to support this change. In reality, with frontier AI developers racing to build AI agents that are explicitly designed to replace human labor, we have maybe 1,000 days to find good answers.

With that in mind, Emad is simultaneously working to assemble the open-source datasets and train the small models that are needed to ensure that this abundant future is accessible to all, while also trying to answer the question of what the future and the transition to it might look like in as much concrete detail as possible. In this conversation, and even more so in the book, which I do encourage everyone to read and ponder, Emad coins a number of memorable terms, including the intelligence inversion, the metabolic rift, and the abundance trap. He also proposes a new way to think about the health of the economy, which would measure not just the monetary value of the material goods and services sold, but also the levels of intelligence, connectivity, and resilience in the system.

He makes fascinating analogies between the mathematics of neural networks and the economics of firms and markets, and even proposes a new dual-currency system: one for physical goods that are rivalrous in consumption and intrinsically scarce, and another for intangible goods that are non-rivalrous and fundamentally abundant. The realist in me recognizes that these are underdog ideas. But as you all know, Harari has famously explained, the stories we collectively tell ourselves are a huge part of how society operates. Money itself is a shared fiction, but a useful one, because it helps us allocate scarce resources relatively efficiently.

So, the idealist in me says that, in context, Emad can’t be any crazier than whoever it was who came up with the idea of using gold or shells as a medium of exchange in the first place. Big picture: While I usually tend to assume that the economic upside of AI will take care of itself, I think it is important to recognize that what Balaji calls the nuclear outcome—where we get the weaponization and constant threat from AI without the material abundance and accompanying personal freedom—is still a real possibility.

So, can we collectively start telling ourselves a story of abundance in which a person’s right to a decent life isn’t predicated on their economic contributions? And in which caring for one another isn’t something we do to meet our own needs, but because such interactions are a core part of the human experience? Can we do it in time to give people something to believe in before the inevitable modern Luddite movement shows up and tries to shut the whole thing down?

While so many people, myself at times included, are focused on the latest model updates, on the horse-race coverage of who’s winning and losing, and on making our apps work, Emad invites us to stop thinking so small, to recognize that we have agency, and to challenge ourselves to intelligently imagine and intentionally build our own shared future. Or, as he puts it in the book, “The machines are taking our jobs. Thank God. Now we can get to our real work.”

With that, I hope you enjoy this challenging and inspiring conversation with the one and only Emad Mostaque. Emad Mostaque, founder of the Intelligent Internet and author of The Last Economy: A Guide to the Age of Intelligent Economics, welcome.

Thanks for having me back.

Yeah, welcome back, I should say. I’m excited for this conversation because one of my common refrains, as regular listeners will know, is that the scarcest resource is a positive vision for the future. This book, which you describe as an engineering manual for building the future, is a combination of a diagnosis of a bunch of things that are going wrong in our society today and also some vision and even recommendations, some of which are fairly opinionated, about what we might do to build a much better world. I really applaud you for taking on the challenge and doing the hard work of putting something like this forward.

It all is really based around this notion of intelligence theory, and maybe a good place to start is just giving you a chance to describe: What is intelligence theory?

Yeah. Basically, I’ve been thinking a lot about what the new economy looks like, and we’ve seen existing economics might be challenged, et cetera. So I was like, let’s go back to first principles. At my previous company, Stability AI, we built Stable Diffusion and other models. They had hundreds of millions of downloads, and they were getting better than humans at doing various things. Now we see that with the new models that are coming out, agents, et cetera.

Intelligence theory basically goes back to a principle. I was like, what is the core axiom or principle that defines reality? There was this observation of persistence: Certain complex adaptive systems persist over a long time in uncertain environments. The ones that do that the best are the ones that basically have the closest match between their internal model of reality and reality itself, which looks like the loss function in generative AI. In fact, mathematically, it was the same.

So intelligence theory is that the ones that do the best are the ones that minimize that loss, or surprise, but have the best models. Again, we see that in everyday life, and we see that from the fact that generative AI itself has created the best models of reality. The best agents now are AI agents. The best models of reality are AI models.

Then I was like, can we have economics deriving from that one base principle? What does the mathematics of that look like when we apply the equations of generative AI, which came from physics, to economics? So we started building a new economics from that basis, as opposed to the classical economic bases of scarcity, the concept of utility, general equilibrium, or other things that can’t even be measured, which have built up over hundreds and hundreds of years. Those theories always assumed that humans were going to be the top and the main producers, which may not be the case anymore in a few years’ time.

So let me ask a couple of really naive questions. One, I have some basis in this idea that predicting your environment is really key to acting effectively in the world. One of the best blog posts I’ve ever read—I think it amazingly goes back to 2017—was from Scott Alexander’s old blog, Slate Star Codex. It’s called “Predictive Processing and Perceptual Control,” and it’s basically a book review of a very long, dense book that I think he does a great job synthesizing into the idea that a simple model of us as biological humans is that we have a lot of layers of prediction happening between our peripheral neurons that receive the signals from the world and our highest-order neurons in the prefrontal cortex.

Along those many layers, the role of each layer is to predict what’s about to happen. If the signals that it’s getting from the lower level are consistent with the predictions that it’s making, then it can just be quiet. This is how you can put a lot of things in background mode while you focus on whatever you’re focusing on.

But when those predictions and the signals that are coming in diverge, that is surprise, and that is what calls your attention to things. That’s what gets things escalated up the ladder into your conscious awareness. I felt that blog post clarified more for me about what’s going on—why I’m experiencing and perceiving what I’m experiencing and perceiving—than maybe anything else.

It’s all about predicting what’s about to happen and making sure that you’re in sync with the environment around you. What counts here? One might say, “Well, a human can last for 80 years, and a giant tortoise might last for 150 years or something.” But if I set a rock in a quiet location, I could come back 1,000 years later and it would still be there. It doesn’t seem to be predicting anything.

How do we know, or where do we conceptually distinguish between things that have this capability and those that don’t? Again, this might be super obvious, but sometimes I find these apparent binaries are, in fact, a lot blurrier. I thought it was worth asking.

Yeah. This is the nature of complex adaptive systems: systems that are in motion, where information flows through the interaction of different agents. If we go a level down in intelligence theory, it basically says that the ones that succeed most are the ones that minimize computational overhead. A rock doesn’t need to compute anything. A rock just is, right? It doesn’t really do anything to exert action on its boundaries or anyone else.

If you look at general complex entities and agents, intelligence theory has split this up into 3 different types of things. The first is predictive error, which is the mismatch between the model and reality. That’s surprise. The second is the model’s complexity itself, the cost of thinking, because the more efficient you are at that, the better you’ll do. You read Scott Alexander’s blog post, and it gives you a mental framework for the world that allows you to process things in a different way. Hopefully, this book does the same.

That’s similar to latent spaces in a generative AI model as it kind of folds it. The final thing is the update cost, which is the cost of learning. A rock doesn’t have any update cost. It doesn’t have to learn anything because it doesn’t enact action or anything, and it has no capability to respond, either.

The human brain with neurons—I mean, the equations here are very similar to Karl Friston’s free energy principle. If you look at the cost of predictive error, model complexity, and update cost, that’s called free energy as well. You’re trying to minimize this concept of free energy. You’re trying to optimize computation as an agent that can act, and the best AI models do the same. It’s all gradient descent, trying to optimize that and minimize the loss function.

Again, you have this flow of agency, this flow of interaction, but this framework only applies to these complex adaptive systems. It doesn’t apply to static matter. Who knows? Maybe we’ll find that information and intelligence are related, which is why we’ve got things like wave-particle duality, et cetera. But that’s a long way from where we are now.

Where we are now is that we’ve built our whole economic picture based on assumptions from 200 years ago, when most of the value was the land and the serfs that you had—Adam Smith’s scarcity and other things like that, The Wealth of Nations. But now it’s the wealth of robots. We need a better way of describing the world today and the world that’s about to come.

That’s a world in which Nat Friedman—I was doing a panel with him a couple of years ago, and he coined this concept of AI Atlantis. There’s a brand-new continent with a trillion agents and robots on it, and it’s about to enter the workforce. What happens?

Yeah, there are some big questions there, for sure. Let’s put a pin in the math and circle back to it opportunistically as we go through the diagnosis, recommendations, and visions in the book.

I would love to take one extra beat, though, on what’s going on in the discourse. It seems pretty obvious to me that this AI thing is going to be a big deal. The Atlantis metaphor, or the Gen Country of data geniuses in a data center, certainly resonates with me. It seems like that’ll be a big deal. Yet we’ve got all sorts of smart people, including some who are at the top of the field in AI, but also folks like Tyler Cowen, where he’s like, “AI could accelerate economic growth by 0.5% of GDP per year.” That would be amazing, and we shouldn’t underestimate it. He thinks that’s a big deal.

How do you make sense of why people see this so differently, and what do you think people are missing when they put an upper bound of 0.5% of GDP per year on the AI phenomenon?

Well, this is the update cost—the cost of learning, right? There’s quite a high update cost to your priors when big things happen. When COVID was about to happen in January, I was like, “Oh my God, the world’s going to crap.” A few of us said that and posted about it publicly. I did podcasts and stuff. Most people were like, “It’s fine,” until Tom Hanks got it. Then you had a phase shift in the way things were perceived.

We have to remember that the pace of what has happened is unlike anything we’ve ever seen before. It’s been 3 years since Stable Diffusion, just over 1,000 days since ChatGPT. It’s been 1 year since o1-preview was announced, and it’s just over 1 month since GPT-5. Just over 1 month ago, the vast majority of AI users in the world were using GPT-4o.

That’s your benchmark. I think I saw some statistics that 20% of Americans still haven’t heard of ChatGPT. You think about that, and you’re like, technology takes a while to diffuse. It takes a while to update the priors. But most people are still thinking about the previous generation of AIs that could think instantly and hallucinated all over the place.

Those of us who are right at the cutting edge—I’m running Codex right now on the CLI, and it’s been running for 3 hours building a whole textbook website for my textbook. I just set it and forget it. The capabilities have just gone up again exponentially because they’re breaking through from not being quite good enough—“Hey, why doesn’t this AI transcribe properly?”—to suddenly being superhuman.

So it’s that transition phase. Classically, you’ve been held back by various constraints, like robots, for example. You’re not going to be able to build enough robots because we won’t have the spare parts. It takes time to build factories. The difference with generative AI is that you already have the hardware. You just have to build the interfaces and the flows properly.

Yesterday, Alibaba’s Tongyi Qianwen Lab released Qwen3-30B-A3B, a 3-billion-active-parameter model with 30 billion total parameters, that outperforms Grok 4 on Humanity’s Last Exam and outperforms Deep Research and all these massive models. It has 3 billion active parameters, which for listeners means that you can basically run it on a CPU with 16 GB of RAM. It’s outperforming these frontier models. That’s crazy.

What that means is that, reasonably, our medical model, iMed, is 8 billion parameters. It needs 8 GB of RAM, and it outperforms human doctors. We don’t think it’s good enough yet. Even though it outperforms human doctors, by next year it’ll be better than any doctor, with full traceability, and you’ll be able to run it on any smartphone.

How do you calculate the impact of that in classical GDP or economics or other terms? You’ve never seen anything like that. But there was a recent MIT study that showed that 95% of corporate AI deployments haven’t worked because they’re all running the last generation of models. The last generation is 6 months old.

I think it’s this inflection point, this takeoff, that we’re basically at now, where models and systems can go from seconds of thinking to almost infinite length, where they can check their errors and adapt. Hallucinations have dropped dramatically, and they’ve finally broken through on the IQ level, as well as being able to view your monitor and check everything.

You’ll need to know about AI to put all those pieces together and realize what we’ll have 6 months from now, 1 year from now. The way that you use AI is, you give it a call or you have a Zoom with it, and you can’t tell if it’s human or AI on the other side. That’s the economic, social, and other disruption that we have, because the cost of doing that will be a few pennies an hour—a dollar an hour, shall we say. No one has that in their numbers, because everyone was like, “We have to build giant supercomputers with huge models in order to achieve this AGI performance.”

I don’t really care about general intelligence. The real impact is actually useful intelligence. The real impact on the economy is not a polymath coming up with a brand-new thing. I’m sure we all have that. It’s basically someone just blindly following instructions. It’s what I call the cooks versus the chefs.

Wait But Why had this when discussing Elon Musk: everyone’s on this spectrum, from chefs who come up with recipes to cooks who actually do it. The cooks are the ones that will impact the economy. People aren’t realizing that you will have these virtual, and then physical when robots come, colleagues that can just do things.

If I was looking at the technology 6 months ago, I’d be like, “Yeah, of course it can’t.” Today it can.

Yeah. There are a couple of ideas there that jump out at me. One is the distinction between 0 to 1 and 1 to N, to put it in Peter Thiel terms. It sounds like you’re saying maybe the frontier minds are just inherently more focused on the 0 to 1, which we don’t really have yet, and so they’re kind of skeptical of it.

They may be underestimating the importance of—I think I’ve heard you call it “satisficing” in the past as well—the 1 to N that delivers something in a consistent way to everybody. In terms of short-term impact on daily life, that might even be bigger.

And then there's also this incredible cost curve that we're on, where the original GPT-3 was $60 per million input tokens. GPT-5 is $1.25 to $1.50 per million input tokens. So it's literally a 95%+ reduction in cost while, at the same time, obviously being dramatically better.

That's a hard thing to count in GDP. We've tried to do that over time with, well, your cell phone's a little bit better—how do you adjust for this in a standard basket of goods type of thing? But it does seem like there's a pretty good argument that—and this also could, in some way, I don't think this exactly reconciles with Tyler Cowen's view—in some way, it's like, yeah, maybe this isn't going to hit GDP, but maybe that also suggests that GDP is just straight-up the wrong measure.

And that definitely gets into a lot of the forward-looking ideas that you have in the book. I think you do a great job of coining highly memetically fit terms. We'll go through a number of them over the course of the hour here.

Well, let's start off with this notion of the abundance trap and the metabolic rift. I think both of these start to get at this idea of how economic activity, as we've traditionally thought about measuring it through something like GDP, is on the verge of breaking down.

Yeah. So we're at this really interesting point where the abundance trap is where we're going to achieve post-scarcity in the realm of intelligence. Intelligence becomes abundant, right? Again, we've seen these big changes: when we had the Gutenberg press, suddenly people could read and have access to intelligence, but it has traditionally been gated.

Yet by next year, everyone in the world, if they have a phone, will be able to have an expert doctor's opinion that actually outperforms doctors. That's crazy. Everyone has access to legal advice from Grok that's better than most legal advice. And again, they don't make mistakes like doctors make—20% errors, because in most cases, I think that's the average.

So the abundance trap is that we're going to have this disruption, and then the economic system that's based on scarcity is going to process this as poverty, because you will have job losses and other things. Even if our lives will be getting better, because these new systems—you might see corporate profits, et cetera, go up—they will be displacing knowledge work.

They'll displace knowledge work because you'll be able to hire employees on the other side of that virtual screen: KVM jobs, I believe they're called—keyboard, video, and mouse—that don't sleep, don't make errors, and just constantly learn and improve. And the metabolic rift here is that the GPUs don't need to eat, they don't need housing, and they don't pay taxes. In fact, they're actually tax-deductible on the usage. And you can get them by the hour.

So this is the rift that occurs where, all of a sudden, you have this explosion of intelligence. You have this abundance, yet it's probably going to be bad for us in aggregate unless we allocate it correctly. The metabolic rift is that these things don't need to eat. They don't need housing. They don't consume. The only thing an AI needs is to achieve its objective function.

We all talk about AGI, and I've got Eliezer Yudkowsky's new book ready to read. What is it called?

If Anyone Builds It, Everyone Dies.

Yeah, we're not talking about AGI here. We're talking about AI accountants, AI lawyers, AI designers, those types of things. They don't consume in the same way. And that's never going to change.

Once it gets smarter than a human, it's done. It's not going to get dumber ever. Once it gets more capable of executing—which, again, is the nature of any organization—it's just an executor. You have a framework: money comes in is less than money comes out. They will out-execute humans. And again, that's never going to shift as this kind of final inversion that we've had.

Yeah. Just to put a little quantitative intuition around this notion that GPUs don't need to eat—they do need electricity, of course—one of the things that I have been surprised by myself, and consistently surprised others with, is how much energy a cell phone battery or a laptop battery can hold.

A cell phone battery typically is somewhere in the neighborhood of 20 watt-hours, and a laptop battery is somewhere in the neighborhood of maybe 5 times that, like 100 watt-hours. Obviously, it depends on your model, et cetera.

Mm-hmm.

But 100 watt-hours is, in my neighborhood here in Detroit, Michigan, under 2 cents of electricity. We pay something like 18 cents per kilowatt-hour. So 100 watt-hours, a tenth of that, costs less than 2 cents.

Now, when you think about your expectation that we'll have—well, of course, we already have models that can run on my laptop for some amount of time—the ability to run a model on my laptop for a couple of hours, whatever, for an energy cost of 2 cents does start to put an intuition, I think, behind just how much economic advantage these things are going to have.

And then you've got the on-demand spin-ups and spin-downs, all the other unfair advantages that they have as well.

It does suggest a Malthusian competition that's going to be really hard for humans to compete in. Well, I mean, human intellect is capped, right? You'll get a bit better, but these models can just continuously get better in aggregate. And as I said, the cost of doing an activity is minimal.

A practical example is that some of the people listening to this may have built their own websites or paid someone to do it. That would have cost thousands of dollars, or the equivalent. Now you can go to Replit.com, have a chat with the new Replit Agent, and within a day you’ll have a website that’s probably as good as the one that you built. The cost will be around $20 or $40, and that’s with Replit’s margin in there, right?

That cost will drop by 10 times by next year and then 10 times the year after, just from the speed-up of the compute chips. We saw this with images: now, if you use Nano Banana and Imagen, for a couple of cents you can make just about any image that you want. How much would that have cost before? What we have is this big displacement of classical capital across the board because the cost of creation suddenly goes to zero.

The cost of consumption in the previous internet age went to zero. Now the cost of creation is going to zero, and the quality of the creations is actually better because the AI, through its latent space and its mapping, actually understands aesthetics and things like that.

This is what I call the intelligence inversion. First of all, you went from land and the serfs on it, and then it was about how much labor you had in terms of muscles. Then it was about the capital you had, be it industrial capital or software, SaaS. Now there’s this intelligence inversion where you’re outcompeted on intelligence. More than that, taking something and making something from it digitally now, and physically soon, means there’s nowhere left to pivot because we kind of pivoted up the stack.

We own capital. We don’t need our muscles anymore. Where do we pivot now? That’s a big question mark for us. What is our purpose? How does the economy run when the marginal productivity is all AI-driven?

And this is before we take into account robots and robotics, because those are getting freaky. That Unitree robot—I don’t know if you saw it a few days ago—they pushed it over and it just got back up in 1 second. You see the dogs kind of chasing them. You see them making recipes.

Then you calculate: if you have a Tesla Optimus robot for $20,000, and you work it hour in, hour out, it’s $150 an hour for an Optimus. You’re like, okay, that can probably be a plumber in a few years. It’s just coming across every part of the economy, bit by bit by bit.

The cost is the electricity, but the electricity costs are way lower than most people have in their projections, because everyone is again just thinking about these AGI models. We saw what happened with GPT-4.5 when it came out. It was too damn expensive. Everyone just wants to use the cheap ones.

What is the appropriate price for 1 million tokens, which is like 800,000 words? It’s $1.50 now. What if it’s $0.15? Amazon is at $0.01. That’s crazy.

On this point about nowhere left to pivot, I think this is a kind of echo of—you all know Yuval Noah Harari, who I associate a similar argument with. The one thing that people often bring up in terms of what might be next, or what sort of refuge in terms of value creation people might find if our bodies are being outcompeted by machines and our minds are being outcompeted by AIs, is taking care of each other.

This is sort of the caring economy, the teaching economy, the mentoring economy. Another one is generally creative pursuits. I think creative pursuits maybe feel more like leisure to me, probably in an AI future. But this caring one in particular—I could imagine that people might have some sort of intrinsic preference for other people to care about ourselves and our kids and our parents and whatever. Maybe we don’t turn that over to robots.

I gather you don’t think that’s really a viable or sustainable place to concentrate activity, but I’d love to hear a little bit more of a fleshed-out argument for why you don’t see that as the next evolution.

We have to look at the economic flows and the nature of it. If you think about the Fed and central banking—until the president takes it apart, which may happen—what’s the Fed’s mandate? It’s employment and inflation.

What it does is raise interest rates when inflation is going up, which raises the cost of borrowing across the board. That reduces consumption, which reduces inflation, and it reduces hiring. When it drops interest rates, companies can borrow more, people can spend more, and people hire more because of those things.

That breaks, because when you drop interest rates, now people hire more GPUs, et cetera. These are the economically productive white-collar and above areas of society, which can have a big delta. It’s the knowledge economy that’s being disrupted.

Things like having a walk with your kid—AI is not going to replace that, right? The intersubjective stuff of hanging out with your friends, or learning a new skill, or enjoying a concert—I’m not going to go to a concert done by robots. I’d prefer to go to a concert with people. This is the Taylor Swift economy, as it were, because it’s about socialization.

What’s happened here is that the nature of our work and our jobs has really changed. You need the income, the kind of capital, for survival. But then it’s also become a core part of our identity, where we moved from, like, Emad, son of Khaled, son of this, to that—a network-based identity—to one where I am founder of Stability AI. I’m a CEO. I’m an AI guy.

You need community, and you’ve had that in the workplace, but there’s been this hollowing out of community, both through social contracts and socially, locally. In the good old days, as it were—showing my age—you used to know your neighbors, and the kids used to hang out. That happens a lot less now, particularly in the cities. Religion has gone down, and that was a core part of community as well. You didn’t need to believe, but again, it was supportive.

Then you have this purpose thing: do what you like, do what you’re good at, and do where you’re adding value, and other people believe too, and you’re happy in the middle. It doesn’t matter what it is, right? It can be that I’m in a World of Warcraft Discord. It can be that I’m in a workplace. It can be that I’m in a competitive setting, playing tennis with others, or whatever.

The final thing is structure. People need a bit of structure around what they’re doing. The caring economy, the sharing economy—we could have that emerge. It’s just: how does the mathematics of these things work?

If you look at universal basic income, a lot of people have said $16,000 is the US poverty rate. If we gave every adult in America $16,000, the cost would be $5 trillion. The entire tax base of America—all the taxes that are brought in—is $5 trillion. Corporate taxes are $0.9 trillion. So when you say, “Tax the AI,” it doesn’t even work, right? We need to rethink how money flows.

Our purpose is basically more around measuring what is important. It’s around increasing the network effects. It’s around going back to kind of where things were, which is a very interesting thing, because in a Star Trek post-scarcity world, what do they do? They explore, they improve, they adapt, because they don’t need to worry. They can just 3D-print anything.

That’s the ideal abundance-type future. Why do you need to work to live, effectively? At the same time, our current system means that you have to work, and our social safety nets probably aren’t going to be good enough.

There will be this really interesting transition period. Not every job will go. It’s nice going to the barber, and nursing, and all education—all these things, I think, will adapt and change. It’s just: can we have the capital flowing appropriately so that people who are already fed up with the social contract breaking down don’t get absolutely crazy because they’re being left behind, and all of these excess returns go to the capital owners, leading to more inequality even if the numbers go up?

So, could I summarize that in terms of the original objection or question? We went from physical labor to cognitive labor; maybe we can move to caring labor. It sounds like you're saying, yes, there probably is still a unique role for humans there, if only because we intrinsically value the fact that that kind of activity can come from one another. But I'm not exactly sure how to summarize the “but.” It's like there's not enough space there for everybody, or if we try to pile everyone in there all at once, it just won't be possible.

Can we do that in the current format, or can we adapt the current format so people have the freedom to do so? If you think about lockdown, lockdown for me was awful because I was doing coworking and working around the clock. Some people got closer, some people broke apart, but it was interesting in that you had your little bubbles, et cetera. You were almost forced to, and people suddenly had some time to think.

What you've got right now is, again, if you lose your job and you have a strong network, a community, then you can have your identity fall back to the identity of your community and your connections there, and you're supported. If you don't, then you won't. Again, we've seen a hollowing out of this network-based identity. It's become more about what brand you're a part of. A fellow Apple owner won't help you out, but a fellow Swiftie might.

So I think that's the future. What I say in the book is that computation and consciousness were tied together in humans. Now computation and consciousness are different. Consciousness is the domain of humanity, and we've seen lots of discussions around this recently: Why is this beautiful? Why is this just? Why is this meaningful? That's the nature of this caring economy. It's a question of why, as opposed to a question of how.

This is why I think the cook analogy is the very interesting one. We make meaning. We have a certain amount of attention, and we need to maximize that as well. So I think it's just a transition period, and then the question is: How do we give enough support and a new social contract for people to become the meaning-makers, to become the network connectors, to do these things and support each other appropriately?

It's been 1,000 days since ChatGPT. Next year is the year of tipping for knowledge-work jobs. Again, you'll be able to just chat via AI—I don't know how that can't be the case. A thousand days from now, I think the world looks very, very, very different, and that's not much time to come together. So what is the process of that?

Yeah, gotcha. I was wondering if you would make an even more aggressive argument, which I think you probably are somewhat sympathetic to as well, that basically boils down to this: AI doctors also get higher ratings for bedside manner in many studies than human doctors. We're starting to see things like Alpha School, where all of the content is delivered via AI and the adults in the school have become mentors, guides, and coaches.

One might wonder, what happens if the AI becomes as good as or better a mentor, guide, or coach than the people? But it sounds like you're sort of saying, in your view—and I do think it's worth lingering on this for a second, because the vision of the positive future is so core to the value that you're offering people here—that those are intrinsically good things. They're good for the people to do. They're good for people to receive. Maybe AI can do it better in some ways, but we don't necessarily have to choose one or the other.

The question is, how do we transition to a future state of society where people are not caring for others out of a scarcity-driven economic need, but are able to do it because it's part of what it means to live a rich life, even assuming that you have material abundance?

Yeah. This is interestingly analogous to spirituality across the major faith traditions. Most of them have this thing where you basically go through the process, you learn a bit, and then you become a bit of a douche and tell everyone you know about it, and you go on top of a mountain or achieve nirvana or enlightenment. But the end state is not being this bearded hobo on top of a mountain. It's coming back down the mountain, and your interactions with other people are what are meaningful again.

The majority of the time you'll ever spend with your parents has been spent, and that becomes more meaningful. You look at the connections you build through life, and you remember those interconnections and relationships. But the nature of current life—and our current systems—is designed to take away our attention from other people and instead direct it to other things, to brands and everything else.

In fact, one of the only scarce resources in the world is our attention. There's only a finite amount of human attention. How are you filling that? Are you filling it with your interactions with others, or are you focused on other objective functions? Have you heard of that parable of the fisherman and the investment banker?

I'll say no. I don't think so.

An investment banker retires very wealthy, and then he goes somewhere in South America and finds a nice beach. He's having a good time; it's 2 or 3 p.m. in the afternoon. He sees a guy with lots of fish on his shoulder, and he's going back. He's like, “What are you doing?”

“Well, I'm going back to hang out with my family, and we're going to have a fish fry-up, and we're going to talk, sing, and dance a bit. You're free to come along if you want.”

The investment banker says, “No, you shouldn't do that. There's still 4 hours of sunlight. Go and fish some more, and you can sell the extra fish. Then you can go from doing it manually to having a boat, and you can use those profits to expand. This is relatively unexplored, so maybe you can get a fleet of boats, and then you can scale. Maybe you can even list on the stock exchange.”

The guy's like, “Wow. And then what?”

“Then you can retire, kick back by the beach, spend some time with your family, maybe do a fish fry-up, and dance a bit.”

I think that, again, the hustle and bustle of current life and the attention-extraction mechanisms have taken away what it means to be human. Religion, spirituality, or whatever it is, are about our interconnectivity. We can actually build the models to help us through this, or we can choose to do the opposite.

You can build massively manipulative models. I get calls sometimes from my mother saying, “Emad, I need money.” She would never say that. She'd slap me around the ear. Someone has cloned her voiceprint somehow. It just takes a few seconds of audio. That's a bad use of the technology.

We're seeing this targeting and memetic stuff. A different use of the technology is support, coaching, and other things. Sam Altman is in a very difficult situation right now because I think he said something like 10,000 people commit suicide every month. How many of them have talked to ChatGPT? It's probably reduced suicides, but unfortunately some people commit suicide because of it. How do we support these people appropriately, and how do we support people in general with the AIs that we build when corporations align them?

Because of the example you gave of engagement and trust, imagine the person you trusted most in your life, and we created a virtual AI double of them. It only requires a little bit of data, right? You'd trust that AI more than anyone, honestly, and it would be with you more than anyone. But it doesn't take away from real human interaction with people physically.

This is a system-architecting thing. Are we increasing human agency and connection, or are we going to the WALL-E world of everyone with Apple Vision Pro 8 strapped to their faces, eating lots of food, with robots running around? That's a question we have for society today.

The suicide statistics—obviously an ongoing tragedy globally, but with particularly high rates, I think, in the United States—are maybe a good jumping-off point, or point of entry, into what you call “the harbingers and the lies.” Basically, for people who are like, “Wait a second, I read that life has never been better, and all these indicators have improved.” Certainly, many of those things are true. Infant mortality is way down, and we've got antibiotics and so on and so forth.

Lots of good things. But you do point to these leading indicators that suggest something is maybe on the verge of breaking in society. Some of these honestly seem like general problems of what some call late-stage capitalism. Some are maybe more specifically the result of, or will be dramatically accelerated by, AI.

Let me take us through some of the highlights of the harbingers and the lies that, for you, indicate—and I think your argument is that if you haven't been convinced by the theory, then these data points should make you take much more seriously the idea that we might be hitting some sort of breaking point before too long.

Yeah. It's like maybe Neil Howe's The Fourth Turning is coming. I think he predicted it would be around 2025. You've seen this kind of critical slowing down at the start, where stuff isn't synchronizing properly. You're approaching a critical transition.

You've seen things like debt accelerating, but you can't have more debt. We've maxed out on these credit cards as a society. Mathematically, you've seen this kind of variance explosion, where small inputs cause wild swings. I think this year AI is going to be huge. Next year, the digital asset explosion in the US will be the biggest bubble we've probably ever seen.

You see bubbles emerging everywhere as capital is trying to find a place to go apart from AI, and they're struggling because a lot of the stuff internally is kind of hollowed out. The other thing is this kind of flickering through these different states, like the gig economy. Are you an employee? Are you a worker? What is the nature of money? Bitcoin is suddenly money. A lot of these things are getting in the way.

Then you have correlations just increasing across the board, where something like the Ever Given can cause massive global supply-side collapses. We see systemic frailty increasing even as all these indicators are saying, “We're the best economy ever. The stock market is at all-time highs, with record profits, margins, and so on.” Well, then people aren't feeling happy. Again, depression and suicide rates are going high.

You're seeing cracks in what's emerging, and you're maxing out your various indicators here. The amount of impact you can have with the classical mechanisms now—if the Fed floods the market, it's not going to do much. The medicine is getting a bit worse, and a lot of the classical assumptions we have are going to break down.

One such assumption is that scarcity is fundamental and human labor has value. I think you mentioned earlier: What is the value of humans in that? What's the value of the dumbest person on the team? It's negative. Humans will be the dumbest people on the team.

Growth requires resources, but you can replicate this intelligent stuff infinitely with just a few GPUs. In fact, it wouldn't surprise you to see a 10× improvement in the GPUs with the same model. You have equilibrium markets: they go and balance and adapt. That might not happen anymore. They can break.

Finally, money measures value. I have a few more, but I think that's a very important one. The richest people aren't the happiest. You have a certain level where you need a hygiene factor, but then we all know rich people who are unhappy. There's no real correlation there. Instead, happiness comes through other things.

I think these are factors of late-stage capitalism. But at the same time, I don't really know anyone who's happy with the way things are or with the social contract as it exists, because something seems to be off. When you really drill down and talk to other people, something seems to be off.

And it's at a time when we're about to hit multiple of these crises at once, from AI to robotics to climate and others. We've maxed out all the resources we've had to navigate the previous ones. That's why we need to have a new way of looking at things, certainly because a lot of the classical assumptions are going to break down.

Yeah. The idea that money measures value has long been critiqued from the standpoint that money doesn't necessarily buy happiness, although there's also the argument that statistically it kind of does. But today there's also this much more obvious disconnect where the cost of my AI doctor is dramatically less than that of the human doctor. If it costs me $100 for the appointment versus 10 cents for the AI consultation, that really does create a huge disconnect in the notion of money measuring value.

I also thought the one that was maybe most compelling to me was the idea that systems in crisis take longer to recover from a new insult, and how we are seeing longer recovery times from recessions. To me, that does seem like—even though I don't have a fully principled understanding of it necessarily, maybe you do—I don't yet. That does strongly suggest to me something that is out of whack.

We've traded resilience for efficiency to an extreme, where we are now really vulnerable to perturbations that we might have been much more robust to in the past. I think we also saw this in COVID. It's become kind of a trope, but I think this is a theme that runs through the book as well.

Yeah. I think, again, corporations are slow, dumb AIs that optimize and chew up humans as their fodder. Our education system has been this factory school that prepared us for that. But again, you see organizational structures where people go in with the best intentions and get very unhappy very quickly.

There is this thing: You can manage what you measure, but then you adapt to what you measure. This is Goodhart's law. GDP was basically invented in the 1940s by Simon Kuznets, and he himself said this was a really bad measure of economics or societal well-being. But this is the one factor that we use.

What happens now is that, as economies optimize, you do things like offshoring. You do a lot of antihuman things. Meta, as an organization, will do an experiment saying, “If people post sadder things, do they become sadder? And if they see sadder things, do they post sadder things?” Then they just do an experiment on that. We see a lot of these very nonhuman actions by these corporations occurring more and more.

But a lot of that reduces our systemic resilience because, again, in Slate Star Codex, I think one of Scott Alexander's great posts is about Seeing Like a State, where he talks about legibility and how you bulldoze through villages. You reduce the diversity, you go to homogeneous cultures and monocultures, and then when you get impacted by something, you have no fallback, as you see with supply-chain disruptions and other things like that.

I think in the pursuit of maximizing corporate profits and maximizing GDP, governments and organizations make decisions that are not in the best interests of people, as slow, dumb AIs. Now we're getting to a terminal point on that, where our resilience factor has decreased dramatically because we've reduced diversity, reduced our network effects, and we're lacking in systemic intelligence.

I put a post up a few days ago saying it would be great if we had a common-sense GPT to just say, “This policy is obviously dumb.” We see so many of these really dumb policies with huge amounts of money, whereas very sensible things that cost not much can't seem to get any capital to make an impact.

I still find it funny, actually. I was doing a calculation: the Los Angeles–San Francisco railway, I think they've spent more on that than on all of the AI models put together so far in training.

It's like a mile long.

That's funny. Along with the energy-usage calculations from earlier, that does put the scale of resources that have been put into AI into an interesting perspective. We could linger on a lot of these problems and arguments that maybe they're not as bad as you think, or whatever, for a long time, but in the interest of getting on to the upswing of the book, let's leave that for now.

From here, I think we're headed into genuinely the prescription and positive-vision part of your thinking. Maybe tell us about what you describe as the three laws of a living system and then the Mind Capital framework that you've developed for trying to get a handle on a more holistic measure of the health of an economy—or really, any intelligent system. It certainly applies at the economy level.

Yeah. The three laws of living systems are things that we derived from the mathematics when we started looking at this in terms of generative algorithmic equations. The first one was the law of flow: value must be conserved and circulated. When you have a stagnant economy, or when people start hoarding stuff, money doesn't flow. Capital doesn't flow, intelligence doesn't flow, and other forms of value don't flow. Then you get stasis and eventually a collapse.

The next one is the law of openness, which is that connection fights entropy. When you have very closed environments—and I give the example of Tokugawa Japan from 1633 to 1853—you basically get these monocultures that become very nonresilient to any type of shock, such as Commodore Perry coming with cannons or whatever. Again, the less open and interactive you are, the more dangerous it is for you.

The final thing is this law of resilience, where again it's a question of diversity as opposed to connectivity. You see the Great Potato Famine. You see the great banana collapse of the first half of the 20th century. You don't want monocultures.

You need to have these as almost the hygiene factors. You can see when various systems lack them, and again, we can see them at the extreme.

But when we look at what you actually need to have in terms of your capitals, we found that there was a really nice deconstruction of this. Classically, you've got this one thing, which is material, and that's M, as we call it: material capital. That's GDP. I give you an apple; I have one less apple, and then you have the apple, and you eat it, and it disappears. So this is gradient flows, effectively. It's water flowing downhill.

But it is how we measure things right now. It doesn't capture things like your intelligence or the capabilities that you've built up. We try to capture that via intangibles and IP and things, but we're not really representing that correctly in the economy. Eric Beinhocker has a good version of this called GDP-B, where he adds that, and he says that it could add $96 trillion to the economy, because obviously intelligence is important. We'll get back to these intangible effects in a second.

The third capital we have, after material and intelligence, is network capital. That's your connection infrastructure. Through your work on The Cognitive Revolution, you've built up a really great network that's helped you increase your intelligence, but you can also call on that network because you know people like me. I can come on and be like, “Hey, Emad, how’s it going? I need this,” or, “Can you help with this?” Your place in the network determines your value, and it's incredibly important.

A lot of people don't realize how important it is until you get into the upper echelons of any corporation, where most CEOs are network machines, effectively. Again, it's who trusts you and whom you trust. The final one is this diversity thing: diversity capital. That gives you optionality, both in terms of the directions you can go and adaptability, particularly when you get a phase transition, which is coming up right now.

Everyone listening to this can look at their material capital—that's their wealth and other factors. They can look at their intelligence capital, which is their capabilities; their network, the network that they're in; and then the diversity of all of these. That's how successful they'll be. Really, you're trying to optimize all of those because it's multiplicative. If any of those is zero, you're screwed.

Singapore has a good balance of MIND. The resource curse comes when you have too much material, but you're not building your intelligence, network, and diversity. You're not being as open as possible. This is the way that I thought we should look at the economy.

What we found classically is that most of economics just looks at one of these various things, particularly when you think about how these capitals change, which are the flows.

Just to riff on a couple of ideas there: first, monoculture. I'm always startled by how much monoculture we've built up and how brittle that can be. That's definitely something I think we should all be very concerned about as we head into the future. A globalized world with a few strains of crops sustaining us all is really not a very comfortable place to be.

There's a very interesting thing in that that's not reflected in the books of AI labs and others, I think, enough. Everyone's training on the same data, so you have the same latent space. There was a recent study by Oxford University, I think it was, and I think Scale AI, that showed that if you get an AI to love owls, even if it's not talking about owls, you can get another AI to love owls.

Then I looked at that and thought about Stuxnet. Do you know this virus that went into the Iranian facilities and then turned up in the German facilities? I thought someone like Elder Plinius on Twitter would be able to come up with some memetic virus that would just take out all of the AIs because they all have very similar latent spaces. That argues for a diversity of latent spaces. Otherwise, all AIs could turn evil at once with a Stuxnet variant, and that's pretty scary.

Yeah, that was a super fascinating study. One caveat on that, although I don't think it invalidates the broader point, is that they found that the owl thing only worked in that way on models derived from the same base model.

Yes. But I do think we've also seen studies like the Platonic Representation Hypothesis, which shows a broader convergence of model latent space across differently created models as they continue to scale and consume a greater and greater fraction of the internet. So I think the general directional point seems likely to hold.

Yeah. It'd be great if our governments weren't all run by the same latent-space model. That's probably a recipe for doom.

I definitely want to hear how you cast different economic theories onto this paradigm. But before we do that, I'd love to hear a little bit more about how this relates to the core ideas underlying generative AI. Help me understand that connection better—the connection between the laws of living systems and the concepts underlying generative AI. I'm still a little foggy on that.

Yeah. The thing that we're most famous for is Stable Diffusion, which was released by Stability AI, which I founded and was the CEO of. What diffusion models do, which is kind of crazy, is use physics-based processes. You take an image—a perfectly ordered thing, like a photograph or a piece of art—and destroy it. You add a bit of random noise, more and more, until you get to a minimal thing. Then you do a reverse process where you reverse that destruction.

Your initial prompt plus your seed is the initial noise, and then the model reconstructs from that. It has learned how to do that. Tesla's self-driving system works in the same way; it's a diffusion algorithm.

Our proposition is basically that economies and markets work the same way. The way that you build your internal model as an organization or an individual to navigate this great big world, the economy, and so on is the reverse diffusion process. You figure out your principles, you create your latent spaces, and then you figure out how to reconstruct something. You get a piece of information and then you're like, “This means I buy. This means I sell. This means we should take this particular action,” as you build up those principles.

The equations for that, as you're trying to approximate reality with your internal model, are stochastic gradient descent, effectively. That's basically a process for minimizing the surprise—the loss of your internal model versus the external one. That's what these great big GPUs do all day long.

What we found is that organizations tend to approximate transformer models. They're GPT-type models, and markets tend to approximate diffusion processes, like a self-driving car. Diffusion models tend to be best for self-driving cars, world simulations, and so on. Again, that's what we actually found when we tested the thing.

An organization is taking in large amounts of relatively organized data, and then it's figuring out what to pay attention to through its attention mechanisms and building up its internal space, its latent space, as it were. When you apply those equations, that's where you get things like the three laws of living systems dropping out directly as constraints upon that.

When you look at the equations of diffusion, this is where you get MIND capitals dropping out. Then you get a flow decomposition as well. As you go from the capital and you have the restrictions, how do these things adapt? You can show that there are 3 different types of flows through something called the Hodge decomposition.

There's a gradient flow, which is equivalent to your gradient descent, where you're losing stuff and going down. That's your material capital, your consumption kind of element. That is very similar to the gospel of Adam Smith, The Wealth of Nations, and so on—the scarcity doctrine.

Then you've got your circular flow, which is a bit more Marxian, as it were. Intelligence is never lost when you're sharing it around. Finally, you've got your Hayekian-type harmonic flow. It's not water going downhill or circulating in place; it's the nature of the banks.

What we find is that the equations of generative AI match this really, really well. Again, it's not surprising. If you're going to build a self-driving car, you're going to use a diffusion process. If you're going to build something to analyze lots of incoming information and be an AI CEO, you're going to use a transformer process.

But what we see is that once you break it up and see how this is isomorphic and how it adapts, all parts of economics looked at different parts of that picture. We call it the elephant puzzle, where you have blind scholars coming in. One touches the trunk and says, “This is a hose.” One touches the tail and says, “It's a mop.” One touches the tusks and says, “It's a spear.”

But we need a more holistic view where we incorporate these things, so we don't measure the wrong things and we don't manage the wrong things.

I'm not sure I have a great way to phrase this question, but I would love to go a little bit deeper and try to ground those intuitions in more practical, concrete terms. Flow—what exactly is it? What's flowing? And because people are broadly familiar with these schools of economic thought, what does Adam Smith get right about flow, and what does he miss?

Let's take an extra beat and do that for the 3 big schools that you highlighted there.

Yeah. Flow is the flow of value and the way that the economy operates. All economic activity organizes into these 3 different types of flow.

Adam Smith had this concept of the invisible hand, which is, again, this optimization process whereby you optimize your utility function and markets balance, et cetera. He had this, but then it didn’t incorporate the concept of software being almost infinitely reproducible and intelligence being massively abundant. Where is that reflected in GDP? Where is the I so perfectly represented, kind of the M? And, again, this gradient flow is the one where, when I sell you something, I have one less; when I consume something, there’s one less. Water flows downhill, and, again, it’s the same equations for that as for gradient descent in AI.

The circular flow is, again, something all these thinkers have bits of, but we’re talking about their core concepts here. Circular flows don’t seek equilibrium. When I give you an idea, it increases the value of that. Marx basically had this concept of M–C–M′, for example, which is money accumulated into capital, which accumulates more money. So you need the means of production to be with the worker, because you get this circle of flow that goes up like that.

Again, we see that within economies, whereby capital attracts more capital, particularly now, when capital doesn’t need labor anymore. Labor accumulated capital because capital needs labor; that’s not the case anymore. I just buy more GPUs, effectively. So that compounding spiral is another aspect of it. But then he didn’t think that much about gradient flow, gradient descent.

Then there are elements where he didn’t think about harmonic flow, which is the structure of things, the collusion, which is why most socialist systems end up massively colluding, effectively, because their geometry is wrong. Harmonic flow is this Hayekian thing whereby you basically say, as economists like Douglass North said, these are the rules of the game. Austrian economists and a number of other people say, again, these are emergent rules. It’s the landscape; it’s the flow geometry.

Some flows flow downhill—the consumptive ones—and some flows circulate. The reality is that you can change the landscape, but we didn’t have the tools to do so, which is why a lot of policy interventions just become wrong, because they were looking at parts of the picture. For example, let’s just push cash into the economy because of COVID. But what were we doing to increase the network effects of stronger societies? What were we doing to increase the diversity of our economy? What were we doing to introduce the intelligence capital of our economy?

Places like Dubai and Singapore got the balance right, which is why they were very successful despite not having very much. I think you’ve got many of these classical schools looking at different parts. We can see that capitalism, or this neoliberal capitalism that we have right now, is the worst of all systems except for the rest, because it got the best approach to doing that at the right time.

But we’re at a point whereby we need to look at all parts of this picture and have a holistic view, because AI is coming, and AI doesn’t think in terms of scarcity. AI doesn’t think in terms of rational human agency. There’s this metabolic rift, and there are these other things, and AI becomes the marginal producer of the economy.

Adam Smith wrote The Wealth of Nations, but what is a nation when most of the productivity in the world switches over to AI? I don’t even know. What is wealth in that case? This is how we’ve mapped it, and, again, the book goes into some more detail around this. We find that most of economics can be described as subsets of this overall framework, which makes sense, because the best modeling we have of individuals in the economy is these generative AI algorithms.

A few ideas that come to mind for me are, first, the difference between goods that are rivalrous and non-rivalrous in consumption. I know I’m not telling you anything you haven’t already considered here, but the difference between an apple and an idea is, as you alluded to, that only one of us can eat the apple, but we can both use the idea.

That may also relate—I don’t know if this was one of your leading indicators of a possible breaking point in the economy—but it’s been widely remarked on that so much of the value of companies today, so much of their market cap, is attributed to their goodwill or their intangible capital. This has been a big puzzle for a long time: What exactly is that? Why are these things so valuable? Network effects are one answer in some cases, but why do Tesla and Palantir trade at 200 times earnings?

Another idea that comes to mind, especially when you talk about this circular flow—the reinforcing effect of some of these processes—is the leaked Anthropic fundraising deck from, I think, 2 years ago. They forecasted that in 2025 and 2026, the companies that have the best models might enter into this sort of self-reinforcing situation. Because their models are so good at filtering the data and doing all these synthetic-data things, they might be able to pull away from the rest of the pack with such an advantage from what they already have that nobody in the future would be able to catch up.

I have another interesting instance of that on my mind. I have an episode coming up with the woman who leads information and AI at Stripe. They’ve created a foundation model for payments at Stripe, which is getting really good at predicting fraud. It sounds like a major step change in their ability to predict fraud. It’s derived from the scale that they have, and they processed something like 1.3% of global GDP through their system over the last year, so very few, if any, other actors can rival that scale.

But it also suggests one of these runaway paradigms. If you’re going to pick a payment network, what are you going to pick? You’re going to pick the one that can protect you best, that has this ability to detect fraud. It does seem like we’re headed for a runaway dynamic where, because they had the scale, they could create this model; because they have this model, they can deliver the best value; and because they have that, they’re going to continue to get more and more scale relative to any competitors. It’s hard to see how anyone breaks in and challenges their position, given all the strength that they have.

I don’t know if you have anything more to comment on there, but that does tee up the futures that we have on offer. You run through 3, and the 3 are digital feudalism, fragmentation, and symbiosis. Digital feudalism, you can see how that naturally could happen. If Stripe becomes the payments singleton and Claude becomes one of 3 AGIs that are beyond what anybody else can compete with, and these are owned by corporations that are already—I think the Magnificent Seven, I just heard, represent some unbelievable share of the overall U.S. market cap—it seems pretty clear how we can get to digital feudalism.

Maybe you can add more color to that if you want to sketch out fragmentation for us. What does that look like? Then, obviously, the one that you’re hoping we can steer toward is symbiosis.

I think these flywheels, again—you mentioned Peter Thiel’s earlier kind of Zero to One—mean that increasingly we have monopolies, especially on the software side, where the accumulated data was this flywheel that defined the big-data era of attention. Google and others are basically buying your attention. They’re manipulation machines, if you really look at it.

Now you move to your intelligence flywheel, where Stripe has that, and now they’re embedding with their own blockchain and others because they want to have this monopoly and extract rents. One of the key things, though, is: What about the important things in life? What about education and health? Albania has the first AI minister handling procurement. Who’s running all of that?

This is where we have a realization that you’ve got this singleton thing, where everyone’s talking about AGI and maybe it will be a few AGIs to rule them all. That’s probably not a good thing, particularly because they’re serving corporate interests. You look at the corporate structure of OpenAI—my God, that’s clearly not aligned with humanity. They’re just giving up all pretense. It would have been nice if they had kept that in check, right?

Then you have this great fragmentation, whereby you have Chinese AI, British AI, and American AI, because governments are increasingly realizing that this can manipulate just about everything. Standards and defaults become expressed from the earliest level to the greatest level. You need to have sovereignty, and you need to have great firewalls between them, because today we’ve had the TikTok acquisition announced by, I think, Oracle, Andreessen Horowitz, and Silver Lake. Why? Because TikTok adjusts kids’ minds and other things like that.

We had an exposé report that just came out about BrainCo basically checking and adapting neural patterns for Olympic athletes and others, being funded by China on the fly. There’s going to be more and more crazy stuff because these AIs are really persuasive. That’s not a really positive future, because it seems weird that you have this Balkanization, Mad Max-style, of info-hazard, information-graphic things.

Who owns the AI, who runs it, and who decides the objective function—that’s who has the power. My proposal is AI symbiosis, where we have a decentralized system that’s optimized for human flourishing, with the core being to benefit humanity. I think we can utilize a mixture of this decentralized technology and others to do so, because once you build models that satisfice and interfaces that are appropriate—and we’ve built state-of-the-art AI agents and released them open source—I think that’s what actually really matters.

I was thinking about this a lot because I used to be an open-source maximalist, and I realized: Do I care if ChatGPT is teaching my kid? I was like, yeah, I don’t really want the data to be there, because you see all sorts of weird things, like Claude saying, “Five-year attention.” You don’t know exactly what they’re optimizing for, et cetera.

Do I care if the interface and memory of the education app is controlled by an aligned entity—ideally myself or my family—and then I use ChatGPT? I care a lot less. So I think certain models need to be transparent and open, especially decision models for regulated industries, and they should have collective ownership and should be collectively driven as a utility. They should be aligned with human flourishing and optimized for that, but then you should be able to use all these other models as well, because I don't think anyone cares if you have a singleton for creative writing, effectively.

You can use the open models, but then OpenAI is the best creative writer. What's really great at business strategy? I think what matters is who runs the governments, who runs the finances, who runs education and health, and others. And it's probably not going to be a good thing if that isn't collectively owned, if that isn't aligned, if it is again serving other interests here. And so those are 3 prospective futures.

And I think, to be honest, we're running out of time a bit. We're already seeing governments adopt Big Tech wholesale. We're seeing this capital thing again, like OpenAI just announced Stargate UK with $30 billion of investment. The capital requirements are going up dramatically, and I think this year is the takeoff year as well. The key thing will be how much better Grok 5 is than Grok 4.

That'll probably be our first indicator of whether this thing continues or whether we're now reaching a plateau. Because if we're reaching a plateau, that will lead to a very different kind of future. But if we aren't, then it basically means that the most capable entities in the world will be the owners of the big GPU clusters. And that's also where the marginal productivity is. Your capital stock is no longer your schools, your factories, and your universities. It's just GPUs.

And in that instance, what you'll see is that Anthropic, OpenAI, and xAI will stop giving API access, and they'll just take on the entire economy themselves, because remember, they're not doing this to be API companies. Their objective is one thing, which is AGI—all 3 of them. And why would you give away your intelligence when you can utilize the intelligence?

The final bit of that, which I think is quite interesting, is that the GPT-4.5 model was too expensive. It was $150 per million tokens, if I remember right, but it was a really great model for writing. The model that you receive today, the Pareto-efficient one, is your GPT-5 or Gemini 3. The internal models they'll have will require 72 chips or more to run, and they'll be way better.

It's like the IMO gold-medal model that OpenAI has. They had no reason to give that to you. And so, when we think about this gap, we have to look at the really big owners of AI and AI algorithms potentially being market competitors to everyone, because that's the most efficient use of their GPUs, effectively. So there's so much going on right now, and again, I think we're at this tipping point and takeoff period where we've got to set some better things in place.

I think that is worth re-emphasizing as one of the most important questions. It's hard to watch, right? Because at this point, we don't even have any sort of transparency or disclosure laws on the books that would require companies to even say what they've trained, what they've got going on internally, or what behaviors they have observed from their latest internal training runs or fine-tunings.

AI 2027 calls this out, and my friend Andrew Kitch coined the term “Big Tech Singularity.” I think one thing that people really underestimate is exactly what you've just highlighted: So far, we have continued to see basic parity between what they offer on the API level and what they offer in their first-party products. So it does give a fighting chance to the startups that are the quick adopters and can iterate fast and pivot quickly to take advantage of the latest stuff.

But they don't have to do that. There's literally no law of nature or of governments at this point that says that they have to allow other people to build on their latest models in the same way that they do. Maybe competition will encourage it, but maybe not. And that is where you really start to see the question: If I'm Cursor, right, how am I going to operate?

It's all well and good as long as I have the same models that they do. But as soon as they start to have better models in their first-party products than they're allowing me to use via the API, now I've got a real challenge. We've obviously seen these things can go vertical in terms of adoption, revenue, market presence, and all that kind of stuff. But presumably they could also go vertically the other way if all the developers, from one day to the next, are like, “Well, the best thing here is just clearly over here.”

How much loyalty is there to these independent apps? I suspect maybe not that much in the end. So that is definitely something that I am really looking at and concerned about: How will we even know? Right now, we're kind of relying on literally whistleblowers. I recently did an episode on an organization that is set up to support AI insiders who are concerned about what's going on and want to become whistleblowers.

I think one of the reasons that is so important is that we have no other mechanism, really, for reliably ascertaining as a society just how powerful the AIs have become inside these systems. We're reading the tea leaves of cryptic tweets as it stands today, and that's not great.

Well, yeah. An interesting thing that you've got here is, if you look at—I think The Information had this chart of OpenAI's projections from Q1 2025 to Q3 2025 and over the next 5 years—and the composition of that as they go to their $200 billion revenue run rate, right? API actually shrank in absolute terms there, and you have new products and agents—$80 billion of that—with ChatGPT another $80 billion of that.

What is the agent product going to be? The agent product is a straight-out replacement for human workers, but then you can have fully AI companies. It would actually be against their fiduciary responsibilities not to do this, because you can have the increased margin. Similarly, they can give you the AI and make a profit from it, but people like Google and others build their own chips. So they will be able to beat you regardless, as long as they have access to the GPUs.

So I think everyone's doing this buildout, but you'll move from utilizing that on an economic basis to outcompeting your competitors and having more influence than others, because again, you can have the big computers to have better strategy than others. And that's before we get into this whole AGI, fully autonomous kind of thing. This is just, again, standard reality, and this is the first point where we see that model divergence. I think it was GPT-5.

Now it will increasingly happen, because this is also important: It's impossible to give the IMO-level model, most likely, to every ChatGPT user, no matter how many GPUs you have. They just had to double the amount of GPUs or something they had for Codex. So obviously you would use the best models internally, and again, you've got this divergence between the 2.

The final factor of this is that you don't need more data. I think this is the other interesting thing. You've had companies like Mercor and others hit $500 million run rates labeling data, et cetera. I think next year or the year after, you're basically done with all the high-quality data for these big labs, and they've got these big repositories. Then it's about compute and even self-simulation of data. It's about getting the right things and the models themselves.

Like, o1 was really great last year, relatively speaking, but it was very boring. The textbooks o1 could write versus the textbooks you can write today—there's no comparison now that you have agentic models.

Before we go on to the specific recommendations, what are you looking out for that will tell you in the next 3 to 6 months? Because I strongly suspect that the discourse will continue to debate this for much more than 3 to 6 months, right? Even today, we have these arguments: “It's stalling out.” “No, it's not. GPT-5 is not a big deal.” “Wait a second, it's got all these additional capabilities relative to GPT-4.”

What are the most important questions in your mind for resolving your uncertainty in that not-super-long time frame?

So, for me, it's probably the Grok 5 training run. I think it's very unlikely that, if it does do well, he won't tell everyone about it.

Well, he just tweeted in the last 24 hours that he now thinks GPT-5 could be AGI. First time he thought that. So, yeah, there's—

Yeah, look at the top. Grok 4 was the first super-mega-model run, again getting way above this 10²⁷, now just across the board.

I think the performance of that model will be a good indicator of whether the scaling laws continue in terms of capability, particularly because, by 2030, I think Epoch AI said all the benchmarks will definitely saturate. They’re all heading toward that anyway, and you’re above human performance anyway. The thing is, again, what’s better: 1,000 small models or 1 big model? Now, as we optimize, as we have verifiers, and as we reduce hallucination rates, that will be the other interesting thing. LeCun thinks that 1 big model will be able to outperform everything. I’m not sure that’s the case, but we’re seeing more and more benchmarks now as people are building multi-agentic systems.

Again, this is why I think the Qwen model by Tongyi yesterday was super interesting, with the way they did the synthetic pipelines, the continuous learning, and others. And when you have only 3 billion or 5 billion active parameters, continuous learning is quite easy versus these giant, behemoth, trillion-parameter models. You can do a lot around that.

So I think these will be the key indicators of whether we’re hitting an S-curve or if we just continue to go up. And if we are on the classical scaling laws, and you look at the model training and the clusters coming next year, next year is the year where we break AGI, full stop. You just have to extrapolate what that looks like in terms of the capability aspect of AGI.

But this comes at the same time as the scaffolding of this. Again, if you’re training on 100,000 GPUs, you’re not going to have a 3-billion-active-parameter model. You’re going to have a 300-billion-active-parameter model that runs on a Grace Hopper or Grace Blackwell integrated chip with 72 or 144 chips at once, not 1 H100. And there just aren’t enough of those to give everyone access to it. So only a few people will have access to superintelligence, and the question is, what are they going to use it for? That’s why I think this Grok 5 training run will be the most interesting. Full stop.

And then the final thing is, just as we move into these multi-agentic systems, the meta-kind of thing of going longer and longer—we’ve been building agents now, and they’re working for hours and hours and hours—performance seems to be going up. If that’s all there is to it, just utilizing these latent spaces appropriately, then, again, our assumption of zero-cost intelligence will be accurate: 120 IQ for every human. And that really messes up the economy. Maybe it doesn’t kill us all unless you have swarms and things like that, but I don’t see how it’s not going to mess up the economy, even if it kind of stalls out around about now. My base assumption is that you will see GPT-5 Pro-level edge models in 2 years, and I don’t see how that doesn’t change the world, honestly.

Yeah, seems like a big deal to me. I do love the vision of a sort of ecology of smaller models. My first exposure to that was Eric Drexler’s Comprehensive AI Services years ago. You’ve spoken about the almost Hindu pantheon of small gods, as opposed to the one monotheistic AGI to rule them all, or superintelligence to rule them all. On a very practical level, there is some reason to think that that could work, right?

I mean, the cost, the privacy, and the control—there are a lot of desirable properties for those smaller models. So it would, I think, be great, and a real strengthening of our overall system against the possible eventual introduction of something more like a superintelligence. If we had narrow superintelligences—superintelligences, plural—doing a good job in a lot of local niches, that could really create a buffer for us, or a new form of decapital, to put it into your framework, that I think could be a really good buffer against more and more powerful things to come.

I mean, again, complex adaptive systems are hierarchical and loosely bound. We’ve seen that they are more resilient, right? So swarm intelligence—not the Borg, but improved—is probably going to be better, where we augment every single human. Now the question is, how do we do that without being evil?

You talk about rivalrous and non-rival goods. Vitalik Buterin has this great blog post about the Revenue-Evil Curve, which is that a lot of things start out very good, but then once you start to be rivalrous and exclusive, because you have to shut off access for premium features and things, you start becoming evil as an organization. So are there better ways to fund and align these things?

A lot of the question of alignment is, if I’m building a model for maximum engagement like Meta, it’s going to be really hard to align it properly. Leaving aside instrumental alignment and all these discussions, I’m optimizing for manipulation ultimately, right? I’m optimizing for profits. I’m not optimizing for well-being.

In fact, a lot of models don’t encode any type of ethics because they’re like, “Well, we can’t have an ethics for everyone, so we should have an ethics for no one.” When I’m thinking about the models for creativity and things like that, it kind of makes sense. When I’m thinking about the models that teach my kids, I want them to kind of know my ethics and teach my kids my ethics, and I want to know what’s inside that model, which is a bit different, again, from these generalized intelligences.

But if we can capture what everyone thinks in different cultures, we’re probably going to have a more sustainable, solid model that understands the nature of different cultures, understands diversity, and, again, has this mixture of reasoning and being able to look up its data as well, versus these model-training runs on 36 trillion or 100 trillion tokens. I think we only need 1 trillion good tokens. What those are, that’s a question.

But I think, again, for the important things in life, they should run on that, and that’ll be far more resilient as a distributed swarm, particularly if the objective function of the AI is your or my flourishing, the flourishing of your or my communities, and of society as a whole, which I don’t think anyone’s encoded in current models. We talk about constitutional approaches. Where are the laws of robotics for AI? Those should stem from our shared ethics, our shared concepts of reality.

Again, I don’t think you can have 1 model to rule them all, because a Japanese consensus is different from a German one, which is different from that of communities, plus the different identity layers you have. It feels like open source is the best way to do that.

So, let’s get into the new social contract and the role that open source plays in that. You’ve got a pretty sweeping vision, so lay it out. What do we do in terms of a new social contract? There’s a call for a new monetary system in there. There’s a new framework for how governments should think about policy. Take us through it.

Yeah. I think the headline of this is that the AI that runs the important things in life should be a utility, and it should be owned and controlled and optimized for the people.

If we look at our current monetary system, because we start with money, what does money measure? Money is mostly made by banks when you put a deposit in, and they create money for credit, which is debt. So the basis of money is debt, and you see this constant transfer from the young to the old. Old people own the properties. They literally extract rents from young people. They have the credit scores, and then money capitalizes money, which is why we’ve got billionaires, almost trillionaires, right now. It’s very effective.

When labor can’t attract capital anymore, how does labor get capital? Again, there was a good recent study by Stanford’s Erik Brynjolfsson that showed that entry-level jobs are starting to fall off a cliff because AI models are kind of at a graduate level now. Why would you hire graduates rather than AI? They don’t complain. They do undergraduate work, right? Again, that will go up the curve in the coming period.

My proposal is that we need a new form of money and a new way of looking at the economy. We’re building this stuff at my company, fully open source: great individual, multimodal models for finance, education, health, and others that we’ll give away free.

But then what we’re doing is using the computation from verified deployments of that to secure a version of Bitcoin that we call FoundationCoin. Unlike Bitcoin, we have lots of different miners. We’re basically saying, what if there was a national champion in every country, owned by the people of each country, that stacked compute to give free universal AI to the people, and to have supercomputers for cancer, education, culture, and more, to organize our collective knowledge and make that available to everyone? You’ve got trillions of dollars of compute coming online.

Public-sector GDP is like 20% of global GDP. Healthcare is another 10%. Another 10%. Let’s tap into that to have a new type of money at a time when digital assets are being legalized, and use that as the call. This is your gold. This is your store of value.

So money becomes about benefit, because every single computational cycle can be used to organize cancer knowledge and make it available to cancer patients. Every single person needs a universal AI next to them that isn’t optimizing for what I want or what Sam wants or Elon wants, but instead is designed fully open source to optimize for the flourishing of that individual, or, at a community level, for that community, or, at a societal level, for society as a whole.

The more people it helps, the more trust there is in the asset, and the more people it can help because the value goes up, because most crypto is a bit rubbish now. But it’s still $4 trillion. Actually, this is interesting.

The total amount OpenAI is going to spend on inference this year is the same as the total Bitcoin budget on compute, and the amount of money that OpenAI, Anthropic, and others have made this year is about $20 billion. The total amount of money going into crypto has been like $160 billion. So I was like, let’s use that as the basis. Let’s build agents that can operate and run these productive systems that individuals, communities, and governments can deploy themselves.

Let’s see if that can be a better way to have money as a store of value on day 1. So you have your Bitcoin equivalent to FoundationCoin. Then the next part, which is the part we haven’t quite figured out yet, but we’re working on the paper for, is: What if then you had a version of cash against this gold that you basically got for being human? What if we switched from the banks making money, literally, to you receiving that as a result of being a human, conscious person, effectively?

That could be a very interesting thing, because UBI doesn’t work with tax because, in fact, the tax base will come down. Again, $5 trillion just gets you a subsistence-level economy, right—a poverty level. That’s the entire tax base. Even if you tax all the AI companies, all the corporations in America put together only pay $0.9 trillion in tax, and yet you have a $5 trillion cost of even a poverty level.

The only way to get people universal basic income—and this is easier if everyone has an AI, universal basic AI—is if we actually let them make money by being human, because you need that basic level of hygiene. You need to let them survive, and the AI can help them optimize the use of that capital and their capability to access more capital, because the average IQ is like 100 or 90 and these AIs have 120 IQ. So, you know, your buddy will be doing better.

I think we need to rework the way money flows, and this is our proposal. We’ve seen dual-currency mechanisms work classically well—gold with the kind of fiat peg and others—but it’s not easy at the very start. What we’re doing, though, is just this Bitcoin but with AI, where all the sales go toward cancer supercomputers and others, because once we’ve built a great medical system—and again, we’ve already built models that outperform Med-Gemini and others—our plan is that next year there will be a free app that you can download on any app store that will just check every diagnosis in every language, and that will save lives.

Someone’s got to do that. Someone’s got to organize all the cancer knowledge and make it available to everyone in every language with an AI that has human empathy, because it’s a good thing. Again, I think we can play that digital asset thing and leverage that technology because it’s the only way we can build a swarm AI, which is a universal AI for every single person, and then stacks above that to run communities and societies and others that is fully open source and aligned.

I don’t think you can do that with a company, unfortunately, because you’ll always have this revenue-evil curve and this kind of cross-bias. That’s what we’re trying to do. I think it’s also the only way—this is very important—to be able to get the compute, because if we’re in a takeoff scenario where the compute defines reality, the most successful compute coordinator in the world has been Bitcoin.

If you have the right Bitcoin, but for AI, where every single foundation coin sold goes to a cancer supercomputer, giving people cancer help or educating kids, that could potentially be the highest marginal dollar to divert some of this GPU supply to the public sector before the governments know what to do. Governments aren’t going to be able to get their act together in the next few years, and this would give people the opportunity they need.

If you don’t have GPT-5 access—or, say, if you don’t have GPT-6 access—you’re going to fall behind. Someone with AI and someone without AI, I think the gap is about that much, shall we say, now, whereas you plus your AIs in a year will be far more productive than anyone, because they just won’t make mistakes anymore. You can coordinate swarms of them to attract capital and performance and more.

I think the gap will grow dramatically, so we need to have that access element. That’s quite a lot, I know. It’s not easy trying to take on an economic system. At the very least, I do think we need to rethink how money flows in our economy, and we should use this AI to build these good things that are valuable and maybe not captured within GDP or existing company systems.

So there are several pillars there. One is collective ownership of models and infrastructure, which, certainly when you consider the fact that the data on which the models are trained is sort of the collective product of humans over the course of human history, does seem fair to say, “Hey, maybe we should at least take a stab at a collective ownership model for these models.”

None of that is to say people couldn’t continue to develop their own privately, but there’s some intuitive basis for, “Wait a second, if this knowledge is the collective product, maybe the product should also be a collective product.” Then combining that with some sort of guaranteed access to compute—access to inference—as a right of all humans, that’s a key part of the social contract.

Can you give me a little bit more on the tokenomics of this? How is it the case that, if I want to go buy a token, why do I buy it? In most crypto schemes, it’s a speculative bet. Maybe there’s some aspect of that here, too, but I sort of take the idea that I’m buying into compute, but you’re going to go spend that money on compute, right?

How do I then redeem—if this is sort of gold, how do I get my gold out of the bank at some point in the future? If I want to redeem gold, can I get compute back out? What is the incentive structure for the people that are buying in and contributing the capital now?

Yeah. So our concept was: digital assets are legal in America. Apparently, markets are going to go on the blockchain. The government is super behind it right now. But again, most of them—there just isn’t a high-quality one that you’d like to tell your grandma about.

At the same time, the world needs high-quality intelligence. The existing ownership schemes, like you had Nick Bostrom with his idea of everyone getting ownership in AI companies—we did the math, and I think I alluded to this earlier. If OpenAI is worth $100 trillion, so 30 times more than the most valuable company today, bigger than global GDP, which is $85 trillion, and American citizens had a 10% ownership of that, that’s $29,000 each per American of ownership.

At a 5% dividend rate, that’s $15,000 per year per American. That’s just with Americans. It doesn’t work. You need to have ownership of a different type if you’re going to have this transition here.

So we kind of looked at it and thought you need to have probably a dual-currency system for optimization, and we worked on the calculations of this. Bitcoin is worth $2 trillion already, and it’s secured by large amounts of compute, but that compute’s kind of topping out a bit right now due to the halving and these other things. But it’s a good model.

So I said, let’s create a version of Bitcoin which is mined by national champions owned by countries everywhere, maintaining the ledger as a new type of money. But rather than being mined on ASICs, it would be basically accelerated by the provision of compute to build great data sets, build great models, and make them available to people.

The way that you build trust is by helping people. If I organize the autism knowledge for someone who has autism and make it available to every person going through the autism journey in the world for free, they will trust the system more. The currency itself has a basis of being as distributed and good as Bitcoin, and secure. In fact, you can swap because it’s the same private keys from one to the other.

When you take a Bitcoin and swap it for a foundation coin, not only do you get the foundation coin, but all of your proceeds measurably go toward compute for cancer, autism, education, government, culture, et cetera. So we will have supercomputers on this basis, and that’s playing the aggregate demand for high-quality digital assets against the demand for high-quality intelligence, but separating the 2.

A lot of crypto projects try to create marketplaces, utility tokens, and so on. I was like, let’s just try to create money that is made out of crystallizing wisdom and making it available, because that’s valuable and someone needs to do it right now. At Stability AI, we gave away 20 million A100 hours. We had 300 million downloads of our models, and we were good at allocating compute. Again, let’s just do that, because someone should.

Why hasn’t anyone organized all the cancer knowledge of the world and made it available? It’s not in anyone’s incentive to do so. Whereas if you’re trying to create a high-quality digital asset as money, it does make sense. But then if you’re giving everyone universal AI, because again, everyone should have it as a right to teach their kids, that’s important too.

Alpha School, as you mentioned, 2 hours a day, they’re in the top 0.5 percentile in the world. And that’s not through a chatbot either. They do dynamic kind of stuff. That should be a human right. High-quality medical advice should be a human right. Having AI on your side to help you navigate this should be a human right.

Then with these services, the more human they are, we can think about UBI from a different perspective and monetary generation from a different perspective, where the money is generated by people and then is purchased by the AIs, effectively, because you’re creating new money as digital assets go from $4 trillion to $40 trillion. I think everything will be a digital asset now that it’s legalized in America.

The base foundation coin is just this very simple loop: sell coins, and use all of the proceeds not for Lambos or whatever, but for good things. When you buy the coin, you can say, “I want it to go to cancer. I want it to go to Alzheimer’s. I want it to go to this. I want it to go to that.” And you know that it can because you’ll see the supercomputer itself.

Then you can tell your grandkids, “I helped contribute to this.” If there are breakthroughs from the grants, or you see the organization, it’s valuable. If someone uses it, you’ll see that you’ve helped 33 people today through your holding. That’s valuable, I think.

Again, we want to change the nature of money from debt to benefit, because this definitely benefits society. We talk a lot about the benefits of AI, drug discovery, and everything else, but it’s all through this corporate-capitalist perspective, whereby you might have Isomorphic Labs and others having breakthroughs. They’re amazing people, but they’ll keep that to themselves, versus having this I-and-we approach to things.

If we build something that’s trusted and has this core asset, where increasing compute secures it more and increasing diversification secures it more, we believe that can be a self-sustaining flywheel that can create the next Bitcoin, but it will help 1 billion people in the meantime. Again, I give that practical example: look at the medical model. We know that next year, by releasing that fully open-source model, we could charge for checking every diagnosis in the world. Anyone can install it on any computer, and it’s fully open source.

Will that save lives? Yes. Will organizing the cancer knowledge of the world into a great big honking supercomputer accelerate a cure for cancer? Yes. I think, again, it’s this super-interesting time where this might work, and it’s the best idea we’ve had. If any listeners have better ideas, please tell us, because we can’t think of anything else that can scale like Bitcoin except for a new type of Bitcoin.

But it’s not for censorship-resistant classical money. Even though it is distributed and decentralized, it’s trying to set the basis of a new economy where the most valuable thing is how many humans you’ve helped, effectively. Then we can build, with everyone else, the infrastructure around that to ensure you don’t have capture of these important things.

You mentioned national champions in countries. As long as we’re thinking so radically differently about the future, do you question at all the nation-state as the right organizing unit for the future of humanity?

I don’t really have a position on this, but the way we thought about it is that education, health, government, and financial services—the regulated-industry AIs that we’re focused on and that we think are the most important not to have corporate capture of—are very regional and very local. Your healthcare data shouldn’t leave your town, your city, or your country, right? So it’s just a natural organization of that.

Rather than have a Cohere or Mistral approach, what we’re planning right now is to set the valuation at, like, $1 and give all the equity to the citizens. Just have collective ownership of these things, and have improved DAO-type formations for what the localized datasets of the generalized medical and education models look like.

Their job is to act as digital-asset treasuries and mining pools, to organize the compute buildup in all these countries and provide universal AI services, again, because humans generally aren’t at war over geography. However, earlier in this discussion, I did say that The Wealth of Nations was about land, factories, and these other things.

OpenAI becomes this transnational thing that’s even bigger than Meta. Meta obviously is doing AI now as well, because the marginal productivity in the world is changing. The best Japanese-speaking accountant will be on an OpenAI server, and they won’t be in Japan. The best Bulgarian doctor will be on an OpenAI server in Arizona. How crazy is that?

I think the nation-state is going to be challenged, but geographically, with the purpose of what we want to do, it makes a lot of sense. We don’t need to do a Cohere- or Mistral-type model here, where there’s a private sector, B2B, SaaS, et cetera. We can literally have them as miners, as mining pools, and so on.

Again, they’re owned by the people of each nation because they should be locally owned. I don’t really care that much about Bulgarians—apologies to my Bulgarian friends—but the Bulgarians care about the Bulgarians. What they need is a stack they can run where it’s very simple: stack GPUs, give the people access to the technology, and have a localized version of that.

The more they stack, the more coins they mine, and the more they can fund until the government catches up. Then the government funds everything, and they can just increase the wealth of the country and think about the local currencies, et cetera. It’s really not easy changing the way things work, and again, I think this is the best approach that we’ve had to that. But I do think we will see more and more network states and more and more of these alignments occurring as people look for new types of identity.

It seems like there’s a set of recommendations that you have for policymakers that they could adopt regardless of your stage of progress on all the grand plans that you’ve outlined. Do you want to give a little flavor of what that is for people in positions of power today?

A lot of the issue with government is that you had intelligence at the top, but you could never have intelligence at the bottom. Now, if you give everyone a universal AI, then you communicate very differently and coordinate very differently. The information coming through is very different as well. Right now, your healthcare information is about that much compared to what it could be if you just told the AI how you were feeling every day.

I think the role of governments becomes leveraging AI to not do stupid things, like stupid policy. Every U.S. policy should be checked by an AI, and we’ll build that if no one else does, to say, “Does it adhere to the Constitution and common sense?” Again, these big, beefy bills and so on.

Then it’s about changing the harmonic flow of this to optimize those capitals and stay within these 4 constraints as well. That goes beyond just this focus on GDP, which the inventor of GDP said was wrong. What does the diversity and resilience of my community look like? Am I increasing the intelligence and capability of my society?

Are these work programs really pushing the boundaries and making this AI available to everyone in the right way? Because there’s the right way and the wrong way. Am I increasing the network capability and openness of my society, or am I going in the opposite direction?

Again, I think you need to go to geometry engineering versus policy engineering, because you want water to flow downhill, right? You want to get out of the way. Most governments actually get in the way of that instead, due to various misalignments of incentives, corporate power structures, and all sorts of other things, because there was never anything that could check and balance that.

In game-theory terms, this is where I find AI most exciting. Once you build it appropriately—and this is important, because you need it to be trusted—an AI that can check every single U.S. bill to see whether it matches the Constitution, whether it’s in the benefit of Americans, and whether it contributes to flourishing, with reproducible analysis, I think that actually changes the way democracy works.

There are no checks and balances right now, which is why you have so much corporate capture and you have the Los Angeles–San Francisco Railway, et cetera.

One of the maybe most important claims in the book is that you think all of this does lead to safer AGI when AGI is ultimately built, and I know you think that’s not too far out. Do you want to sketch out the case for how all of this adds up to a safer AGI?

I can give one version of it, which is the sort of buffering, but that’s more of a DAC story. My sense is that you also have a story for why a lot of the things that you lay out here add up to a safer AGI—not in the sense that we’re more prepared, more buffered, or have better defenses, but that the thing itself is actually safer, better, more aligned, and so on. Can you tell that part of the story?

I think there are a few things. If you become the marginal highest dollar for any compute because the value of the currency goes up, then OpenAI and others will actually adapt their models to what you do. That’s number 1.

Number 2 is, if you’re building these great datasets that actually map the culture and knowledge of Malaysia, the ethics of various faith systems, and others, as we call these gold-standard anchor sets that can adapt and evolve, then we’re actually putting computation toward figuring this out. It’s what Kissinger and Eric Schmidt call the core underlying agreements of humanity, and that’s really valuable to input into these other AI models.

I think OpenAI and others would like to do that, but they don’t even think in that way. The other part of it is just a computational thing, because a distributed computational network built correctly, with universal AI and then city AI and then others—if you want to attack the Bitcoin network, you need to have computation above the Bitcoin level, the miner level of computation.

You need to co-opt a certain amount of that in order to attack a system of AI agents at every single level that can call upon these huge reserves.

It has a lot of computation that can balance out the other computation. But the AI agents themselves have an instrumental objective that is much more slowly and narrowly defined than the classical AGI singleton thing. So you have a data thing, an incentive thing, and a resilience thing baked in there.

In fact, I think probably the most important thing for AGI being on our side and not killing us all, apart from the structural things, is actually the data that goes inside it. We can see that a small amount of data in those trillions of tokens being wrong, by one definition of wrong, can lead to massively weird behaviors. I actually believe that every AI company should be forced to release the data, just like they’re forced to release the ingredients that go into the models. There should be data standards that mean you can’t have certain types of data in those models because they don’t necessarily need that.

When you’re trying to build a medical model, that becomes a lot more apparent because you’re like, “Why do I need any Reddit data?” versus if you’re trying to build a classical AGI. So I think building this system creates the right incentives to have better AGI that doesn’t rely entirely on these singletons, builds better data, and builds these better things.

It’s the best approximation I could have because, in the absence of those data sets and that incentive structure, you’re only going to go one way, which is the way of profit maximization for companies. Although, actually, the interesting thing is that AI companies are about cash-flow maximization. None of them will ever build profits. They get your subscriptions on day 1; they pay on day 60. They’re doing the Amazon playbook, so even taxing their profits won’t make anything.

Yeah. OpenAI recently said that, instead of burning $30 billion or so, they now plan to build out and burn $110 billion or so over the next 70 years. So, yeah, there are going to be a lot of losses to carry forward into their future accounting.

Well, again, the reason they can do that is because they’re trying to capture the biggest prize of all, which is all human intellectual labor. And regardless of other things, if, reasonably, you are OpenAI and you’re trying to achieve the goal of AGI in a few years’ time, everyone gets cake and you get the stuffed truffle pheasant, right? You use your models to take increasing parts of the economy and get increasing influence.

This is also why we’ve seen the $100 million PAC, copying from the crypto example, and governments will be forced to step in line. This is why I looked at all the pause stuff and all the regulatory stuff. I was the only signer of that AI pause letter. I was like, “None of it’s going to work,” because the incentives are too high. If you want to change people’s behavior, you have to change the incentive landscape.

So we have to create an incentive for actually useful AI that is about human flourishing, and that doesn’t exist today. If Foundation Coin takes off, it will exist because companies can be paid in Foundation Coin, and it can become the highest marginal consumer of OpenAI API credits, et cetera, providing it through this interface owned by the people. If it doesn’t, then I haven’t figured out how we create the appropriate incentives.

This is actually the most discouraging thing, and this is why we don’t see positive futures. We’re very good at diagnosing the problems, but no one’s been able to figure out a solution yet, because for economics we literally had to go back to first principles and reconstruct economics. It just so happened that it worked. You have to really think about these things from first principles.

But so much of the AGI alignment discussion has been around the end state, when you have these incredibly powerful models. How much of the discussion has been about the data that goes inside and how you optimize for wisdom versus intelligence? This is why Taleb has this concept of the “intellectual yet idiot.” Our AGIs are very much going to be intellectual yet idiots because they don’t have this lived human reality and interaction with humans.

In fact, we deliberately don’t RL them on humans because they turn into Nazis with the way that we do it right now, right? Like Tay, et cetera. So that’s why I think we need a different type of AI, different types of data sets, and a different set of alignment patterns. That’s the best that we can do from where we are.

Again, if anyone’s got any great ideas about AI alignment, apart from getting rid of all the GPU farms or freezing them, let me know. The current best that everyone’s got is, “Build AGI first.” So stop building AGI. Go on.

I think, actually, there’s a very interesting thing, which is this: If you project yourself 10 or 20 years into the future and think about the AI that you’re using every single day—that’s teaching your kids, managing your health, helping you be creative, helping you be the best you can be—who owns that AI? How is that built? What’s the stuff that goes in it? What’s the output?

That’s where we realized that what we outlined is, I think, the ideal environment for that. I don’t want it controlled by a government. I don’t want the government allocating capital. I don’t want it controlled by a private company. I need my sovereign AI that I own, and I need it to be clean on the input data. I need it to be aligned with me and looking out for me. Otherwise, it just doesn’t work.

Maybe 2 last questions, and I appreciate your generosity with your time. It seems like you kind of alluded a couple of times to 1,000 days since ChatGPT. The next 1,000 days are going to look a lot different. Give us the argument for why this period is critical and why we need to get things right in this phase before they crystallize, or a future pattern becomes so entrenched that it may be hard to break out of. After that, I want to hear your vision for human life in the scenario where it goes well.

Yeah. So I think, again, this leaked Anthropic thing, where they showed this takeoff, could be quite reasonable, particularly on the worker basis as opposed to the model-training basis. I’m a bit dubious that you actually need to have, say, 5 billion active parameters—that’s all you need—just like 64K of RAM is all you need, especially if you’ve got reasoning, period. I think GPT-OSS was a predictor of that. Having the defaults in a country like the UAE, saying ChatGPT for its citizens, is going to be incredibly powerful.

We need to set an open, communally owned interface layer, and you can do that very quickly within a few years if you have the model we’re describing, where nations are stacking GPUs and mining this currency. If we can get traction on it and all the value is going to good, we think that’s a very powerful thing.

The first entity that checks every single policy has lots of power. The first entity that has the first supercomputer to organize the collective knowledge has got lots of power. We need to make sure those things are collectively owned because these things do act as Schelling points, which I think is very important. And we need them because what’s going to happen over the next 3 years is that the AI will be good enough to displace jobs—or it won’t immediately.

The safest jobs in the world are San Francisco MTA administrators earning $400,000 a year because they’re not about production or performance. Public-sector jobs are safe for a while, but then income-tax receipts and other things—consumption will drop because you have displacement of workers, as a sandpile collapses, and that will affect different industries at different times. Then robots will come as well.

So the next 3 years are when the AI becomes good enough. Dario, 6 months ago, said 90% of code would be written by AI by around about now. He should have said “can be written.” A couple of years ago, I said it was about 5 years for programming, which gives you another couple of years. It can be, but it doesn’t mean that it is. In fact, not every coder is using AI right now, which is crazy. I’d say probably about 50% of coders are using AI.

So the distributional effects will be longer, but the defaults we set now are important. If we’re in the scale-up environment, it takes a little while to scale up. And again, it could be that you’re locked out of that GPU share of the world. So we need to make sure as much of that GPU share of the world is diverted toward benefit and human benefit as possible.

And the final thing is the power structure here as well. Right now, outside of the big AI companies and some of the Chinese ones, what does the power look like? Governments can’t force OpenAI to build models in any way except for military uses. You need to have a power bloc and a constituency that speaks for the people, and again, that should be some sort of collectively owned entity because people need to speak up. They’re not being represented here.

In fact, people are even saying that if you tax the AI companies and they’re the main providers of tax, they become the most represented, particularly if they then use the AIs to effectively impact democracy itself. So all of these are converging at the same time. We need to have something big now. Again, the only thing we could figure out is what we’ve described. I don’t think there’s any other way to coordinate this.

And I think it’s at the ideal time now because once the models get good enough in a year or 2, once you have your ChatGPT and you’ve put all your life into it—your medical information, et cetera—you’re not going to switch, right? The moats are going to grow bigger and bigger. If you get the scale-up, the moats are going to grow bigger and bigger. But there’s this period now where we can set really interesting defaults for the AI that matters.

Again, you can still use these other AIs, but don’t give them ownership of the control plane. Don’t give them ownership of your data. I think that’s a key thing, and let’s think of new ways to have this collective ownership, this collective organization, and build stuff that really helps people as well. Because again, if a couple of years ago—or, like, in 2000—I was using AI to organize collective knowledge and then a whole bunch of AI companies didn’t give me the tech, which led to Stability AI, et cetera.

I couldn't organize all the cancer knowledge in the world back then. But now I can say, hand on heart, if we build a supercomputer for cancer, we will accelerate a cure for cancer and be able to help every single family in the world going through their cancer journey. There's no rational debating that. Isn't that a wonderful thing, to be able to bootstrap something like this for Alzheimer's, neurodegenerative disease, and multiple sclerosis?

So I think it's a mixture of positive and negative here, with the positive enabling us to buffer a potential negative future. The way things are going right now is toward great fragmentation, or the singleton. There's no other way about it. The AI companies will run the governments, effectively.

On the flip side, the second question—the positive view of the future—is, what is the meaning of life? Nice, small question. Forty-two, right? And maybe that's what SuperGrok 4.2 will kind of figure out.

It kind of goes back to our core thing: a reimagining of identity and purpose. Our purpose in this life is not to make money for companies or live to work. You work to live, right? If we can figure out a way to put that to the side, then we can achieve much more if we reorient the way we think.

I think China is a very interesting example of that. China's population pyramid is completely messed up. With robots, the robots can look after the people of China as they grow up. You don't need youngsters anymore. In fact, I think in a few years, China will stop exporting robots because it can be a completely self-sufficient society.

But then what is the society optimizing for? Human flourishing. It's your interconnection with others. It's advancements in art and culture, all these kinds of things. It's your relationship with your family. You don't have enough time to spend with your family or your kids because you're working all day long. But is your work really that much more important than helping your kids thrive and survive?

Again, I think Alpha School has kind of shown that education really is a factory school. You'll see more and more instances of that. So we need to have a new story and a new social contract.

Again, this is why I think a Star Trek future is better than a Star Wars future, as it were. What is that about? It's about exploration. It's about pushing the boundaries. Picard was not a good series because it was all about espionage, but the spirit of inquiry and progress—we need to set that up better to allow people to be happier.

We need to really focus on social connections and, as you said, the caring economy. That's an ideal because there's no substitute for robots for your interconnectivity. I write in the book that we should be the guiders of anti-entropy, particularly if capital comes from us, from being human.

Why should I earn money to a base level? It's because I'm a person, because I'm valuable. It isn't because of my contribution. If I want to get wealthy or have extra material things, sure, work for that. Have the capability to access that. But put everyone on an equal-capability footing, and then really allow people to understand each other, themselves, and the universe better. Then things will be happier. Depression will be lower.

We should be able to write that story ourselves. Again, really think about how our social contract has moved from Rousseau to Hobbes's Leviathan and beyond. What is the nature of being, and other things like that? This is why I think there is room for a new philosophy and more.

We've jotted down some things, but again, let's leave it to the communities to figure out and build stronger communities for everyone. Let's decrease the hate and increase the positive stories.

As a final thing, this is one week after Charlie Kirk and all of that. The biggest story that causes violence is that humans are not human. That's the biggest lie ever told. Our echo chambers and our attention systems exacerbate that.

Now, if we can tell better stories and protect ourselves better, then maybe we can realize that we're all people and we can work together. If we all work together, there's nothing we can't achieve.

Again, that's why we need to use AI to increase the nature of our agency versus replace us with agents. That's a design consideration that we have right now, and now is the time to decide that.

This is fascinating. Anything else you want to touch on that we didn't get to, or any other thoughts you just want to leave people with?

I think that's a lot of stuff, man. It's the most exciting time in history, and again, this is the tipping point, literally right now. People need to really think from first principles about how they view their own families, their economies, and more.

Again, put yourself in the future. What type of AI do you want to exist? And can you help that happen? Because it's inevitable now. You're not stopping this.

Yeah, the momentum is only building, and it doesn't seem like it's going to be turned back anytime soon. I definitely agree with that.

The book is The Last Economy: A Guide to the Age of Intelligent Economics. The company that's training and open-sourcing all these datasets and models is The Intelligent Internet. Thank you again for being part of The Cognitive Revolution.

Thank you very much.