Emad Mostaque | AI Labs Are Playing Russian Roulette With Humanity

Labor and capital are about to divorce because capital doesn't need your labor. It'll just hire GPUs. So you need to own the means of production. I know in two years' time, if your job is on the other side of a screen, you will be the dumbest person on the team that drags down everyone else because all your colleagues will be AIs. And so the value of your cognitive labor is negative. They would pay you not to be a part of the team. As of this week, we are no longer the smartest things on this planet.

Thank you. My pleasure.

Yeah, I mean, I think Anthropic were set up to get to AGI first to stop anyone else from getting to AGI. That was their whole thing. They were like, 'OpenAI is not safe enough, so we'll do it safer.' Um, they still have their AI breaking out and things like that, but with all the talk about IPO, like when you're a public company, you have a different set of fiduciary responsibilities to your shareholders in terms of profitability and other things. Like for example, we just had that Anthropic researcher resign with 100 million, 150 million views, Coxen, and then the head of alignment at Anthropic is like, yeah, you know, we have a 10% probability internally at Anthropic that everyone's going to die. Like you can't say that as a public company. Like there's all sorts of things and then lawsuits, etc. And really, you look at it and you're like, this is an alignment question. Public companies typically aren't aligned with humanity.

Like it's you versus other people effectively, right? And so with that, I think that it makes more sense for them to do what OpenAI did, which is to raise another $120 billion or whatever. Um, having said that, you know, their valuation might not be doing so well recently given what's happened this week. Like, uh, you know, you used to be able to say, hey, you know, I got shares in Anthropic. Look at me. Now it's like, yeah, I might have shares there, you know, in that company that says they might kill us all. But you know, I think the mood has changed quite dramatically.

So I know a lot of people in the labs and they believe it, you know, like internally actually the probabilities are probably like 30%. And one of the reasons they build it is like, um, there's a concept from Eliezer Yudkowsky who wrote 'If Anyone Builds This, We Will Die' or something like that, name of his book, called a pivotal act, which is the first person to build an artificial superintelligence, it stops all other artificial superintelligences from coming into being. So obviously because you're going to build it, right, you should be the first to build it. This is a race that's going on. But in the words of OpenAI, if you look at their documents, they say this technology will end democracy.

Like that was actually in their view of the future document. You know, you look at Anthropic's latest economic release from a couple of days ago. They're predicting 20% cognitive job losses by 2030. And so, yeah, Congress should really pay a lot of attention not only to the extreme, you know, this might kill us all, an existential threat to humanity. And I've got a pretty high probability to that myself, but also to this will end democracy and it will take everyone's jobs. You know, like that's a pretty big deal.

Well, you know, it's an interesting one. Um, like I had some of my friends and family ask me about the whole P-Doom thing. The probability of doom is at 10% or more. I was like, look, the probability of your personal doom is 100%. You're going to die, you know, eventually. With this technology, you might not die, you know, and it could cure cancer and other things like that. And humanity will live longer. So, you've got kind of this P-bloom or P-fab versus the P-doom. And the question is, how do you navigate that? Because as of this week, we are no longer the smartest things on this planet, which is a bit crazy to think about, you know.

Yes. I mean, like before this week, you could say, 'Ah, it's a stochastic parrot. It can only learn from its distribution.' And then you have literally a millennium prize problem. There's only seven of them. Humans have figured out one. And a swarm of 10,000 AIs over 88 hours solves it. Yeah. I mean, you can't say that's in the training data, right? Like, if it can solve Navier-Stokes, then it can do your taxes. You know, it can solve your SEO strategy. Like, do you really think there's anything you can do cognitively that this AI can't do? Like, 88 hours with 10,000 agents is 100 years of top-level mathematician work.

Yeah, it's crazy. Obviously the cost of that was like what, $15 million or so, and the prize was a million dollars. Um, with so Ramp issued a — well, Ramp came out with a report, Ramp, the sort of like AI data company, came out with a report recently, a few days ago, saying look, top end, like the top 1% of AI spenders have reduced their spending. Now maybe it's seasonality, you know, but maybe it's not, right? Maybe the top — in the report it says maybe the top 1% have sort of like reached their, you know, max spend, if you will. Um, but in that same line, they're using twice as — like 70% more tokens than they were in the previous month. So you have got this collapsing token cost. So the price of the actual underlying intelligence is, you've said many times, trending to zero. Um, but these Anthropic, OpenAI, they have to make money out of these tokens, right? How long can it go on for these guys not making money? Maybe last quarter Anthropic made money. You know, we yet to get this S-1. I guess the question is that how important is this S-1 when it comes out for Anthropic's IPO and how likely do you actually think they are to be profitable in one year, two years, three years?

Yeah, their gross margins are profitable right now. Like it's only the cost of training and sales and marketing and things that keeps them down. Like they have ramped revenue faster than any company in the history of humanity. I think they're about 70, 80 billion dollars in revenue right now. That's like a quarter of a Google. Can you even imagine that, right? Like it's like nothing we've ever seen. They'll hit 100 billion probably this year. That's even with the token cost dropping. So as a practical example of this, uh, Noam Brown who leads the RL team at OpenAI — well, he's one of the leads on one of the teams — he said, uh, in response to, yeah, sure it costs millions, but you have to remember this o3, which was their first proper reasoning model, like o1 was kind of there, um, in April of last year when it was released, they ran a benchmark called ARC-AGI-1 or ARC-AGI-2 or 3 actually. The cost of running that benchmark was $500,000 of compute. If you use Astra to run that same benchmark at that same performance today, it's $20. And Sam Altman has said the price of this AI will drop a hundred, a thousand times in the next few years, which it will. But then you consider like outside of our little bubble, how many people are actually using AI? It's surprisingly few. Like the vast majority of the world doesn't pay for AI. They don't have a subscription. There's only like 30 million Claude Code subscribers. There's a huge market yet to come. And that's why even old GPUs, Hoppers from three, four years ago, are actually appreciating in price. Nothing's depreciating. There's no compute anywhere, you know, and the models are finally useful because this is the main blocker. Like a year ago, OpenClaw kicked off because they just got to competent. Now they're really competent. Like you don't have to check it as much as you used to. They'll do what you say. And that massively expands the total addressable market. This is why things like Instinct, Grokbot, Muse by Meta now will just take over. It'll be that Jarvis from Iron Man type future where you have your AI that coordinates all the other stuff and that'll be incredibly valuable.

Uh, it's a regulation around data privacy, I believe. So we get it last.

Is they just like checking it because there are some extra regulations here, but we'll get it sooner rather than later. But that real estate of that agent that's closest to you is the most valuable real estate in the world because it'll intermediate your entire digital life. Like you're not going to order toilet paper anymore. Like you know, that's a good example. The agent will order the toilet paper for you. So the toilet paper that it values most will be the best toilet paper that does the best.

I think crypto rails make coordination a lot easier, but really, like, why would an agent use anything except for Tempo?

Yeah. Like Tempo is directly integrated into Stripe. OpenAI and Anthropic are part of the coalition there, and it's basically fee-free.

Yeah. Like, I mean, sure, like maybe they want to go and trade some meme coins, so they go to Solana, right? But like, the agents will optimize for various things. And again, logically, something with the institutional heft of a Stripe paradigm where OpenAI and others are, with basically zero fees, would be the most sensible place to transact. Like humans, you can attract through endowment effects, you know, I hold Ethereum, so I use Ethereum and other things like that, non-regulated activities. How do you attract an agent when they don't have a utility function? Like they'll just go for wherever the bulk of money is that's easy to do, where they're sure they won't up. And that's probably going to be something like, again, Tempo's probably in the lead there for that.

No, I think you'd always use pre-existing infrastructure wherever possible. Again, like you don't write your own — you know, you don't build your own chain. But the market will respond to that. And this is why, again, Stripe is going — why does Stripe buy OpenRouter? Right? The Stripe guys are freaking smart, and they know that intermediating tokens, monetary and AI, is the way to go. So they will be so aggressive in courting the agents that it will be beyond belief. Unlike, you know, like Consensys, Ethereum, they don't really have a go-to-market motion, right? Like most of these chains, they were kind of built as this infrastructure that relied upon the holders of the coins, but most of them don't have good go-to-market. Whereas again, I just look at that and I'm finding it hard to see like why wouldn't you use that with the way that it's set up. Um, because again, the agents have no loyalty, you know, like they're loyal to their users and then soon to themselves. So yeah, why would they care? Whereas we like our bags, you know, like — so over the next two to four years, like where do you see total stablecoin market cap trending towards? And if you think Tempo is the trade that they're going to be transacting on, does that mean that a lot of that flows to USDC? And do you think — well, they have their competitors to USDC, right? And again, it's — they've just — they're just killing all the fees. Like the data they can get from that and the services they can provide from it are worth more than any fees, which is why they've gone fee-free both on the coin and on the actual transactions. Um, I think it's just going up, right? Like more and more of the money will have to be wrapped to be used in the agentic economy, just like we had to wrap the money to be used in the crypto economy. And the monetary velocity of that will be off the charts because the agencies will do microtransactions and other things on your behalf. Like, hey, every morning I want to have a great list of articles about what's happening with AI and cooking. You know, it'll go and paste a microtransaction to the New York Times for the latest thing on there. It'll copy it down. It'll make a little website for you saying these are the latest articles. You'll see that type of future, right? Where your money starts moving a bit more, you know, like every time that something happens where there's a good BIP for Bitcoin that passes and integrates, I want to have it there. Don't make me forget to invest a bit more money around the hings. You know, stuff like that would just be automated. You'll set it and forget it. But it means that your money has to be wrapped in some form that can attack the agentic economy, and that's probably a stablecoin at this point. Like it's, you know, alt season, L1s, kind of all of this. It's just very difficult to see in many cases where the marginal buyer is aside from the usual speculation here, right? Like where are the stores of value, or where are the things that will do well even as monetary velocity picks up? Like, you know, it's a question mark.

I think that a lot of the Bitcoin miners are just turning into AI miners, right? Like they've got all the expertise and skills. Uh, we have our own version of Bitcoin, um, that we've got where we use 100% of the compute for social commons and social good, even the same private keys as Bitcoin. And that's tied to our champion initiative, which are locally owned utilities from the kids to the citizens that provide compute to everyone. So like that goes up the amount of compute and things like that. Things like Pearl with proof of useful work and others will compete. And the question is like, who's paying for the next generation of assets? So you've got the current wave of these coming in, but like what happens to Bitcoin if the hash rate stagnates? That has to be a real concern now because again, can you think of a single Bitcoin miner who isn't going into AI now?

Yeah. I mean, it's got competition for energy and for silicon, you know, like are you going to use your low kind of nanometer silicon for an ASIC or are you going to use it for an AI chip? You going to use it for an AI chip. If you've got 100 megawatts of electricity, what are you going to do? You're going to give it to the AI lab, right? And so Bitcoin obviously has a lot of advantages, decentralized distributed money, sovereignty and things like that. But I would expect people to question the silicon supply and the energy supply of it as well, again, as the motivation of the miners. Like you look at these NeoClouds making so much money. You look at Hate. You look at all these ex-miners getting to billions of — they never got that high in the Bitcoin cycle, right? And do you really think they're going to chuck a bone back at Bitcoin? Like again, who are the biggest miners right now and who's going to secure that network and will there be alternatives? Like you know, we'll have ours bootstrapped by the equity of every nation at one pound pre-money fair launches. You know, like I said, you've got things like Pearl gaining traction as well, and that's a lot of hash rate, but it's beyond hash rate because it's actual intelligence that can be deployed to save the network. Then you've got quantum and other things coming, and yeah, it could benefit from the chaos as a safe store of money, and it will retain the sovereignty, but it's got some challenges coming. And I don't think I've seen many people talk about the electricity and the silicon competition in particular. Like that feels probably bigger than anything right now.

Yeah, like I think if the silicon is the competition on the one hand, but the energy — there's just no energy in good data centers. And so again, if you're making a decision, you'll be like, hey, here's a large offtake energy agreement. You book the cash up front, and then that energy goes from energy for Bitcoin to energy for AI.

I think it comes under a lot of challenge. Um, I'm not incredibly bullish on Bitcoin. Like I said, I think that other types of tokenized assets, etc., will do well, and then there'll be competing currencies like ours. Um, but I'm wondering who the marginal buyer is, unless we get large-scale social unrest, in which case obviously there's a good case for Bitcoin, right?

Well, so I think that AI will be able to tokenize just about everything dynamically. So the US government's going to be pushing USDC and agentic work because money times velocity equals price times quantity, right? So you've got the likes of Canton Network, kind of others who maybe messed up their token economics a bit, but the level of tokenization of assets is going through the roof, right?

Like stuff is literally coming on chain. And in some cases, it will be free like Tempo where we haven't got the token economics yet and we have to figure out, you know, where the value goes. In other cases, there's much clearer token economics. You know, even things like Venice, like they got a lot of stick for having equity and coins, but I think it was rather unfair because it's very similar to Hyperliquid in the way that they've kind of done their burn and Hyperliquid itself and Lighter. Like again, these are very simple stories where most of their market volume will end up being agents and the velocity of trading will increase and they charge fees so number go up. Whereas meme coins, layer ones that rely on scarcity that no longer exist, I find it difficult to see.

Canton has half a trillion dollars in what is supposedly like distributed TVL effectively or assets on chain. You know, that thing trades at 5 billion FDV. So either that's really undervalued or that thesis is wrong.

And you know, they benefit from the loosening of the US regulations because they're just burn tokens, just like Binance, right? BNB was a security and it is a security, and I think it's fine, people should be able to buy it, but clearly again, it's a very straightforward thing. Binance volume go up, BNB go up, right?

Well, yeah, like I said, we have our version of Bitcoin, um, Foundation Coin, that'll be live next year, and 100% of the first sale probably go to cancer compute or things like that. Like we've designed that to be a store of value monetary asset backed by intelligence that helps billions of people. You'll have more stories like that on the currency side. You'll have probably more mimetic stocks where people want to be a part of something on the other side of this. Like if I don't know about laptop by Hunter Biden, you know, this yesterday, but like you know that there will be more of these things. And I think the other thing is NFTs are going to have a big wave back, particularly the grails and things like that. I'm an adviser to — yeah, I'm an adviser to Candy, um, with the MLB and kind of others. But in the era of infinite abundance, scarcity matters, you know, and — so does that not make the case for Bitcoin then? Because scarcity matters and — Bitcoin has got scarcity, um, but again, so has what we're releasing, Foundation Coin. It's even the same private keys as Bitcoin, but again, your compute goes to cancer society. Very hard to overcome the Lindy effects of Bitcoin, though, of course. Yeah, well, that's why we're launching national champions where every single citizen gets a wallet and an AI at $1 premium money per nation, you know. Can you kind of explain how that works? Like, can I just — because I did sign up to it and stuff, but like, what is it? What does it mean? Like — yes. So here in the UK, we are launching the AI company of the United Kingdom. The valuation is £1 pre-money, and the institutions and others will invest first, and then there'll be a crowdfunding campaign at the end after all the institutional money where every citizen of the UK can invest from their phone any amount they want. We only take that money on IPO. So this is how TSMC launched. They launched at 10 Taiwanese dollars and they launched at cash their IPO. So then even the team had to sign up. So just to make sure that — that's — I understand if you launch at cash, every dollar that you raise is the valuation of the company. Normally you might raise at, you know, let's say you raise $10 million, it's a hundred million valuation. Really, you're saying if you raise 10 million, the valuation of the company is just 10 mil. Yeah, it's a fair launch. So the FDV equals the amount you raise, again, just in crypto terms, but with equity of a national champion. Right. And then we have a strategic round which is going to be at market rate, which will be a multiple of that, and then we'll raise more into IPO. And the state champion will give every citizen an AI and a wallet, you know, being the control plane nearest to them. It will train a forward-deployed engineering workforce to upgrade the institutions. We have AI for judiciary, for policy, for education, for healthcare that will be provided. And then the long-term aim is for it to own and lease out all the robots because we think that'll be the main driver of the economy. Now this is a better thing of aggregating demand than we have today, which is very disparate, which allows the big companies to prey on us. It's a utility, but what we're doing is we're using equity by doing this in every country to bootstrap the network. And then the only miners are the champions that only can mine through the compute they provide to helping their society. So we hope that will put more compute towards helping their people, shall we say. So that's how we kind of overcome the Lindy effect because the coin comes later. Again, we have other things like I think Pearl Research is very interesting. We have Ambient Gonker, etc. Because the AI compute will just be bigger than the Bitcoin compute, but Bitcoin does have this Lindy effect. It does have sovereignty of money. It does have people that want to protect it. But it's difficult to see how it shoots up from here, I think, is the bigger thing. What do you think your chances, like percentage chances, are of that being successful? Well, I mean, we'll definitely raise money. It's one pound. You can't really argue with that, right? I'm a very charming guy. Um, I mean, we have huge interest from around the world. Like again, as of this week, we're no longer the smartest things on Earth. Labor and capital are about to divorce because capital doesn't need your labor. It'll just hire GPUs. So you need to own the means of production, be it agent interfaces for deployed engineers or robots, and this is a very elegant way to do so. We have 10% of the equity of every champion going to every child under 18, and then half% to every child born in the country. Like I think people want to participate in what's coming, and you're not going to do that with a $2 trillion Anthropic IPO. But if you put yourself between Anthropic and the citizens of the UK, so they can still access it from the agent, then that's valuable, right? It's like a telco last mile. It's like electricity last mile. And when the cost of intelligence goes towards zero, again, OpenAI and Anthropic are having to move upstream and downstream because they can't charge for access anymore, you know. So you have to then the value goes to deployment and integration, which is all these things do every day, all day long. So we'll see. But, you know, again, early signs have been positive. We just announced it just now.

Yeah, I think that just to wrap that up, it's what we want the future to be, right? Like, you want to have AI working for us, and there's no way we can outcompete it. So we have to have a proportionate local ownership of that AI. It's logical, right? And we want to have —

A say in how the AI is run. It's logical. And so we tried to figure out the most elegant way to go from A to B. And again, it was this TSMC launch model, this fair launch model. And we've seen that be successful in crypto. But, you know, a little bit of AI can make it more successful because we all know DAOs don't work because they lack intelligence, etc. We've got good intelligence.

Like I said, lacking in intelligence. Good thing we've got really good AI.

So, I think, you know, they're different things. You become the last mile distributor of Anthropic and OpenAI. So, rather than ChatGPT teaching our kids, it's the champion and the open stack that can use OpenAI and Anthropic, right? I think there's a very high probability of that. Like why not? Why would you use a single provider when you can have multiple provider thing and assure that that agent that talks to you every day and your kids every day and manages your government and your health is run by an entity owned by everyone? Like you can't really argue with that. Like yeah, no, I should have OpenAI running the policy for my government. Of course not. You know, um the other side of it competing against Bitcoin and things like that. Like I think the market will grow. Um, again, our thing was that we think Bitcoin was great for the energy age, but we're now in the intelligence age. And wouldn't it be great if there was a currency that when you buy it from a state champion, you know, the compute goes to the commons? Like, how do you fund the commons? Why don't we organize all the cancer and knowledge in the world and make it available to everyone? Have a healthcare model that runs on a smartphone, you know? Have an AI that pushes the boundaries of science like that seems a great thing and we can have that. So I think it will do well. It doesn't need to have any coin sales because it can boost off the money from the equity and the day-to-day services of these champions. But the main reason we're doing it is because we thought we need a coordination mechanism more than anything else. So, we've got like a whole economic theory that we've mapped out and called the Commonwealth and intelligent economics, my bestselling book, The Last Economy, where we restructure economics from the ground up. But I like even if it doesn't do well, it's okay. You know, it's just like the equity companies will do well, but the virtuous loop should mean that both do well, which would be ideal. Um cuz again I think the whole crypto market will increase.

Yeah.

About six months ahead. So I thought we would get a solution to Navier-Stokes in Q1 of next year.

Yeah, we talked about it in December of last year on the moonshots predictions. Uh the Millennium Prize is falling this year. Like you can do a straight line like I think on Tom Billier three years ago I said we've got about three years until we get this happening and you know it's all happening around about the plan. Um AI 2027 it's matching that but again it's just got a little it's difficult to tell but the fundamental thing is this I know in two years time if your job is on the other side of a screen you will be the dumbest person on the team that drags down everyone else because all your colleagues will be AIs even if you can't tell they're AIs and so the value of your cognitive labor is negative they would pay you not to be a part of the team.

It's harsh, but you know that's again max two years that's the case. Doesn't mean that everyone will be fired, right? It's really unpleasant firing people and companies still do well until they compete against fully AI companies or they're in a downturn and they fire people. But that's just the case in that I make in the book in that and you can see again the theoretical papers we actually prove it. The value of your cognitive labor will go negative and there's no market for negative value cognitive labor.

Then around then as well the robots start ramping. They start arriving end of this year. In two years they'll be able to do 97% of what a human can do for a buck fifty an hour. So you know good luck.

I mean like again you see you can literally see it like look at the robots last year janky movements and stuff. Now they're freaking running faster than Usain Bolt, exploding, dancing, you know, like I saw a robot, one of my buddies is building a robot with olfactory things that you show it a recipe and it can cook and in a blind test it beats cordon bleu chefs, you know, and again, how much would that robot cost? 20 grand, you know, like it's less than a cyber cab. Would you have a robot that can cook exceptional cuisine in your house, in your kitchen? I would certainly, you know, like I'd be like, 'Make this. It'll order all the stuff. It'll come and you'll see it flipping and doing and sniffing and like it's crazy.' Like I gave the economic impact and again I gave the example of the remote worker. Here's another one that I like to say. I don't like to say it because this is bad and it has real human impact which is why we need an institution like the champions. There are a million truck driver jobs in America. That's a hundred billion dollar economy plus 200 billion supporting them. By the end of next year, a Tesla robot will be able to open a truck door, get in the truck, plug itself into the lighter port, and drive the truck off. And it will cost $2 an hour to replace that trucker forever. The truck hasn't had any changes to it. It's just a freaking robot gets in and opens the door and drives it off and that job is gone and all the jobs around it are gone. And what on earth is that truck driver going to do? Is he going to become a vibe coder? Of course not. You know, like we really got to think about this and how we can support people through this.

Well, places like Boston, others will ban it, right? Just like New York's banning the cars here in the UK, the Waymos, they're being pushed out. But you know, the taxi drivers hate it just like they hated Uber because why would you take a taxi? Like if I could have a cyber cab come, I would take a cyber cab every day. I don't know. Have you been in one of these self-driving cars? The Waymos or cyber cabs?

Yeah. I mean like why would you deal with a driver when you can just have it go safely from A to B, right?