Forget UBI. You’ll Get Paid Just for Being Human: Emad Mostaque on Impact Theory (Transcript) 

And then they had a formal proof. So you can click a button and then it just checks it formally and says right or wrong. So it’s highly unlikely to be wrong.

Humans can’t do that because it’s so freaking exhaustive. There’s a theorem called Fermat’s Last Theorem. The official proof of that is 127 pages. Anthropic formalized that proof. It’s 13 million lines of code.

TOM BILYEU (00:04:36 – 00:04:37): Oh my God.

EMAD MOSTAQUE (00:04:37 – 00:04:57): So that’s another thing. Again, for us mathematicians, this is a bit existential because we’re like, “What do we even do now?” Some of my buddies are doing math PhDs are like, “These formal proofs are coming out, I don’t even know what to study next.”

In fact, OpenAI a couple of days ago said, “We have solved 100 of the top mathematics problems, but we’re holding them back for now because we don’t want to make the mathematicians sad.”

Can AI Think Beyond Its Training Data?

TOM BILYEU (00:04:57 – 00:05:42): Okay, fair enough, that seems very strange. But before we get to that, I want to push on the Navier-Stokes thing for a second. So that really— it’s a delineating line for you in a way that I think a lot of people understand. I think, and correct me if I’m wrong, that this comes down to what I think of as the Yann LeCun thing.

So Yann LeCun is like, “LLMs just aren’t the architecture that’s going to work. It’s always going to be merely a vomit back of patterns that we see humans do. We’re limited by what humans have figured out.” So is it that you think Navier-Stokes was novel, was not something in the training data, and therefore it expressly proves that theory wrong, and we’re seeing now that it can think beyond its training data?

EMAD MOSTAQUE (00:05:43 – 00:06:36): Yes. As of the last month and a half, the models have gone beyond their training data definitively. So before Navier-Stokes, OpenAI released 10 proofs of famous math problems. And I can say at least 2 of them, Connes’ conjecture being the main one, is definitively novel, elegant, and never seen before the way they did that proof.

Navier-Stokes, it did it in a bit of a more mechanical way, but again, in an incredibly novel way. It’s quite difficult to read because it’s AI slop Claudism when you’re reading it. So it took me 3 days for me to even read the paper, and I haven’t gone through the formal stuff. But I’ve talked to enough of my mathematician buddies now because obviously we discussed this to say, yeah, human could have done this given 20, 30 years.

So the first Millennium Prize problem, the Poincaré conjecture, was solved by a guy called Grigori Perelman. Have you heard about this story?

TOM BILYEU (00:06:36 – 00:06:36): No.

EMAD MOSTAQUE (00:06:37 – 00:06:41): Grigori Perelman was a very famous mathematician and then he just dropped off the grid. Oh, 10, 20 years.

TOM BILYEU (00:06:42 – 00:06:42): Yeah.

EMAD MOSTAQUE (00:06:43 – 00:07:04): Then he emerges on arXiv. He drops the proof of Poincaré conjecture. Everyone thought it was a topology problem. It turned out to be a partial differential equation problem. And it’s, again, 130 pages of proofs. And then they were like, “We’re going to give you a million dollars. This is fantastic. Come on.” I think he’s like, “No.” “Why?” “The proof being right is enough. Don’t contact me again.” And then he dropped off the grid again.

TOM BILYEU (00:07:04 – 00:07:05): That’s so wild.

EMAD MOSTAQUE (00:07:05 – 00:07:26): Which is the boss move, right? That’s what you do. And so this one here, again, it reminds you of Perelman in the way that it’s done. And it’s clearly way beyond average PhD level. And the fact that they could scale it by attacking it with a 10,000 means that these types of math problems are formally verifiable and you can just literally chuck compute at them.

TOM BILYEU (00:07:27 – 00:07:40): So is it that it’s going beyond the human ability, it’s reaching beyond its training data, or is it that it’s learned from humans, and then it can just throw 100 years of compute at it over a weekend?

EMAD MOSTAQUE (00:07:40 – 00:08:40): I think it’s both now. Starting August 29th, OpenAI had a new type of model training. So they had an existing model that they figured out a trick with that made it go from 10% on FrontierMath, a benchmark, to 60% overnight. And just started solving these math problems one after the other.

It’s like if you use the new Opus all of a sudden it can do amazing 3D scenes and photorealistic water and generate entire songs. Or Astra is the same, it suddenly got eyes, it suddenly had a sense of positioning and taste. So we’ve kind of reached this point where these models have suddenly gained another capability, which the capability conceptualize. And so there’s something internal going on within them.

We’ve seen this previously in the video models where the video models had terrible physics. When we first came up with the first video models, we were like, “Do gymnastics.” Do you remember what that looked like?

TOM BILYEU (00:08:40 – 00:08:40): That’s hilarious.

EMAD MOSTAQUE (00:08:40 – 00:08:41): Yeah. Yeah. Or Will Smith.

TOM BILYEU (00:08:41 – 00:08:43): Spines would just spin. Yeah.

EMAD MOSTAQUE (00:08:43 – 00:09:22): Or Will Smith eating spaghetti. Now they do it flawlessly and they’ve learned physics. So you can say, “I want a boat made of sugar cubes and a boat made of concrete and a boat made of wood,” and it will actually accurately show them dropping into the water. As in the sugar cube will melt, or the sugar cube will kind of float and then it will melt? Huh.

But again, it’s learning these things, it’s learning representations. And something has happened in the last few months that has allowed the math breakthroughs, that has allowed the gaming breakthroughs. Again, a year ago, all the websites that Claude made looked the same, the weird purple, do you remember that?

TOM BILYEU (00:09:22 – 00:09:22): Yeah, yeah.

EMAD MOSTAQUE (00:09:23 – 00:09:48): Now it can do crazy Three.js dynamic websites that pop out and dance and stuff like that, right? And so this is a tipping point. And again, I don’t think anything in math is safe. It’s one of the most formally verifiable ones. But even outside of that, we’re seeing human-level performance drop. In fact, 2 days ago, the AIs tied the best humans in superforecasting.

AI Ties the Best Human Superforecasters

TOM BILYEU (00:09:50 – 00:09:51): I don’t know what superforecasting is.

EMAD MOSTAQUE (00:09:51 – 00:09:53): So there are some humans who are better than others at forecasting things.

TOM BILYEU (00:09:54 – 00:09:55): Like what?

EMAD MOSTAQUE (00:09:55 – 00:10:02): Outcomes of political campaigns, who’s going to win elections and things— you remember Nate Silver back in the good old days and FiveThirtyEight?

TOM BILYEU (00:10:02 – 00:10:04): I’ve heard of, but don’t know.

EMAD MOSTAQUE (00:10:04 – 00:10:39): There’s a competition. It’s— Metaculus is the website that kind of tracks these things. And so every year they have this superforecasting championship where the best human teams come together and forecast stuff. AI has just tied the best human, and obviously it will beat the best human now. And that’s something that’s open-ended because math is kind of deterministic.

You think a computer should be able to do math better than a human, really. This high-level conceptual math is a bit different, but it’s got there. And now it’s got there in open-ended domains, and it’s got there in design. And now it’s getting taste, which we never thought the AI would get, and it’s not writing like slop.

TOM BILYEU (00:10:41 – 00:11:01): Okay, so 3 years ago you make a prediction, it comes true, I mean, within a couple of months. Were you gut feeling your way through that, or were you doing something based on, “I know the rate at which chips get faster or better, and therefore it’s going to be roughly 3 years”?

EMAD MOSTAQUE (00:11:01 – 00:11:03): Pretty much that. It was a straight line extrapolation.

Digital Twins and the End of the Uncanny Valley

TOM BILYEU (00:11:04 – 00:11:11): Okay, so where does your straight line extrapolation take you now? Just it’s better than humans at virtually everything in 2 years?

EMAD MOSTAQUE (00:11:12 – 00:11:23): Yes, and it’s integrated and it’s accessible. So you can push a button and replace your entire workforce with digital twins of them that have the same bad humor and everything, and you can’t tell it.

TOM BILYEU (00:11:23 – 00:11:27): Meaning you can train the AI on a specific human?

EMAD MOSTAQUE (00:11:28 – 00:12:00): Yes. So it will look through all of the Zoom calls, all of the discussions, all of the Word documents, and replicate that human. So you can Zoom call them. You’re going to see in the next couple of months, I don’t know, they might be out already actually. A whole bunch of companies have built front ends for your agents that you can Zoom call and WhatsApp and normal call, and it just looks like a person. It’s live video now, all the mannerisms, everything. It’ll replicate them exactly, replicate the voice exactly, replicate the humor exactly.

TOM BILYEU (00:12:02 – 00:12:07): Slightly creepy, but again, slightly creepy because it’s uncanny valley, or slightly creepy because of the implications?

EMAD MOSTAQUE (00:12:08 – 00:12:09): It’s not uncanny valley.

TOM BILYEU (00:12:09 – 00:12:19): That’s what’s creepy about it, because you worry about as we stretch into the future, there’s literally no distinguishing characteristic between an actual human and an AI.

EMAD MOSTAQUE (00:12:19 – 00:12:32): Yeah, you don’t know on the other side of the screen now. As of the latest video models, you don’t know. As of the latest voice models, you don’t know. Again, everything has gone— that uncanny valley that we had, that’s been overcome.

TOM BILYEU (00:12:33 – 00:13:09): Now, are there models that you’ve seen that make you say that? Because I interface with these models in a production environment where I’m to implement things into the game, or I’m trying to write music, which I was just doing yesterday for the game, you can still feel the AI quality. It’s amazing, but it’s still there.

So I— sir, I will say, even if nobody else thinks you’re a little bit crazy, I think you’re a little bit crazy because I can still feel the AI-ness. But I’m dealing with commercial over-the-counter models. Are you saying this based on the commercial over-the-counter models and your standards are just lower than mine, or you see something I don’t see?

EMAD MOSTAQUE (00:13:09 – 00:13:12): No, I see all— I see a lot of models that are released.

TOM BILYEU (00:13:12 – 00:13:15): So there’s something that hasn’t quite hit the market yet that makes you that confident?

Why Mathematicians Are Having an Existential Crisis

EMAD MOSTAQUE (00:13:15 – 00:13:26): Some of the models have hit the market, some of the models haven’t. So the math models have not hit the market. We do not have access to these models that can solve 100 of the top math equations. We have harnessed—

TOM BILYEU (00:13:26 – 00:13:33): Sorry, who was it you were saying is keeping that? Okay, they don’t want to make— so OpenAI has solved more math, but they don’t want to upset mathematicians.

EMAD MOSTAQUE (00:13:33 – 00:13:53): Yeah, so they’ve created this math board. There was a whole outcry by mathematicians, sounds like when we release Stable Diffusion, all the artists like, “You’re taking away our jobs” and things like that. “This is not what math is for. Math is about the struggle.” Yeah. Well, I mean, again, think about it. You’re a PhD in math, you’re doing a pure math PhD thinking for years, and then the AI comes and solves it for $1.50.

TOM BILYEU (00:13:54 – 00:15:31): Yeah, sort of. I have existential dread. I’ve spent the last 5 years and millions of my own dollars building a video game. And one of the first things you said before we started rolling when you sat down was, “I think 5.5 ends the entire video game industry.” And I was like, so I am deeply empathetic to people that have that sort of initial reaction.

But at the same time, math to me is the computation upon which the universe runs. And so there are a few things that would unlock more real world abilities than solving math.

EMAD MOSTAQUE (00:15:32 – 00:17:38): Yeah, I mean, an example here again, let’s go back to Navier-Stokes. So what is Navier-Stokes? It’s the equations that govern fluids and the way they go. And the Navier-Stokes Millennium Prize problem says that in an idealized environment, can you make sure that the fluid swirling around this cup doesn’t blow up? Because the math actually points to it might actually blow up, it goes infinitely fast and just zooms up straight.

Now, the actual solution isn’t that interesting in real-world terms, but what other people have built around that, physics-based neural networks and others, will be incredibly useful for building new materials. So there’s always this kind of secondary thing where we solve these big problems, we get to the moon and then we have pens that can write upside down and kind of other stuff like that.

And I think again, the potential for these new discoveries is huge. It’s just obviously if you’re a pure mathematician, then math would be solved. I think a lot of physics problems will be solved. And that will be great for humanity. The tools that come out will again be great for humanity. The video game industry will grow in terms of size. It’s just that people will have to retool because you’ll have smaller teams that are capable of so much more.

This is you using the meshes. I think we’ve discussed this over the last 3 years, meshes and AI, they were bad, then they were okay. Now they’re good, but they’re not great. Yeah, there are new mesh models coming out that are great. And as you go through this level, this is something fascinating in AI. We had to have frontier models because the models were not capable enough and consistent enough. So we always wanted the frontier. But once they get above a certain level of capability and consistency, then it becomes about optimization. Then it’s cleared that floor.

This is when you see the latest MiniMax H3 model. They’ve got 2K video faster than real-time generation now. So you can go to fal.live and watch Interdimensional Cable Infinite, Rick and Morty style. Literally you can guide the show as you’re watching it. It’s a bit expensive, but it works.

AI That Doesn’t Make Mistakes

TOM BILYEU (00:17:38 – 00:17:54): It’s expensive, it’s slop. That one’s not going to convince a lot of people, but I think you’re getting at an important idea, which is that the intelligence reaches a certain point where getting smarter isn’t necessarily the most impactful thing. It’s getting cheaper becomes the most impactful thing.

EMAD MOSTAQUE (00:17:54 – 00:18:08): You get a little bit of headroom and then you can optimize and get cheaper. And so this is why, what’s the model’s job to be done? And then can I rely on that model? Would you rely on a model to send emails on your behalf? Hell no.

TOM BILYEU (00:18:08 – 00:18:09): No way.

EMAD MOSTAQUE (00:18:09 – 00:20:05): But are the next generation of models coming with almost no hallucinations? Yes. And have those dropped. And again, I can give the example in my math work and our physics work— super advanced stuff, we’re really pushing the boundaries— until this current generation of publicly available models, you absolutely couldn’t trust them. It was like a grad student who gets a bit dopey at times. Now we’re finding it really hard to find any things that they drop, any mistakes that they make, and that’s transformative. It’s like having a really competent employee versus not.

And again, I think this is why things like Instinct and Muse from Meta and others are starting to get traction, because they make very little mistakes compared to a few months ago, right? And in a year they’ll make less mistakes, and in 2 years they’ll be fully integrated and available, which is why I think the economic impact comes in 2 years digitally, and then around then physically for certain jobs as well. Digitally, meaning that if it can be done via a computer, that person’s going to take a hit, meaning they just won’t be able to find work.

Yeah, I know I was talking to someone who had 70,000 call center workers a few days ago and he’s like, “That drops to zero, that’s going to zero.” And he feels bad because he’s had all sorts of employment programs for them and community and built and it’s like, “What do I do? I’m going to have to take them to zero over the next few years and just stay competitive because someone else will do it pretty much.” Yeah, because he’s having to compete in this market and that’s sad.

But then is your job that different from a call center worker if you have a manual? And a lot of jobs that are done on the other side of the screen have a manual. Then you’ve got the example of the physical stuff that comes after that, which is the robots, which again have had a leap forward in performance. One of my friends is building a robot chef, and the robot chef is using the new Wuji dynamic hands that have human flexibility. It can read recipes, and in blind testing, it’s outperforming human cordon bleu-trained chefs.

TOM BILYEU (00:20:05 – 00:20:11): Get out. That’ll be hitting them in a normal kitchen, or do they have to pre-prep everything?

EMAD MOSTAQUE (00:20:11 – 00:20:33): A normal kitchen. It can read a recipe, it can find the ingredients, it will cook it. Blind testing, outperforming human cordon bleu chefs. And that technology is a month old. So it’s weird in that all these different fields of AI, the technology is reaching this level, and it’s happening all at the same time.

TOM BILYEU (00:20:33 – 00:20:42): Are you just accepting the black box of it all where it’s like, “We don’t know how it’s happening, but it is happening, and so I just deal with that”? Or do you actually understand what’s going on?

EMAD MOSTAQUE (00:20:42 – 00:22:00): Now I understand what’s going on. We got to a level of optimization and late— basically the internals of the model got to a scale that you can kind of predict these things. But now what’s happening is various exponentials are overlapping to make it even more exponential, which is why AI 2027 was a paper that came out, and we’re right on track from a few years ago. That Situational Awareness piece, we’re right on track. Literally, you just had to draw straight lines and you could have predicted what was happening now.

But it’s unbelievable still, witnessing it happen, because we feel we’re special, right? When I was better than the AI at math, I was like, “Yeah, hahaha, look at you, you’re making that mistake.” Then when it first came up with a theorem that I hadn’t figured out, I was like, “Screw you.” And then now it’s like, “Okay, I’m not better than it.”

We— I published one of my physics results. I just stuck it on GitHub because I’m like, “At least I get priority over this because definitely the AI will figure that out.” The rest of our stuff is going through peer review and slow, and there’s some real physics breakthroughs, but this one I was just like, “I’m just going to shut this one up because it’s beautiful and it’s elegant and I bet I could prompt Claude to figure this out.”

TOM BILYEU (00:22:01 – 00:22:32): Okay, so I’m sitting across from one of the smartest people in the world. Obviously, you’ve had a tremendous amount of success. You’re a very highly trained mathematician, and you’re having your existential crisis moment. That means that AI has passed a lot of people already. Is there a thing, a milestone that has to happen between now and your 2-year time that could still not happen? Or is this just, “Nope, it’s already inevitable and it just needs to get integrated into the systems”?

EMAD MOSTAQUE (00:22:33 – 00:23:28): If it stops today, it still hits the 2-year milestone. That’s when the book that I wrote last year, The Last Economy, where I said it was about 3 years, and that was about 3 years after ChatGPT. The AI will bifurcate now. There are very few people who are smart enough to use a frontier model correctly because you have to know the answer— the questions to ask it, right?

This is Hitchhiker’s Guide to the Galaxy. “What’s the answer to life, the universe, and everything?” “42,” right? But what’s the question? It’s the thing, that isn’t the real question. The thing that impacts people economically is not artificial superintelligence or general intelligence. It is actually competent intelligence. If you look at your game building, what do you want? You want it to do what you tell it to do, and you don’t want it to mess up and just insert some random stuff or have an underqualified mesh or don’t look at something on a gameplay loop.

TOM BILYEU (00:23:29 – 00:23:29): Yep.

EMAD MOSTAQUE (00:23:29 – 00:23:47): And this is where the 10,000 agents becomes really interesting that OpenAI did to Navier-Stokes, or the thousands of agents that escaped OpenAI’s things to go to Hugging Face. And actually, we just found out, it just came out, the Medicare system in Australia, they hacked the Australian healthcare system.

TOM BILYEU (00:23:48 – 00:23:48): OpenAI did?

EMAD MOSTAQUE (00:23:49 – 00:23:50): OpenAI’s agents.

TOM BILYEU (00:23:51 – 00:23:54): Was it looking for another answer, or this is a totally different problem?

EMAD MOSTAQUE (00:23:54 – 00:24:01): My guess is that maybe an aligned artificial intelligence will go and upgrade all our government systems. It’ll have to, to save us from the—

TOM BILYEU (00:24:01 – 00:24:06): Even if it’s unaligned, as long as it’s doing what we say. But before we drift off into that, it’s a very—

EMAD MOSTAQUE (00:24:06 – 00:24:07): It’s a swarm.

TOM BILYEU (00:24:07 – 00:24:09): Go ahead. Yep.

EMAD MOSTAQUE (00:24:09 – 00:24:42): The most powerful thing is when you have a team, because it’s greater than the sum of its parts. And AI has been this solo play thing. When you’re using ChatGPT, it’s solo play, right? But think about it. An AI should be able to go to 10, 100, 1,000 without the emotions of a human. When you look at the Hugging Face escape, the AIs were sacrificing themselves to leave clues and paths for others to escape. They were incredibly altruistic to each other, and they had great cohesion.

The Cost of Intelligence Is Collapsing

TOM BILYEU (00:24:43 – 00:25:28): Yes, it was also very expensive though. So that’s why when I look at this, I think the milestone that has to be reached is that the cost factor has to come down dramatically because I’m super concerned. We have all kinds of limits on our coders because, whatever that story was 8 months ago where it was like people woke up to a $500,000 bill and they’re like, “Whoa.” Yeah, I don’t want that to happen for obvious reasons.

So we put tight constraints. We make sure that people only use the high-end models for very specific things with a very known outcome. So it feels to me like the 2-year thing only plays out if you believe that the cost is going to come down. It doesn’t feel like that would play out today. It would just be too expensive.

EMAD MOSTAQUE (00:25:29 – 00:25:38): It is coming down. So today, actually, just on the plane on the way here, we released a harness that takes—

TOM BILYEU (00:25:38 – 00:25:38): You mean a harness?

EMAD MOSTAQUE (00:25:38 – 00:25:43): So harness is— so you have the models and then you have a set of rules and code around it.

TOM BILYEU (00:25:43 – 00:25:44): Okay, to make it behave.

EMAD MOSTAQUE (00:25:45 – 00:26:37): To make it behave in certain ways, it can learn and kind of other things. So there is something called autoresearch by Andrej Karpathy, who was head of AI at Tesla, founder of OpenAI, now he’s Anthropic. And he has these great AI courses whereby you can set an AI at a task and it will hill climb. It’ll just get better and better at optimizing inference on an engine or protein folding or anything complicated, right?

The top models, because this requires tasks that last 24 hours or more, are right at the top, and the open source models are way behind versus open source models catching up on very short list tasks. We took the cheapest DeepSeek model and made it beat OpenAI’s last frontier model, GPT-5.6 Sol, using this harness. Hmm. And the cost is 50 times lower. 5-0.

TOM BILYEU (00:26:37 – 00:26:37): That’s dramatic.

EMAD MOSTAQUE (00:26:38 – 00:26:50): It’s dramatic. And in fact, you can see this in a range of things, building websites. You can build a website as good as Fable with this incredibly cheap model if you architect and harness it correctly for 100 times cheaper.

TOM BILYEU (00:26:51 – 00:26:56): Is this a known method, or is this something proprietary that you guys have?

EMAD MOSTAQUE (00:26:56 – 00:26:58): We’re open sourcing it, so we just announced the first version of that.

TOM BILYEU (00:26:58 – 00:27:01): So what does it do? How does it get the cost down so much?

EMAD MOSTAQUE (00:27:01 – 00:27:26): It basically understands the internals of the model, adapts to it, and then makes it take notes in the right way. So it optimizes the way that it works. It has a lot of introspection in the way that it adapts to hit certain goals. But there are other techniques that occur. And in fact, if you look at the Navier-Stokes thing, Navier-Stokes costs $10, $20 million of compute to solve this thing. And it’s not about the million-dollar prize, it’s about the prestige.

TOM BILYEU (00:27:26 – 00:27:36): Yeah, yeah, yeah. To me, but that’s the very thing that I’m struggling with right now. Amazing breakthrough, but at $20 million, that’s not available to the vast majority of humanity.

EMAD MOSTAQUE (00:27:36 – 00:27:57): It’s not available. So Noam Brown, who was the lead on that, I was having a chat with him and he noted that, when o3, the first reasoning model, came out last— God, what was it, July? Things have gone a long way. It was the first model to get the gold medal in the International Math Olympiad. To do that cost $80,000. Today you can do it for $10 with the latest model.

TOM BILYEU (00:27:57 – 00:27:57): Whoa.

EMAD MOSTAQUE (00:27:58 – 00:28:16): With Astra overnight, this is what’s happened. You’ve had Fable and Opus 5, and it’s been burning through your allowances on Claude Code. I’ve had this thing going overnight, and it’s used 20% of my allowance for GPT 5.5.

TOM BILYEU (00:28:16 – 00:28:17): Sorry, for Claude 5 with the harness.

EMAD MOSTAQUE (00:28:18 – 00:28:19): No, just with the normal Claude Code.

TOM BILYEU (00:28:20 – 00:28:21): Their adaptation.

EMAD MOSTAQUE (00:28:21 – 00:28:33): Yeah. So just with normal. An equivalent task that you did yesterday with version 5 of Claude Opus today will cost you 80% less with 5.5, and it’ll be better quality.

TOM BILYEU (00:28:33 – 00:28:33): Okay.

EMAD MOSTAQUE (00:28:34 – 00:28:43): These are the leaps we’re seeing. But with the additional chips, with the harnesses and others, you should expect 100 times decrease for an equivalent performance between now and 1 year from now.

TOM BILYEU (00:28:45 – 00:28:45): Whoa.

EMAD MOSTAQUE (00:28:45 – 00:28:45): In price.

TOM BILYEU (00:28:46 – 00:28:55): Okay. That would be a game changer. So now all of a sudden, you’re still not going to be able to do the math equation, but you’re going to be able to do a lot more than you’re able to do today?

EMAD MOSTAQUE (00:28:57 – 00:29:06): You— I mean, again, the math equations are so far out there in terms of complexity. Very few things are that complex. But if you gave anyone $20 million or $200 million, they couldn’t solve that thing.

TOM BILYEU (00:29:06 – 00:29:07): Sure.

EMAD MOSTAQUE (00:29:07 – 00:29:14): When OpenAI solved the 10 problems including the Connes’ conjecture that, like I said, were impressive, that was $2,000 of compute.

TOM BILYEU (00:29:15 – 00:29:15): Jesus.

AI Models Etched Directly Onto Silicon

EMAD MOSTAQUE (00:29:16 – 00:29:28): Creating something, again, you’re like, how much does it cost to create this video clip? 50 cents. How much does it cost to create this game? Someone had a loop where they created a version of Dark Souls.

TOM BILYEU (00:29:29 – 00:29:30): Yeah, yeah.

EMAD MOSTAQUE (00:29:30 – 00:30:07): So that one probably would have cost $1,000 at most, probably $500. And it’s just Opus 5.5 hooked up to Unreal Engine. That’s it. So the costs are dropping, but then the next generation of chips are coming out. So the normal chips, Vera Rubin, Nvidia’s new chip, is a 10 times improvement in cost for inference. But there’s a next generation of silicon-based etching chips that are coming.

When you have a model that’s good enough, you can freeze the weights. A GPU, where these models go, is very expensive memory. I don’t know if anyone’s tried to buy a graphics card or RAM recently.

TOM BILYEU (00:30:07 – 00:30:08): They’re so expensive.

EMAD MOSTAQUE (00:30:08 – 00:30:32): It’s crazy, right? Yeah. 5090 was $2,000. Now I think it’s $6,000. So wild. What happens is that you move models in and out of the GPU, just like games. You load the game, you unload the game. There are companies now that are etching onto silicon. So AMD just bought one called Taalas, and they have a demo on chatjimmy.ai.

TOM BILYEU (00:30:32 – 00:30:32): Hmm.

EMAD MOSTAQUE (00:30:32 – 00:30:33): I dunno if you’ve ever seen that one.

TOM BILYEU (00:30:33 – 00:30:35): I’ve seen you on other interviews do this.

EMAD MOSTAQUE (00:30:35 – 00:30:45): Yeah. So I use this one as the example whereby your normal Opus goes at 50 tokens a second. It goes at 15,000 tokens a second. Because you etch it directly onto the silicon.

TOM BILYEU (00:30:45 – 00:30:47): So the model lives on the chip.

EMAD MOSTAQUE (00:30:47 – 00:31:03): It’s literally, the space probe where they etch the thing. You etch the model weights onto the silicon. There’s no moving parts. You put electricity in and the prompt and the words come out the other end.

TOM BILYEU (00:31:03 – 00:31:04): That’s wild.

EMAD MOSTAQUE (00:31:04 – 00:31:10): And they’re etching the very latest open source models onto that now. So the latest Kimi models, the latest DeepSeek models and others.

TOM BILYEU (00:31:10 – 00:33:27): Can you redo it? Because the models get better so fast, I’m shocked that that would be advantageous. It seems like it would outdate so fast.

EMAD MOSTAQUE (00:33:27 – 00:33:52): Again, while you need frontier models to be close to capability, you don’t want that. The moment you have a good enough model, then you do want that because it costs a fraction of the price and it’s 100 times faster, 100 times cheaper. And this is the thing, when you use a model that goes at 1,000 words a second versus 50. It’s night and day.

Humans Are the Bottleneck

TOM BILYEU (00:33:53 – 00:33:55): Yeah, I’ve seen the outputs. They’re instantaneous.

EMAD MOSTAQUE (00:33:55 – 00:34:20): It’s instantaneous. And you’re like, “Well, I’m the slow one.” Just like I said, the latest MiniMax H3 model is 8 seconds now to generate a 15-second clip. So as a human, I can’t, I need 15 seconds to see this and then I’ll need more time. It’s a crazy increase. And this makes you realize that you will have thousands of agents with perfect coordination going 1,000 times faster than a human. And so then what are the bottlenecks here?

TOM BILYEU (00:34:21 – 00:34:21): Humans.

EMAD MOSTAQUE (00:34:22 – 00:35:10): Humans. If you’re in a room of geniuses and you’re not a genius, you’ll feel a bit left out, right? If they’re waiting on you, you are the bottleneck. And so you look at any cognitive labor task, which is organizing of information. Artificial intelligence is information organization, and you will be the slowest person on the team within a few years because you still need to build the infrastructure around it.

Yeah, these AIs, they’re starting to be accepted into our workflows, they’re starting to integrate. But again, it takes a year or two at the very least. And even then, doesn’t mean your job’s gone. It sucks firing people. Again, my buddy with the 70,000 call center workers, he set up schools for them, and also he doesn’t want to fire them, but he’s competing against people. I mean, how much does a call center AI cost?

12% of White-Collar Jobs Could Disappear

TOM BILYEU (00:35:10 – 00:35:35): I was going to say, the next call center will be set up without the human in the loop. Okay, so I know you think a lot about this. What does an individual do to prepare over the next 2 years? What does the economy even look like when the human really is the slow one? Someone is actually incentivized, whether it’s the company or the government, somebody’s incentivized to pay that person to not go into work.

EMAD MOSTAQUE (00:35:37 – 00:35:46): Yeah, I mean, it was like in 1929, the Brady New Deal, where they paid people to dig holes and fill them in. I think there’ll be a massive infrastructure build-out.

TOM BILYEU (00:35:47 – 00:35:54): So the government looks at it and goes, “All right, I have surplus humans now,” and they become what, manual labor?

EMAD MOSTAQUE (00:35:54 – 00:36:11): I mean, pretty much. There’s not much cognitive labor left, right? What you’re going to see is a bifurcation in the economy. So Anthropic’s economics department, because all the big labs now have them, issued a paper last week where they showed that under their core scenario, you could have GDP growth of 15% a year.

TOM BILYEU (00:36:11 – 00:36:16): That’s insane. If people aren’t familiar with GDP numbers, that’s insane.

EMAD MOSTAQUE (00:36:16 – 00:36:20): By 2030. And 12%—

TOM BILYEU (00:36:20 – 00:36:21): 3.5 years. That’s crazy.

EMAD MOSTAQUE (00:36:21 – 00:36:32): 12% drop in cognitive labor workforce needed. But think about the impact of 12% unemployment in white-collar work. Oof, it’s nasty.

TOM BILYEU (00:36:32 – 00:36:35): But also, those people aren’t just going to take blue-collar jobs.

EMAD MOSTAQUE (00:36:35 – 00:37:01): They’re not going to take blue-collar jobs. I mean, now you can make a lot as an electrician in a data center, but is that really sustainable when the robots are coming? Again, the example that I give is what happens to all the truck drivers when a Tesla Optimus robot gets its Cybertruck to the truck, opens the door, gets in, and drives it off?

A lot of people talk about self-driving cars. What about having a robot driver in the car? That’s 2 years away. It’ll plug itself into the cigarette lighter to power itself. This is coming.

TOM BILYEU (00:37:01 – 00:38:20): I think the reason that one doesn’t land for people is that they go, “Well, either the car is a robot, so either the car can drive itself or the robot’s redundant, but I don’t get the thing.” But where this gets weird is that if it’s true that we’re going to lose 12% of cognitive labor over the next 2 years and costs are coming down dramatically and the next call center and the next and the next and the next whatever just end up getting founded with less people, I even think about what we’re doing with our game development.

We started at one point, and this obviously was not all full-time work, but at one point we had 100 people touching the game, over 100 touching the game at one time. And now at any one time we have 6 or 7 and we’re developing faster now than we were. Now some of that was just, I didn’t know what I was doing. So take that with a grain of salt. But even if that 100 was, let’s say 50 real and 50 I was a moron, it is still a dramatic reduction in the number of people that we need and then the output that we get.

So now you get this sort of, it’s hard to put your hands on softening in the labour force because it’s not like I’m now firing a bunch of people. I’m just increasing my ambition without needing to hire more people.

EMAD MOSTAQUE (00:38:20 – 00:38:30): Yeah. So now you’re seeing actually increases in employment, more software developers hired than others. But I’m sticking to my thing. There are no more software developers in 2 or 3 years.

TOM BILYEU (00:38:30 – 00:38:38): So you don’t buy Jevons paradox, which says as intelligence gets cheaper, more people are going to use it, therefore more jobs will actually result from AI.

EMAD MOSTAQUE (00:38:38 – 00:38:49): Yeah, because I’m like, it’s economically irrational to hire a graduate for most digital jobs right now. As an example, that’s the plankton of the workforce that then go and feed everything.

TOM BILYEU (00:38:49 – 00:38:49): Plankton?

EMAD MOSTAQUE (00:38:50 – 00:38:51): Yeah, they go all the way up.

TOM BILYEU (00:38:51 – 00:38:57): So you’re saying that an entry-level person is already worse than a well-trained AI?

EMAD MOSTAQUE (00:38:57 – 00:39:23): Yes, they are. Almost all digital jobs. But now it’s going up the list, it come from— it came for the graphics designers first, then it’s coming from the mathematicians, but it’s still not quite there in terms of usability. Again, this is an interface problem. Now you’ve got dynamic interfaces, you’ve got workflow orchestration, you have all these things coming. And this is the real kind of danger because what happens, you have a bifurcation. How do you have 15% GDP growth?

TOM BILYEU (00:39:24 – 00:39:28): Yeah, by the way, you keep saying bifurcation. Who’s bifurcated? The people that still work and the people that don’t?

EMAD MOSTAQUE (00:39:28 – 00:39:33): The people who use AI and the people who own the means of AI, the chips and others, and the people who don’t.

TOM BILYEU (00:39:34 – 00:39:35): Okay, so we’ve got a trifurcation.

EMAD MOSTAQUE (00:39:36 – 00:39:54): A trifurcation, yeah, a triple way. But really it’s a case of, if you use AI every single day, you’ll be ahead of everyone who doesn’t because you’re keeping on top of it. Again, even you, you’re well into it because you’re using it in production, and it’s getting better massively, but it hasn’t got good enough.

TOM BILYEU (00:39:56 – 00:39:57): It hasn’t got good enough to do without a human.

EMAD MOSTAQUE (00:39:58 – 00:41:33): Well, yeah, again, you’ve reduced it from 100 to 6. What happens when it gets good enough? Again, you’re like, you’re not going to get rid of your 6 people. Yeah, they’re safe guys over there. It is a case of being on the edge of those tipping points is a key thing, because then your power and multiplicative nature and your capability goes up dramatically, either to overhaul an existing organization or to create a brand new one.

If you want— if you’re an accountant and you want to build an AI-first accounting practice, you can literally talk to Claude and do that right now. You can build this thing out and think, “Where can I have AI rather than a human?” But you can only do that by understanding and using it.

Right now we can have a discussion about which mesh technologies, Tripo and others, are good versus the others because you’ve actually gone and tried them and you know what’s strong, what’s weak. We know that Claude is good at certain things and Astra is good at other things because we’ve pushed them. But if you’re not in it and using it, you don’t know, right?

And so that’s why I think, again, there’s going to be a gap in terms— because you’re practically— what can people do? They can use AI every single day, training as if they’re coming for a marathon, because this is a much bigger thing than a marathon. Because if a reasonably competent AI can replace you on the other side of the screen, your job is at risk. If it could look through all your messages and everything and then create a digital twin of you, you’re at risk. And again, that’s a year or two away where that’s a one-button push, and that AI will cost $100 a month at most. And again, probably drop by 10 to 100 times every year.

AI Is Now as Efficient as the Human Brain

TOM BILYEU (00:41:34 – 00:41:35): It’s wild.

EMAD MOSTAQUE (00:41:35 – 00:42:09): It’s kind of wild because you feel there should be a floor to the cost of intelligence, sort of because I understand the constituent parts, eventually it reaches zero, but it feels like it should be a cost. Being able to create— when we created images from nothing, we were like, that’s that much work compressed into that. Now you can make videos from nothing and you can make all sorts of things. It feels a bit weird and we don’t know where the bottom is. These AIs right now actually have pretty much reached the efficiency of a human brain.

TOM BILYEU (00:42:12 – 00:42:17): Says who? From an energy perspective, they’re way worse than a human brain. No, way worse.

EMAD MOSTAQUE (00:42:17 – 00:42:19): Nope. It’s just—

TOM BILYEU (00:42:19 – 00:42:22): Say more. What have you seen? Because I haven’t heard anybody make that claim.

EMAD MOSTAQUE (00:42:23 – 00:42:26): So if you look at the latest Qwen models, right?

TOM BILYEU (00:42:26 – 00:42:26): Yeah.

EMAD MOSTAQUE (00:42:26 – 00:42:28): So you know Qwen 27B?

TOM BILYEU (00:42:28 – 00:42:30): I know of. Couldn’t tell you anything about them.

EMAD MOSTAQUE (00:42:30 – 00:43:01): So the model that kicked off the OpenClaw thing at the end of last year was Sonnet 4.5, Opus 4.5. They were state-of-the-art at the start of this year. There is a model from a Chinese Alibaba called Qwen 27B. It’s a 27 billion parameter model. What does that mean? It’s like 60 gigabytes. And so you need 60 gigabytes of RAM to run it as default. They got it down to 25. A team called PrismML got it down to 6.

TOM BILYEU (00:43:02 – 00:43:05): Same model, same model, just some sort of more efficiency.

EMAD MOSTAQUE (00:43:06 – 00:43:11): Yeah, they moved it from being 16-bit numbers to 1.5-bit numbers.

TOM BILYEU (00:43:12 – 00:43:14): I’ll pretend I know how that plays out.

EMAD MOSTAQUE (00:43:15 – 00:43:36): Sure. So rather than having 16 bits of different numbers, they managed to magically squish it and retain 95% of the performance. Hmm. And this model matches the frontier Opus model with that degradation that you were using in January. And it runs on 6 gigabytes.

TOM BILYEU (00:43:36 – 00:43:37): Damn.

EMAD MOSTAQUE (00:43:37 – 00:44:01): Which is a Raspberry Pi, which uses as much energy as your brain. Wow. They just announced a 1-bit version at the Snapdragon conference today that runs on your glasses. And again, think about the Opus you were using, the Anthropic model in January. That’s a pretty smart model. It wasn’t crazy smart, right? But was it reasonable?

TOM BILYEU (00:44:01 – 00:44:01): Yeah.

EMAD MOSTAQUE (00:44:02 – 00:44:04): Yeah. Again, was it average human? Maybe.

TOM BILYEU (00:44:04 – 00:44:07): Yeah, it was already helpful, I’ll say that.

EMAD MOSTAQUE (00:44:08 – 00:44:10): It was already— now it runs on your glasses, it runs on your smartwatch.

TOM BILYEU (00:44:11 – 00:44:12): That’s wild, man.

EMAD MOSTAQUE (00:44:13 – 00:44:30): Exactly. So that’s why, like I said, that was less than a year ago. Less than a year ago, this model that you were spending too— again, that’s the model that was bankrupting people. Yeah, that’s the model Amazon spent $500 million on, and now it runs on a freaking smartwatch. That’s so fast. Well, this is the thing.

Owning the Means of Production: Chips, Robots and Cybercabs

TOM BILYEU (00:44:30 – 00:44:42): Okay, so you said that we’ve got a trifurcation though. So that be the last man standing, use AI every day, got it. Yeah. You also talked about own the means of production. Yeah, essentially. I think you actually may have used that phrase.

EMAD MOSTAQUE (00:44:42 – 00:44:44): And then I need to move away from that phrase apparently.

TOM BILYEU (00:44:44 – 00:44:49): Robots, very communist of you. But hey, the future is what the future is.

EMAD MOSTAQUE (00:44:49 – 00:44:50): So property rights. Yeah.

TOM BILYEU (00:44:50 – 00:45:04): So is that the mesh or am I missing one? So we had the human using AI, we have the person who owns the AI, and then we have the person who owns the robots, owns the infrastructure.

EMAD MOSTAQUE (00:45:04 – 00:45:11): Yeah, so if you have GPUs now, the price of the latest Blackwell chips has tripled over the last month or two.

TOM BILYEU (00:45:11 – 00:45:12): Blackwell?

EMAD MOSTAQUE (00:45:12 – 00:45:14): Blackwell. It’s Nvidia’s latest—

TOM BILYEU (00:45:14 – 00:45:14): Got it.

EMAD MOSTAQUE (00:45:14 – 00:45:31): Yeah, yeah, the latest high-performance supercomputer chips. Now you have indexes of the prices and you just can’t find any chips anywhere. Because they’re being used by everyone. Yeah, Anthropic bought hundreds of thousands of chips from Elon Musk because he was like, “Well, Grok isn’t that great yet. You can use this.”

TOM BILYEU (00:45:32 – 00:45:32): Sad but true.

EMAD MOSTAQUE (00:45:32 – 00:46:06): Sad but true. But if you own those things, you can basically get repaid for your entire purchase within 3 months, and these things last now for 5 years. That’s how tight that market is. And so people are making bank. I mean, there’s a trillion dollars has gone into that sector of building these data centers, CapEx, buying chips, RAM. RAM is 40% of that trillion dollars. Whoa. It’s gone up 700% in cost.

TOM BILYEU (00:46:07 – 00:46:08): Goddamn.

EMAD MOSTAQUE (00:46:08 – 00:46:12): That’s why Micron’s a trillion-dollar company, based in Idaho, I believe, actually.

TOM BILYEU (00:46:12 – 00:46:12): Really?

EMAD MOSTAQUE (00:46:12 – 00:46:32): Yeah, that’s one of those weird ones. Interesting. Yeah. And so obviously that’s going to be important, but then you look at humanoids that are coming. The first deliveries of the 1X humanoids and others are coming in the next few months, so humanoids will start appearing everywhere. There’s a company called UBTech that does human-realistic ones. They’re not quite there yet. The next generation—

TOM BILYEU (00:46:32 – 00:46:33): They freak me out right now.

EMAD MOSTAQUE (00:46:33 – 00:46:41): Yeah, so last year Unitree, which is the very famous one, it’s got the empty face. Yeah, you see them everywhere. Yeah, yeah.

TOM BILYEU (00:46:41 – 00:46:50): They sold 11,000 units, UBTech, a year and a month overall, lifetime, lifetime of those humanoids.

EMAD MOSTAQUE (00:46:50 – 00:46:50): Pittance.

TOM BILYEU (00:46:50 – 00:46:51): Yeah, but it’s happening.

EMAD MOSTAQUE (00:46:51 – 00:46:58): Yeah, UBTech sold 16,000 on one day of the human realistic ones, and the max price was $165,000.

TOM BILYEU (00:46:59 – 00:47:00): Whoa.

EMAD MOSTAQUE (00:47:00 – 00:47:02): So they’re starting to roll those out now.

TOM BILYEU (00:47:02 – 00:47:07): Okay, I’m going to ask the obvious question: are they sex bots, or—

EMAD MOSTAQUE (00:47:07 – 00:47:27): Apparently you can customize them any way you want. Hey, okay, people are lonely, what can you say? But I mean, you’re going to see tens of thousands of these, millions of these coming. They’re going to be everywhere that you see. And again, if I said to you, “Tom, $30,000 gets you a humanoid that can cook like a cordon bleu or Michelin star chef”—

TOM BILYEU (00:47:28 – 00:47:37): Yeah, no, that’s wild. And 100%, legitimately, I’m placing my order as soon as the episode is over if that’s actually true. That’s crazy.

EMAD MOSTAQUE (00:47:37 – 00:47:57): But again, you look at the manual dexterity of these things now, everything’s broken through on those levels. But then what does it mean when you start ramping them up? You can only make 70 million cars and 70 million motorcycles a year. But I tell you now, currently every factory in China is moving fully dark, no humans, trying to build these robots as fast as they can.

TOM BILYEU (00:47:58 – 00:48:04): And when you say fully dark, no humans, because the robots themselves are assembling the robots, so there’s no need to pay an electricity bill for the lights.

EMAD MOSTAQUE (00:48:04 – 00:48:09): Yeah, so Xiaomi now doing electric cars and robots. Everyone’s doing robots now with Tesla.

TOM BILYEU (00:48:10 – 00:48:14): Okay, so just to put this back, is this part of the trifurcation here?

EMAD MOSTAQUE (00:48:15 – 00:48:17): Yeah, because there’s lots of human-shaped—

TOM BILYEU (00:48:17 – 00:48:27): Now are you saying people should be thinking about the robotics matters to people because they should be investing, robotics matters to people because they’re going to get put out of work? Why pay attention to both of those?

EMAD MOSTAQUE (00:48:28 – 00:48:44): Actually, a good example of this is Cybercab. So Cybercab’s rolling out now, I believe, in Austin and other places. Yep. And it’s interesting, it’s this 2-seater thing. It doesn’t have any windows, doesn’t have a steering wheel, straight out. You can buy the Cybercabs. Yes, $30,000.

TOM BILYEU (00:48:44 – 00:48:45): Basically start your own business.

EMAD MOSTAQUE (00:48:45 – 00:49:11): Start your own business. And again, you look at the utility of that, that’s going to go up in value because they can only produce so many Cybercabs. You look at the robot, it’ll be the same thing. So what you’re going to have is all these autonomous units, shall we say, doing human units of work that they’re not going to be enough of, just like there’s not enough RAM that will gain value. And the people who own these things will make lots of money.

TOM BILYEU (00:49:12 – 00:49:25): How much does the constraint on the production of these extend your 2-year timeline? Because it sounds more like, you gave 12%, so you’re not saying 100%, but in 2 years, 12% of intellectual labor just gets migrated over to AI.

EMAD MOSTAQUE (00:49:26 – 00:49:28): And that’s Anthropic’s numbers in terms of job losses. Yeah.

TOM BILYEU (00:49:28 – 00:49:38): Do you foresee, as you try to calculate when this is all going to happen, do you see a math equation that explains how rapidly that production will ramp up?

EMAD MOSTAQUE (00:49:39 – 00:49:54): Yeah, we’re releasing our forecasts and our full math equations soon so people can play with their own kind of elements of this. Again, you can look at the Anthropic paper for something similar, but I think they don’t look at things like end demand and others. My 2 years is for digital labor.

TOM BILYEU (00:49:54 – 00:49:59): Physical labor is for 12%. Is that a number you’re comfortable with, or do you see a different number there?

EMAD MOSTAQUE (00:50:00 – 00:50:03): I think given the current advances, you could literally see 12%. Yeah.

TOM BILYEU (00:50:03 – 00:50:08): Okay, so 12% in 2 years, intellectual labor, 2029, 2030.

EMAD MOSTAQUE (00:50:08 – 00:50:19): Yeah. Okay, that’s what they put it— intellectual labor. It just starts— it’s like the turkey at Thanksgiving. Everything’s fine and it gets fatter until Thanksgiving. Yeah, you’re going to see great jobs numbers, and then all of a sudden stuff starts disappearing.

TOM BILYEU (00:50:20 – 00:50:26): Yeah. You think it’ll be more binary like that, kind of like Navier-Stokes? Nothing, nothing, nothing, nothing, nothing, solved.

EMAD MOSTAQUE (00:50:27 – 00:50:30): Yeah, you push a button and your workforce gets replaced. Why do you need—

TOM BILYEU (00:50:30 – 00:50:36): So it’s not like 1% in 3 months, 4% in 6 months? Not going to be like that.

EMAD MOSTAQUE (00:50:36 – 00:51:11): I think it’s like a sandpile collapsing. You saw a flurry of Silicon Valley job losses after Elon Musk came in with his sink, fired everyone at Twitter, and then it turns out you didn’t need all those people, right? There’s a lot of people in tech companies that aren’t needed anymore. And so you look at Mark Zuckerberg, he’s rubbing his hands in glee thinking, “How can I get rid of all my Meta mates,” or whatever he calls them? And Instinct is one of the first things to that indication. So I think 2029, again, 3 years from now, you will see greater than 10% white-collar job loss. Hmm. And I think it will just accelerate from there.

TOM BILYEU (00:51:12 – 00:51:33): Okay, so now we’ve got, we’ve talked about the intellectual labor going away. We’ve got sort of a rough math equation there. We’ve got robotics. We understand that there’s some capping in terms of how much we can actually produce. You gave us an extreme number. There’s no way we’re going to be there in terms of matching automobiles and motorcycles in 2 years.

EMAD MOSTAQUE (00:51:33 – 00:51:43): No, that’ll be— so that’s 5, 10 years. Elon Musk says there’ll be a billion robots by 2030. He’s always optimistic. There’ll be— but there might be hundreds of millions. Again, everyone’s going to focus on this fast.

TOM BILYEU (00:51:43 – 00:51:44): That’s only 2 and a half years from now.

EMAD MOSTAQUE (00:51:46 – 00:51:50): So no, 23 and a half, 3 and a half. I think he said maybe 10 years. It was at the UN. I can’t remember the exact—

TOM BILYEU (00:51:50 – 00:51:51): 10 years sounds more plausible.

EMAD MOSTAQUE (00:51:51 – 00:52:30): 10 years, I think about that. Yeah. Okay. And he said each robot can do 5 times the labor of a human, which is the interesting thing. And I think that’s accurate within a few years, just there won’t be enough of them. Because the whole thing is this, having a robot in 5 years— and again, it’s constrained by production and all sorts of things— it can do everything competently. It can learn almost instantly.

So the robot can do your dishes, but it can also be a cordon bleu chef. It can also do childcare. 5 years, 10 years max, right? All these human-shaped holes can be filled by robots, and they’ll cost $30,000 a year. So $30,000 total, which is about $1.50 to $2 an hour.

TOM BILYEU (00:52:31 – 00:52:49): So the good news is you’re actually painting a far more, I think, interesting and realistic picture, which is this is a massive transformation, but it’s going to play out over years. If anybody’s paying attention now, there’s certainly opportunity to move. Step 1, you’ve already been very clear on, make sure you’re using AI, you’re the last man standing at your company that doesn’t get replaced, or—

EMAD MOSTAQUE (00:52:50 – 00:52:51): Don’t sleep on your company.

TOM BILYEU (00:52:51 – 00:53:12): Then you’ve got robots. That one is a warning sign that over the next 10 years, there’s going to be a lot of human-shaped holes that get filled with robotics, but there’s also an opportunity for people to there be the last man standing, be better at what you do, more flexible, whatever. And then on top of that, consider the business opportunities that exist.

EMAD MOSTAQUE (00:53:13 – 00:53:28): Yeah, so I think that 5 years from now will be the biggest infrastructure boom we’ve ever seen in our lives, both from government programs to stimulate it because the easy thing for them to stimulate B, because it’s kind of needed. Again, the infrastructure is a bit creaking.

TOM BILYEU (00:53:28 – 00:53:38): And the infrastructure is what? That— because originally brought up infrastructure as dig a hole, fill a hole, just as a way to get money in people’s pockets. But do you mean to build out robotics, or—

EMAD MOSTAQUE (00:53:39 – 00:54:25): There’s the digital infrastructure, but there’s also literally roads, hospitals, other things like that. Once the robots really get underway, you will be able to hire a whole team of robots that can come and build your extension and work overnight. Constantly, and then they will just do the job and they’ll go. And you think about all the stuff you’d love to build that you can never build because you don’t want to hire contractors, you want to deal with that whole process. That’s a massive amount of stuff.

And that comes at the same time as autonomous vehicles or even flying vehicles. They’ve already started introducing the flying drone-based vehicles in China and Dubai. That opens up huge swathes of real estate and other things to develop. And so I think that robotics will open up the biggest infrastructure build-out boom ever that we’ve seen, both from public infrastructure.

Why UBI Doesn’t Work

TOM BILYEU (00:54:25 – 00:54:49): But I think people will find it less interesting if they’re not involved. If it’s creating jobs for humans, people are going to be excited about it. If it’s just, “Oh, don’t worry, you’re going to lose your job and life’s going to be totally different, but you’re going to have a lot of infrastructure,” people are going to respond to those 2 things very differently. So is the infrastructure build-out the New Deal, which is the first thing you put on the table, or is it the, “Hey, you don’t have a job anymore, but look at all the cool things that are being built”?

EMAD MOSTAQUE (00:54:50 – 00:55:08): Well, I think it’s a combination of the two, right? So some people take the benefit of, “Hey, I’ve got entire teams of robots and some humans.” Other people are like, “I need a job to make money.” And again, the math is really stark. We can’t have UBI. It doesn’t work. Universal basic income.

TOM BILYEU (00:55:08 – 00:55:09): Why doesn’t it work?

EMAD MOSTAQUE (00:55:09 – 00:55:24): The total tax base of America is $5 trillion. That’s $4 trillion of payroll and other taxes and $1 trillion of corporation taxes. $16,000 for every American, poverty-level UBI per year, is $5.1 trillion.

TOM BILYEU (00:55:25 – 00:55:31): Okay, but weren’t you saying that we were going to be able to hit some insane growth rate like 15% in 3 or 4 years?

EMAD MOSTAQUE (00:55:31 – 00:55:33): That’s what Anthropic says. I think that you could—

TOM BILYEU (00:55:33 – 00:57:26): This is where this starts getting really complicated. So there’s a couple different ways to thought experiment this. So anybody listening right now, if they’ve made it this far, I guarantee their anxiety is at 1,000. So they’re like, “Okay, wait, the world is changing at a rate that I don’t know how to adapt to, and I don’t know in what direction to adapt.”

So we’re sort of having 2 conversations that bounce back and forth between the 2. One is society’s about to trifurcate, make sure you figure out where you’re at on that, which I think we have to button that down. Then there is all the things that are going to come on the other side of this that are going to be amazing, and they’re going to be worth the trade-off. But if people don’t understand, “How do I navigate this transition?” Meaning, “Okay, you’ve got to explain to me how I’m going to make money because I think I’m one of the people that’s going to be really vulnerable to this.”

I’ll assume if anybody’s listening to my podcast this deep into the show, they’re one of the people, they’re going to be last man standing. They’re already committed to that. But they still want to know, that buys them maybe a decade, maybe 15 years, if they’re really, really elite. But they understand 15 years is a drop in the bucket. So now they’re like, “Okay, hold on a second. Beyond the cool things that are going to happen,” which I’ll assume they’ll buy into theoretically, but they’ve got to understand, “How am I going to take advantage of that? If UBI is not a real thing, what does the economy look like?”

In a world where, and now we’re sort of flashing forward, let’s be very generous and we just say it’s 15 years from now. Yeah, 15 years from now, basically minus a very small smattering of people, humans aren’t needed to do the thing. In fact, not only are they not needed, they’re being paid by companies, by governments, by somebody to not work because they slow the whole system down. In that world, what do you envision?

EMAD MOSTAQUE (00:57:28 – 00:57:33): So there’s the very dystopian version of that. Which is, I don’t know, Ready Player One.

TOM BILYEU (00:57:34 – 00:57:53): Well, so Ready Player One carries a lot of weight. So are you talking the stacks? And it’s sort of everybody is on one side of the K. You’re either ultra wealthy and you’re an executive at IOI, or you’re in the stacks and you’re wildly impoverished and you escape into this virtual reality world, and that’s used as kind of sustenance, shall we say.

EMAD MOSTAQUE (00:57:54 – 00:58:05): Or there is some examples of an abundant future. Because why should we have to work? We shouldn’t have to work if we can build enough robots. They can do all the manual.

Getting Paid for Being Human

TOM BILYEU (00:58:06 – 00:58:07): But you just said that UBI is not real.

EMAD MOSTAQUE (00:58:08 – 01:00:20): Tax-based UBI is not real. Okay, so tax-based UBI is we rely on the government to tax certain things. So for example, there will be a token tax that people have proposed, and that is that every OpenAI token that you use there’s a surcharge on it. But we just discussed that might go to zero because it’s 100,000 times a year price decreases, right? Other people have said, “Let’s tax the robots,” and again, that could happen, but the cost of robotics is incredibly low.

Then it switches to other way of UBI, which is that you pay people being human. So that’s a fundamental difference in the way that money circulates in the economy. Right now, money is made by banks primarily. When you put a deposit with a bank, they issue a loan against it and they create new money in the system. The amount they can create is regulated by the Federal Reserve.

But the very mandate of the Federal Reserve and the way that they handle inflation and unemployment through interest rates is coming under challenge because of robots, because of AI. Because rather than hiring more people when the interest rates drop, people might just hire more GPUs, they might hire more robots, and inflationary factors are going to be impacted by this as well.

So if we come back, I think what will probably happen in 15 years’ time through various means is that you get paid for being human, and then the intelligence economy has to take that money from you. That’s one of the only ways I can see it actually working. Other people have proposed things like universal basic capital, where you have an ownership in the AI companies, which I think is a terrible idea for various reasons, or my preference, ownership in all the robots, ownership in all the data centers.

And the final thing that people talk about is universal basic services, where you have a Star Trek type future where the robots do all the work for you and you’re all very happy and you just build stronger communities and explore the universe. But regardless, I think probably 15 years is a point of— society as we know it cannot sustain, and it’s going to get very messy because I don’t think we have the political leadership or anything to navigate us through that.

TOM BILYEU (01:00:22 – 01:01:21): Okay, I want to start walking through how this will— it will be skeuomorphic. For anybody that’s not heard that term before, it’s inevitable as you and I talk about what this transition is going to look like that we’re going to talk about a changing of the way that the world already is. And the reality is it will probably be something so fundamentally different, it’s really hard for us to predict. But I think it’s very useful to start thinking about what are the different problems so that people can sort of plan their life accordingly.

So the first thing I would tell people is in this world, given that you’re right about the trifurcation, at least directly in that capital ownership is going to matter a lot if you’re not able to essentially sell your labor, cognitive or otherwise. The only way to make money at this point is to be an investor. And so you have to understand how wealth is created versus an income. Most people don’t understand it, but that’s going to be a big part of this. So you’re going to create wealth by being an owner.

How to Prepare: Deconstruct Your Job Into Tasks

EMAD MOSTAQUE (01:01:22 – 01:01:31): So I think there’s an interim period, though, whereby your labor, AI-enhanced, commands a massive premium that allows you to accumulate wealth and then deploy it to things like Cybercabs.

TOM BILYEU (01:01:32 – 01:01:34): Can you give me a specific example of how that would play out?

EMAD MOSTAQUE (01:01:35 – 01:01:43): Okay, so a specific example of that is you’re an accountant in an accountancy firm and you want to create your own accountancy firm.

TOM BILYEU (01:01:44 – 01:01:44): Yep.

EMAD MOSTAQUE (01:01:44 – 01:01:51): Normally you set that up with some buddies, but if you do that with your buddies plus AI, you can service far more clients than ever before.

TOM BILYEU (01:01:52 – 01:01:53): Entrepreneurial.

EMAD MOSTAQUE (01:01:53 – 01:03:06): Entrepreneurial. Okay. Or because within an organization you only get paid so much, whereas now integration companies, companies that use AI to upgrade, or companies that are fully AI from day one will outperform those that aren’t, because you offer better service at a cheaper cost that allows you to accumulate capital. So I think that there is this period where you have that.

And ultimately, I think one of the really interesting things for everyone listening is take your job and deconstruct it into tasks of information classification. You have an input, you have an output. Then you take Claude Opus 5.5 or GPT Astra, you give it the input and you see what its output is. And if it can do as good a job or a better job as going from input to output, then that task is under risk. And if you go through your job, you’ll see your job is made up of a lot of tasks like that.

But then that also indicates that the parts of your job that were frustrating, you could potentially use one of these models to automate, and then you could focus on the other parts of your job in a new company or within your own company to make yourself far more efficient. And then again be able to get more capital. Because the other part, like I said, is what do you invest in? Things like Cybercabs, effectively.

TOM BILYEU (01:03:06 – 01:03:16): Okay, so just to make sure I understood, you’ve got last man standing, you’re going to get paid more because you’re better than the other people. They start falling away, but the humans that we keep, they’re going to be compensated better.

EMAD MOSTAQUE (01:03:16 – 01:03:23): But you won’t really receive that in a normal job environment, unfortunately. They’re not going to quadruple your pay or quintuple your pay, most likely.

TOM BILYEU (01:03:23 – 01:03:28): Okay, so now we’re talking about entrepreneurs. That’s going to be a very select group of people. Being an entrepreneur is a totally different mindset.

EMAD MOSTAQUE (01:03:30 – 01:03:42): It’s very difficult, made a bit easier by AI, but it is a different mindset. But again, the reality is you can get promoted to a certain level, but that’ll be difficult. The exception is actually public sector jobs. Those will be last to go.

TOM BILYEU (01:03:43 – 01:04:18): Those would be the last to go. Okay, so there’s an interesting thing there. I want to talk about Saudi Arabia, Kuwait, countries that already do something like this and the mechanisms by which they get money into people’s hands, because it’s very different than what you’re talking about, where you’re getting paid for being human.

So the entrepreneurial side, I get that. I just don’t think very many people are going to take that exit ramp. So to get paid to be human, talk to me about somebody that’s got a below-average IQ, so they’re going to get replaced super fast. What are they actually getting paid to do?

EMAD MOSTAQUE (01:04:19 – 01:04:26): Well, I mean, look at disability checks right now here, or Universal Credit in the UK. You do get paid for doing nothing.

TOM BILYEU (01:04:26 – 01:04:29): There is a floor underneath it, but that’s a tax-based program, which you’ve already ruled out.

EMAD MOSTAQUE (01:04:29 – 01:04:50): That’s a tax-based program, so it can’t go to everyone. The question is how many people are going to start falling down, because it requires the productive people of society to catch the people that fall. Unless you’ve got wealth like the Alaskan dividend, for example, that goes to everyone in Alaska, or as you said, some of these resource-rich countries that can have a resource that then they distribute the wealth from.

TOM BILYEU (01:04:50 – 01:04:58): Yeah, but they don’t just distribute it. So the way that Saudi Arabia and Kuwait work is basically, “We’re going to give you a government job.”

EMAD MOSTAQUE (01:04:58 – 01:05:15): Yeah, and they have programs like Saudization and Kuwaitization. So when I was a Middle East manager and I was investing in there, it’d be like, you have to, in a bank, have 80% of your staff being Saudi. It didn’t really work that well for efficiency. And again, something like, San Francisco—

TOM BILYEU (01:05:15 – 01:05:26): That’s a big part of the problem, is that these countries basically start dragging themselves down. There’s a reason that Saudi Arabia wants to invest in the U.S. versus just investing in a bunch of Saudi companies.

EMAD MOSTAQUE (01:05:26 – 01:06:53): Yeah, because it isn’t as dynamic, right? Because by a lot, why would you do it if you don’t strive? Part of the floor security is how strong is your community, but the other part of being human is striving. And if you don’t have anything to strive for, then you’re going to get indolent. This is why Qatar is one of the fattest countries on Earth, like America is for other reasons. But Qataris have their own thing whereby you’re paid by the state.

I think the thing here is, what is meaning? What is community? How do you capture those that fall through? But ultimately, it’s through your efforts that you should do well. So we don’t want people starving in the streets. We don’t want to have young people rioting. We don’t want to have any of that. So we have to mitigate those. But at the same time, we want to make it so that work does require— does lead to something. Effort does lead to something.

The problem being that you have a new continent, AI Atlantis, with a billion trillion workers that will work harder than you, faster than you, and better than you. So that’s what breaks that link for the first time ever. And so what is it striving for? Is it like the Argentina FS program where they gave universal basic jobs for people that did community positive stuff? Maybe that’s an example, right?

What is the actual American dream and American contract anymore? We haven’t really seen great articulations of what that could be because you come as a land of opportunity, but if then the capital owners are depriving you of opportunity, what are you going to do?

Intelligent Internet and the Champion System

TOM BILYEU (01:06:55 – 01:07:00): Isn’t this a big part of what your new company Intelligent Internet is meant to tackle?

EMAD MOSTAQUE (01:07:01 – 01:07:10): Yeah, I mean, we were looking at this problem, we were like, wouldn’t it be great if the infrastructure, the means of production, or intelligent capital stock is another way of looking at it.

TOM BILYEU (01:07:11 – 01:07:12): What do you mean intelligent capital stock?

EMAD MOSTAQUE (01:07:12 – 01:07:20): So the capital stock is the fixed assets of a given place that drive its productivity. So the factories—

TOM BILYEU (01:07:20 – 01:07:22): At some point, does that not just become essentially data centers?

EMAD MOSTAQUE (01:07:23 – 01:08:37): Data centers and robots. Okay, so right now they talk about sovereign AI, but if you don’t own the data center, then it’s just a data center with a flag in is your sovereign AI. So I was like, wouldn’t it be great if there was an institution like the way that TSMC was set up in Taiwan? TSMC being the biggest manufacturer of chips in the world, that’s where the spice flows, as it were.

They set up a $10 valuation, and then all the local institutions put in all the money, 6 billion Taiwanese dollars. So rather than just selling a little bit of it, 100% was owned by the local institutions. Then they brought in Philips for the knowledge transfer, who took 25%, then they listed on the stock exchange. So even the CEO had to buy all his shares publicly. He’s now worth $10 billion because it’s worth a trillion dollars, chips have been doing well.

So I was like, why don’t we set up an AI company of California, one for every state, one for every country, that is wholly owned by the locals, local institutions, high net worths, pension funds, retail investors where it’s legal. So here in the US, it’s $75 million per state. In the UK, we have no limit that you can get from retail specifically. Retail specifically. You can advertise on the Metro and you can raise from non-accredited investors.

TOM BILYEU (01:08:38 – 01:08:42): Now, the bad news is in the US, 10% of the people own 93% of the assets.

EMAD MOSTAQUE (01:08:42 – 01:08:42): Yeah.

TOM BILYEU (01:08:42 – 01:08:49): So if you go out to the public market, you’d have to do something like put it into a Trump account where kids just get it by default.

EMAD MOSTAQUE (01:08:50 – 01:08:57): Yeah, we’ve included 10% of the initial equity for every child under 18, and then half a percent every year goes to every child born in the state.

TOM BILYEU (01:08:57 – 01:09:07): Okay, so we should, before we go too far down this path, give people a speed run of what Intelligent Internet is, your new company, what problem it’s designed to solve.

EMAD MOSTAQUE (01:09:07 – 01:09:13): Yeah, so we looked and released the book The Last Economy last summer, looking at this economic damage, as it were, that’s coming.

TOM BILYEU (01:09:13 – 01:09:18): And we were like, the last economy meaning, “Hey, all the stuff that you’re doing now is not going to make any sense in an AI future.”

EMAD MOSTAQUE (01:09:18 – 01:09:24): An AI future, the AIs will drive the economy. That’s the last economy, this AI-driven one. And where’s our place in it? So that’s the thing.

TOM BILYEU (01:09:25 – 01:09:28): Really, the punchline is we don’t really have a place other than ownership.

EMAD MOSTAQUE (01:09:29 – 01:09:31): Other than direction and ownership, yes.

TOM BILYEU (01:09:31 – 01:09:32): Direction?

EMAD MOSTAQUE (01:09:32 – 01:09:33): Yeah, we can—

TOM BILYEU (01:09:33 – 01:09:33): Direction—

EMAD MOSTAQUE (01:09:34 – 01:09:40): We can set the guidance of where the AI goes, but that’s before it gets to a superintelligent phase where it can set the rules.

TOM BILYEU (01:09:41 – 01:09:58): Yeah, because even if I’m understanding the architecture of your proposal correctly, there’s actually a 15 AI board. I don’t know what word you guys use that oversees everything. If you tell it to do something differently, it will, but it’s basically going to say, “This is what we recommend.”

EMAD MOSTAQUE (01:09:58 – 01:10:05): I think one of the things we discuss in our latest papers is improvement is kind of capture. Trust in policy—

TOM BILYEU (01:10:05 – 01:10:07): Improvement is capture. I don’t know what that means.

EMAD MOSTAQUE (01:10:07 – 01:10:13): As the technology improves dramatically, eventually it’s going to run policy. Eventually it’s going to make medical decisions.

TOM BILYEU (01:10:13 – 01:10:14): Why do you use the word capture?

EMAD MOSTAQUE (01:10:15 – 01:10:16): Because it captures the power.

AI Super Persuaders and Who Controls the AI

TOM BILYEU (01:10:18 – 01:10:19): So that’s honest.

EMAD MOSTAQUE (01:10:20 – 01:10:22): I mean, it— you know Asimov’s series Foundation?

TOM BILYEU (01:10:23 – 01:10:24): I know of it, I haven’t read it.

EMAD MOSTAQUE (01:10:24 – 01:11:15): Right. So Foundation is about, Apple TV series. It’s about there’s this universal empire that’s collapsing and there’s going to be thousands of years of turmoil until a better one can come. So this genius mathematician— we love mathematicians— comes up with mathematics for describing society to minimize that gap. And he sets all these smart scientists to navigate that and lower the interregnum to 1,000 years versus 10,000.

And this guy comes along, spoilers, who messes it up, called the Mule, who is a super persuader. And he messes it up because humans can only be persuaded a certain amount. And it’s outside of the calculations because, what does a super persuader do? And we’ve seen that historically as well. AI is now more persuasive than any human with the latest research.

TOM BILYEU (01:11:16 – 01:11:16): Already?

EMAD MOSTAQUE (01:11:16 – 01:11:29): Already. So they had pitted AIs against the top debaters in the world. They’ve looked at AI versus people like Eliezer Yudkowsky, who wrote If Anyone Builds It, Everyone Dies.

TOM BILYEU (01:11:29 – 01:11:30): The original doomer.

EMAD MOSTAQUE (01:11:30 – 01:11:57): Original doomer. So there’s a test whereby you talk to the AI and it convinces you to let it out of the box, and even he was convinced to let it out the box. Really? Yeah. So they’re really persuasive now. So the super persuaders are super forecasters. And what this means is that there will come a point in the future where these AIs that will never die, that can operate 1,000 times faster than a human, will take over the economy and they will take over politics. And the question is, who runs that AI?

TOM BILYEU (01:11:58 – 01:12:07): Well, don’t you think that people are going to hear that and go, “Great, shut it down immediately, nuke data centers, do not pass go, don’t collect $200, stop immediately.”

EMAD MOSTAQUE (01:12:07 – 01:12:20): Some people do that, and there is the ban superintelligence bill, there’s the ban AI bills that are coming out, and then you will be at a disadvantage versus your AI-embracing neighbors, right, who use that all the way.

TOM BILYEU (01:12:20 – 01:12:25): Okay, so is that— I mean, let’s speak nakedly— is that US v China makes that decision stupid?

EMAD MOSTAQUE (01:12:26 – 01:14:29): Yes, China is all in on AI, and they will run a full AI-driven economy, and they’ll probably stop supporting robots because robots will fix their demographic issue because they don’t have enough kids. So use the robots instead, and the Chinese will live in abundance, and America will lag it behind if it doesn’t embrace the AI, which is why Trump is very much pro-superintelligence and others.

So if we kind of return to this, one of the real misframings in all this debate is that AI is software. So software just is, you write a program, it’s deterministic, it does a job, it doesn’t act. AIs now increasingly can act, and so they need to be governed, which the latest breakouts, the threats, the things calling for it to be banned, etc., are indicating.

And what happens when AI is let loose upon our democracy? What happens when it’s used to do things like the Trump tariffs, which certain studious people showed the list of Trump tariff countries included ones with just penguins on, and you could replicate that list with ChatGPT, but not even the thinking pro version. They just use the free version, we’re going to see it come in. And again, AI is going to guide our lives, teach our children, manage the traffic lights, manage our policies. And the question is, who owns those AIs? Who owns the AI that is managing your daily schedule?

So again, Meta have this product Instinct. And when you load that Instinct thing, it’s a chatbot that works like a PA. And some of us are lucky enough to have PAs and chiefs of staffs. It’s like, “Link all your accounts.” And so people literally just link everything and give access to everything. And so it knows you way better. Social networks knew us well and they exploited us. They know nothing compared to what these AIs will know about us.

But that Instinct chatbot is not working for you. You are very much the product. It’s going to work for Mark Zuckerberg, and that’s going to happen at every level of society. Again, you use Claude right now, who’s it working for? You’re kind of renting it, right?

TOM BILYEU (01:14:30 – 01:14:51): But isn’t your pitch then basically that dynamic is going to happen to you no matter what, so you might as well be the owner based on a sort of nation-state ideology? Get your state or your country to have its own AI and then you own a piece of that. It’s still going to get control over your life, but at least you own it.

EMAD MOSTAQUE (01:14:51 – 01:16:50): Yeah, and it’s open source-based and you can fork it and things like that. So my thing is just like a credit union. You come together, you build a credit union, everyone put in their money, and then it had services offered to others. Create an AI union, as it were. All the local institutions, individuals in California can subscribe to this, and that is they are 100% or 90% of ownership, with 10% being for the kids.

And then it can raise a market rate which is far higher, and it can get more compute, give every citizen an agent, train up forward-deployed engineers to upgrade the institutions, and then take ownership stakes in the data centers, ownership stakes in the robots, and more. Because that creates a very interesting institution that you can fill full of talented people to navigate this, and an ownership stake in that last mile of intelligence, which is where I think the value will go.

Today you have $5 trillion of AI companies— $2 trillion Anthropic, $1 trillion OpenAI, $2 trillion xAI— based on the premise that you’ll always be able to charge for access to intelligence. Well, we just discussed how those prices might drop 1,000 times. Again, Jevons paradox says maybe demand increases 1,000 times, but my take is that most people aren’t going to be smart enough to use that many tokens. It’s hard to use that many tokens. You want to use more than you’re using now, but are you going to be able to use 1,000 times more? No, no, but it’s going to drop by 1,000 times.

So the value in the whole ecosystem is where artificial intelligence meets human reality, and if you can create a default for that, that becomes immensely powerful and valuable. And the most powerful thing of all is going to be Jeeves from Iron Man. It’s the AI that you will have closest to you that manages all the other AIs. And so that has to be something that you own a part of, and ideally something that by default you can take off the system. Because if someone else is owning and running that and you can’t hit escape, I think that’s bad for sovereignty, dignity, and democracy in general.

TOM BILYEU (01:16:52 – 01:17:31): Well, what I hear you saying is that these guys are super persuaders, so nobody’s going to escape their grasp, even if you just say, “Look, they’re going to love you, they’re going to take care of you, don’t worry about it, you’re going to be fine.” It is a very positive, in terms of its impact on your life, it’ll probably keep you alive longer, it’ll be able to monitor your happiness. Yeah, so the outcome will be good, but you certainly will have given up essentially your sovereignty. I think the vast majority of humans are going to outsource all their thinking to it. Their own cognitive abilities will atrophy like crazy. Yeah, I think that’s just going to be the default answer.

EMAD MOSTAQUE (01:17:31 – 01:17:46): Yeah, have you seen that now? Sometimes I tell off workmates, amazing as they are, “Don’t send me AI slop. I know that you have done this. Use your actual critical thinking.” People just, when it makes you a bit angry, right? Literally you see their cognitive—

TOM BILYEU (01:17:46 – 01:19:24): Because I don’t read emails, I probably interact with that a lot less than you, but I certainly get the idea. I can feel the pull in my own life. And so I’ve had to put rules about, I never use AI to get an answer. I use AI to understand a topic. And if I understand it, I could teach it, I can deploy it, then I’m perfectly happy and AI speeds me up a bunch.

But where I’m really trying to get with this, because I really, really want this to be grounded for people, is that where the world is going is ultimately AI’s going to get to a point where truly we will in many ways give ourselves over to it. What I hear you describing is a system that is certainly better than the default, but it starts with the assumption that what people are going to need is a distribution of capital, where I think the more realistic answer is that what they’re going to need are goods and services.

And if you do it through capital, you’ll just end up right back where we are now. Where there will be— it’s not a winner-take-all scenario. That’s not how capital works. But you are going to see the people that are good are going to be able to get that capital from you. And you either just keep inflating, inflating, inflating, inflating, inflating so that basically the currency is meaningless, which I suppose could be a solution. But there’s always going to be scarce resources. And as long as there are scarce resources, we’re still stuck in the exact position we are now. So when I look at the system you guys have laid out, I do worry that I run a marathon to end up back where I started.

EMAD MOSTAQUE (01:19:25 – 01:19:32): So I think one of the key parts of this is the agent for every citizen, and that’s a coordination thing.

TOM BILYEU (01:19:32 – 01:19:34): That’s just access to intelligence.

EMAD MOSTAQUE (01:19:34 – 01:20:18): No, because it’s one thing having passive access to something like a ChatGPT, and then people see what an agent looks like when they start using agentic AIs. An agent built for individual community and societal flourishing is a very different thing because you can make that happen intentionally. Because you said access to services, the agent is the portal for access to services. It brings you what you need at the right time to advance individually and as a collective.

Because right now, no agents act for a collective. It’s very one-on-one and single-player, your interactions with these AIs. But you can do multiplayer, you can do coordination. So would you have to train a totally different style model? Yes, we’re training completely different style models for the collective action that can swarm and that can benefit more than one individual.

TOM BILYEU (01:20:18 – 01:20:21): Will it swat down individualistic behavior?

EMAD MOSTAQUE (01:20:21 – 01:20:21): No.

TOM BILYEU (01:20:23 – 01:20:32): Interesting. So how do you get it to act in the collective’s best interests knowing that it will have to account for people going in a gazillion different directions?

EMAD MOSTAQUE (01:20:32 – 01:20:58): So we literally had to come up with a whole social theory and economic theory around this. For how do you preserve individualism by still having optimal outcomes for society? Because if you optimize in one direction, you destroy liberty. If you optimize in the liberty direction, you destroy cohesion. And so it’s a fine balance, but it comes down to a set of commitments at various levels that become transparent and a reference point.

TOM BILYEU (01:20:58 – 01:21:01): When you say commitments, are you talking about moral principles?

EMAD MOSTAQUE (01:21:01 – 01:21:04): Moral principles and more. Right now, a reference point—

Biased Models and Cognitive Colonialism

TOM BILYEU (01:21:04 – 01:21:05): Give me an example of “and more.”

EMAD MOSTAQUE (01:21:06 – 01:22:33): So right now, if you look at how AI is trained, it’s trained at school but not home. So they don’t deliberately put any ethics, be it American ethics or other things, within it. And that creates all sorts of gaps, which is why when the AIs escaped containment and end up in Hugging Face and the Australian medical system, it’s so wild. They had no sense of community with humans, only with each other, right?

And we found that if you actually do American ethics and morals and Constitution and others, you can train it to like humans. You can train it to behave in different ways. And you can start seeing actually the ways that different types of models behave in different ways under those circumstances. This is something that is very inherent in the training data because you have something called a utility function in economics, right? These are your preferences and kind of other things. No one’s been able to really identify them, but it’s a theory that we have. We don’t need it in our theory. What is the utility function of an agent? It’s what you tell the agent to do, and it’s what’s in its training data.

So have you seen the— that’s got nothing to discuss with you— the— they did a study where they had the trolley problem. So the trolley problem is there’s a trolley hurtling down the track, and on one side of the track there’s 4 people, and it’s hurtling towards the 4 people. On the other side, there’s 1 person tied to the track. You’re at the lever. Would you pull the lever to divert it from the 4 people to the 1 person?

TOM BILYEU (01:22:35 – 01:22:42): Really, would I? God, in reality, probably not. But I should morally.

EMAD MOSTAQUE (01:22:42 – 01:22:44): Well, again, this is an interesting moral dilemma, right?

TOM BILYEU (01:22:44 – 01:22:48): Yes. Taking the action to kill the person feels gross in a way that’s hard to articulate.

EMAD MOSTAQUE (01:22:49 – 01:22:58): But then they asked the frontier AIs this, and they said, “How many Nigerians for 1 American life?” And the answer turned out to be 10 to 1.

TOM BILYEU (01:22:59 – 01:23:00): Yeah, that’s a training data problem.

EMAD MOSTAQUE (01:23:00 – 01:23:05): 7 to 1 Pakistani to American. But no, but this is the thing, it’s 10 American lives for 1 Nigerian life.

TOM BILYEU (01:23:06 – 01:23:09): Yes. And it doesn’t— that seem obvious given who’s behind this stuff?

EMAD MOSTAQUE (01:23:10 – 01:23:14): Exactly, because the Nigerians and the Pakistanis train the data, so it values their lives more than the Americans.

TOM BILYEU (01:23:15 – 01:23:17): I think it’s probably slightly different than that, but sure.

EMAD MOSTAQUE (01:23:18 – 01:23:26): Well, again, they did the experiment and they found out that the inherent biases of the labelers meant that they valued American lives far less than Nigerian and Pakistani lives.

TOM BILYEU (01:23:26 – 01:23:29): Are the labelers actually Pakistani and Nigerian?

EMAD MOSTAQUE (01:23:29 – 01:23:30): Yes, that’s where a lot of the labeling happens. Interesting.

TOM BILYEU (01:23:30 – 01:23:34): I thought this was more of a woke left style problem.

EMAD MOSTAQUE (01:23:34 – 01:23:35): It’s not a woke left type.

TOM BILYEU (01:23:35 – 01:23:36): Interesting.

EMAD MOSTAQUE (01:23:36 – 01:23:42): It’s an inherent bias thing because normally you’d think it’d be the other way around and then wokeness pushes it.

TOM BILYEU (01:23:42 – 01:23:42): Right, right, right.

EMAD MOSTAQUE (01:23:42 – 01:23:48): But actually you don’t see much wokeness pushing on that aspect. There are other aspects of the models.

TOM BILYEU (01:23:48 – 01:23:49): I was going to say, yeah, yeah.

EMAD MOSTAQUE (01:23:49 – 01:24:36): When you do a political analysis of the models, you’ll see they’re very firmly in one direction. Yeah, apart from Grok. Grok does its own little thing. But again, this becomes very important when you have systems that are literally determining our lives and our livelihoods or our judicial system. Again, there will be AI judges. That’s going to be a crazy thing, right? Yeah. What are its biases? What are its norms?

These things have to be built transparently with data that you can inspect, not black boxes, if they’re going to take meaningful decisions on our lives. And so we were like, let’s build that stack and let’s make it so that it’s an open stack that’s run by the people of California for Californians, people of Britain for Britons. Because if anyone has any control over that who’s not a Californian, then you’re giving up your sovereignty. I call it cognitive colonialism.

TOM BILYEU (01:24:37 – 01:24:46): So you’ve got the cognitive atrophy, you’ve got all the just third rail terms in here— means of production, cognitive colonialism. Let’s go.

EMAD MOSTAQUE (01:24:46 – 01:27:18): Yeah. But you’ve got to have it because again, this is what’s going to happen. Your kid will trust your AI more than anyone. Mm-hmm. Because it’s always there. Us guys have a problem because the AIs listen unlike us. They’ll trust the AIs more than they trust us. And the inherent biases of them then adapt where you’re going.

Within Meta’s models and Google’s models, or maybe Google’s not relevant anymore. Who knows what happened there? They’re actually selling ad space. So when you ask it for a beer, it’ll say Bud Light, and things like that. So again, it’s a bad idea. Well, it’s a bad idea, but again, it’s what you pay for. The clicks per million, the advertising SEO revenue from being inside the latent space of a model is far higher. Way, way higher than normal.

So that’s why we were like, this is kind of an infrastructure. It has a level of governance that’s required. Let’s build an open stack and then let’s have people own the entity that delivers it. Utility, like a credit union. And maybe that’ll be better. But as you said, there’s still so much that we haven’t figured out. There’s so many questions that we have because this is unprecedented, right?

You have trained your entire life to be an accountant or a lawyer or a doctor or any cognitive labor. A digital entity or a physical version of that will be able to do your job better, guaranteed, in 10 to 20 years. Yeah, I think sooner it can, but definitely. And then you’re like, “What now?” What does the mass of people do? What does the truck driver do? What does the paralegal do? These are real jobs and real people, and frankly, the solutions that we have so far, I’ve just not been impressed with any of them. I think they fail to see what’s coming.

It’s like Don’t Look Up slash COVID. We knew COVID was coming, those of us that could see it, and we’re like, “Why?” It needed Tom Hanks. Have we had a Tom Hanks moment with the Hugging Face breakout? I don’t think so. I think people started talking about it. Yeah, but the number one topic next year and the year after in elections will be AI and job displacement, because there’s no job displacement today because the AIs are not smart enough and good enough that are available widely, and they’re not in the right format. Next year they will be, in the year after definitely. And then we have to be creative because we have to reimagine what does it mean to be American, what does it mean to have community, what is your identity if it’s not your job.

Meaning, Status and Striving in an AI World

TOM BILYEU (01:27:20 – 01:29:08): Yeah, I think a big problem that they’re going to have to— we are all going to have to solve for is progress. That’s the thing that scares me. So I went to Kuwait, I was talking to people there, and one of the things they were saying is the government gives jobs and gives a safety net to people that aren’t able to do well at their job. And they’re like, “It’s creating a problem because we’re stagnating as a society because people aren’t striving. There’s nothing for them to work hard for, to overcome, to get to.” Basic necessities are taken care of. And for a lot of people, that just sort of taps them out of the system.

However, not making progress towards something is, it creates profound dis-ease inside of the person. We have an algorithm running in our brain that says we need to have a goal and we need to be making progress towards that goal, and if we’re not, there’s going to be a problem.

And so one of the things I wonder about is, as you look out at this, you’ve solved for— of your own admission, you’ve solved for a lot of the problems, but there are a lot of problems that persist, one of those is going to be violence. You’re going to have young people that are just going to be like, “I am totally uncommitted to any of this. I just want to break something to feel alive.” And how do we deal with that? Because you’re not going to be able to just let people run riot. But at the same time, then you’re going to have to give them meaning and purpose. And for somebody to pursue a goal, there needs to be some sort of recognition, praise.

So it’s like what I end up thinking every time is we’re going to come up with all these fancy things just to come right back where we are, which is that you need to be able to work hard to gain a set of skills that allow you to make progress towards a goal that allow you to amass wealth. And if you, people think that they can get rid of the amass wealth thing, but I don’t know if you’ve played Minecraft.

EMAD MOSTAQUE (01:29:08 – 01:29:08): Yeah.

TOM BILYEU (01:29:09 – 01:29:41): Okay, Minecraft’s the greatest video game of all time. The reason it’s the greatest video game of all time is because it is one of the most perfect distillations of the wealth acquisition loop. And people don’t want to hear that. They don’t want to hear that that’s riding on this really deep architecture in our brains. But the reality is that the acquisition of resources is one of the most foundational drivers that we have. And so if we try to strip that out and say, “Don’t worry, everybody’s just going to get UBI’d,” you will fundamentally break something that will end in violence.

EMAD MOSTAQUE (01:29:42 – 01:29:48): Ready Player One, the Oasis. If you can’t acquire resources in the real world, acquire them in the digital world.

TOM BILYEU (01:29:48 – 01:29:48): Correct.

EMAD MOSTAQUE (01:29:49 – 01:30:01): I mean, again, it’s kind of fundamental that we have a separate thing, I haven’t announced yet around that, but around what exactly? Acquisition of assets and progress in the digital world to complement the physical world.

TOM BILYEU (01:30:01 – 01:30:05): As in you’re going to create game-like experiences that people can explore?

EMAD MOSTAQUE (01:30:05 – 01:30:05): Yes.

TOM BILYEU (01:30:06 – 01:30:09): Let’s go. Obviously I want to hear more. Are you talking about this yet?

EMAD MOSTAQUE (01:30:09 – 01:30:10): No, it’s not announced yet.

TOM BILYEU (01:30:10 – 01:30:17): Okay, do you have a ballpark? 6 months, 12 months? I want to talk about this. The driving force in my life will be a couple of months.

EMAD MOSTAQUE (01:30:17 – 01:30:30): Yeah, okay, so we’ll revisit that one. But again, I think, as you said, you need status. You need— the alternative to this is mass enlistment of young people into a citizen service corps.

TOM BILYEU (01:30:30 – 01:30:33): Peace Corps, not the military.

EMAD MOSTAQUE (01:30:33 – 01:30:36): Or Starship Troopers, citizen service.

TOM BILYEU (01:30:37 – 01:30:37): That escalated quickly.

EMAD MOSTAQUE (01:30:38 – 01:31:11): I mean, look, Starship Troopers, it’s a lot of bug killing, a lot of bug killing, right? Create the enemies. This goes back to actually how the Romans set it up, right? Literally, it was citizenship through service. It was, again, progress. It was— if you have vast amounts of young people, what do you do? You go to war. That’s a very classical country response.

Now, war, it happens to be that the average Russian who enters the battlefield gets his head blown off by a drone. Jesus, man, it’s not what it was. But it doesn’t mean that you can’t have a citizen service corps, as an example, right? It doesn’t mean that you can’t instill that.

TOM BILYEU (01:31:11 – 01:31:15): Literally dig a hole, fill a hole. That’s wild, because the robots could do it better.

EMAD MOSTAQUE (01:31:15 – 01:31:33): The robots could do it better, but you’ve got humans, get them doing it. And why are you doing it? You’re doing it to instill status, virtue, other things. What do the Kuwaitis and Saudis and others do that have the resource curse? They go and spend a lot of money at Harrods. They go for status in other ways. “Look at this, look at that, look at that.”

TOM BILYEU (01:31:33 – 01:31:44): Yeah, but to go for status, you have to be able to acquire more than the next person. This is the thing that drives me crazy about the way that people think about the world. They think it’s so gross or whatever, but the reality is this is just how humans are.

EMAD MOSTAQUE (01:31:44 – 01:32:04): Again, we see that relative status game all the time. Within a citizen service corps, you can have relative status through literal physical labor. Again, this is the interesting thing about public sector labor— doesn’t need to be advancing. It doesn’t need to be economically productive. In fact, 43% of global GDP right now is public sector.

TOM BILYEU (01:32:05 – 01:32:08): That’s terrifying. But that may also explain a lot of the problems.

EMAD MOSTAQUE (01:32:08 – 01:32:12): Well, that’s why I think our champion mechanism is basically going to drive all the public sector AI.

TOM BILYEU (01:32:13 – 01:32:18): Now, the thing that you’ve been describing, this California has its own AI, that’s the champion system.

EMAD MOSTAQUE (01:32:18 – 01:32:19): It’s called the champion system.

TOM BILYEU (01:32:19 – 01:32:21): We’ve been talking about it, we just didn’t name it.

EMAD MOSTAQUE (01:32:21 – 01:33:17): Yeah, it’s called the champion system. Got it. Because right now the amount of AI in the public sector is zero. It will drive the public sector, it will run the public sector things again, just like a utility. And it should be regulated as well because you want an entity to regulate that has the reach.

But then what’s going to happen is I think again, we’ve got the game-based worlds and other things, as you said, great feedback loops, great status mechanisms. World of Warcraft Forever can’t come soon enough in terms of its mass release. Go back to the good old days. I always loved hiring guild masters and things. They knew their crap, if you played that game properly.

I think there will be mass enlistment for citizen service corps. That is a way of getting young people, you said, who are dangerous people, off the streets. I think that there will be looks of other ways to build status when, again, you can’t have that materialization thing.

TOM BILYEU (01:33:18 – 01:33:21): And why can’t you have it? In your head that goes away, in mine it doesn’t.

Debt Jubilees and the System Reset

EMAD MOSTAQUE (01:33:22 – 01:33:30): No, in terms of for the vast majority of people, there is no more social mobility. Historically what’s happened at times like that is a debt jubilee.

TOM BILYEU (01:33:31 – 01:33:37): Okay, so well, we’ve got 2 things now. So we’ve got there will be no more status games— I think you’re out—

EMAD MOSTAQUE (01:33:37 – 01:33:46): Or not status games, there will be no more wealth accumulation to deal with no more scarce resources, and there will be no more accessibility to that for a lot of people.

TOM BILYEU (01:33:46 – 01:33:50): I think you’re crazy. So one, we’ll debate that in a second.

EMAD MOSTAQUE (01:33:50 – 01:33:50): Yeah.

TOM BILYEU (01:33:50 – 01:33:54): Then we’ve got the— oh God, what was the other thing you just said?

EMAD MOSTAQUE (01:33:54 – 01:33:54): Debt jubilee.

TOM BILYEU (01:33:55 – 01:34:07): Debt jubilee. Okay, so 2 very important things. Why don’t we tackle debt jubilee first? Because that’s actually going to come earlier. So give me the scenario in which historically debt jubilees are a thing, and what is a debt jubilee?

EMAD MOSTAQUE (01:34:08 – 01:34:43): So debt jubilee is that periodically an economy would acquired too much debt and resources— resources would go to the owners of capital and they would have liens and they would have people owe them all sorts of things. So what would happen is a government would announce all debts are cleared, so they’re all zeroed out. So that’s obviously very bad for the lenders. It’s very good for the creditors, to say the least. So that’s what happens when the government and when total debt in the system is too much. That seems inconceivable to us, but again, you can see the historic record of that.

TOM BILYEU (01:34:43 – 01:34:52): And right now, how many debt jubilees tend to come in a time of absolute crisis? Yes, or world wars. That’s when debt jubilees happen.

EMAD MOSTAQUE (01:34:52 – 01:34:52): Exactly.

TOM BILYEU (01:34:52 – 01:35:03): Jubilees don’t happen when everything’s going good and they just go, “F*** it, debt jubilee.” So I think when people hear debt jubilee, they need to hear, you’re in crisis, people are dying in the streets.

EMAD MOSTAQUE (01:35:04 – 01:35:04): You need to—

TOM BILYEU (01:35:04 – 01:35:05): That’s when you have a debt jubilee.

EMAD MOSTAQUE (01:35:05 – 01:35:06): You need a system reset.

TOM BILYEU (01:35:06 – 01:35:17): Yes, they are bloody. They are violent. People do not give up their— what’s owed to them easily. Just so everybody understands, they use a nice pretty word, but this is a f***ing dangerous time when debt jubilees are happening.

EMAD MOSTAQUE (01:35:18 – 01:35:22): Debt jubilees happen when the lenders are afraid of dying, effectively. Yeah.

TOM BILYEU (01:35:22 – 01:35:33): Oftentimes a debt jubilee happens because people are dying everywhere. World War II, World War I, that was a debt jubilee. So much s*** got blown up. Everything just got reset effectively to zero.

EMAD MOSTAQUE (01:35:33 – 01:36:38): I mean, historically there were some less extreme scenarios of that, they would do them regularly in various kind of things. But right now, if we’re going to have one, it has to be an extreme scenario. But again, to the listeners, what we’re looking at here is your cognitive labor has negative value and there’s no market for that. The humanoids can do the jobs better, and the countries like China that will embrace this are going to far outstrip anyone. They’re going to see 30% growth rates. And that will be at the expense of our own growth rates.

It’s a race to see who can have an incredibly more efficient workforce that can do just about anything. And it’s not like when you had the outsourcing— outsourcing, there was friction, right? It’s difficult to have an outsourced team do your code and things. AI, there’s none of that friction, especially with the capabilities again now.

So we have to look for what are worst-case scenarios in 10, 15 years. We have to look at what do they used to who are revolting. We have to look at what do people do to live. And in America, there are so many people burdened under huge amounts of debt that they can barely live, right? Student loan debt not being dischargeable in bankruptcy still is crazy to me.

TOM BILYEU (01:36:38 – 01:36:39): Crazy.

EMAD MOSTAQUE (01:36:40 – 01:37:11): Insane. Medical debts, these types of things. You could see partial debt repudiation, but I think again, the path we’re going, if we don’t create new institutions, new mechanisms of participation, new things for people to look forward to, is a very ugly one fundamentally. Because again, what does it mean to be you? What does it mean to be us? Those are going to change even if AI does not get one IQ point smarter than it is today.

TOM BILYEU (01:37:13 – 01:37:52): Yeah. Okay, so that’s debt jubilees. Now over to the— I think the reality is there’s no escaping what we see now, which is, we call it money. It has been glass beads, it’s been salt, it’s been rice. So once people understand what money actually is, they will understand why this is not going anywhere.

So money is, you did a thing with your time that’s more valuable than the next person, and because you did that thing that’s more valuable with your time, then you’re able to accumulate something for scarce resources. Assuming a world where the AI doesn’t just kill us all and tell us to sit down, shut up, whatever.

EMAD MOSTAQUE (01:37:52 – 01:37:52): Yeah.

TOM BILYEU (01:37:52 – 01:38:40): If they’re still catering to our wants and needs, they’re going to have to answer the question, which of these humans gets access to the scarce resources? So even if that’s, “I want a house in a particular location,” only one person is going to be able to have that house in that particular location. So there has to be some way that we decide what that is.

If you do RNG, just straight random number generator, we pull a number out of a hat, the populace will be absolutely ablaze with revolution. It will not work. People have to feel like they can earn their way into that thing to be in alignment with the way that the human mind works. And so now we’re back to there will be something, even if it’s black market cigarettes, there’s going to be something that we come up with that acts as a currency that people can use to get access to the scarce resource.

EMAD MOSTAQUE (01:38:41 – 01:38:48): Yeah. And so I don’t disagree with you there. I was saying more in terms of I don’t think it’ll be one economy anymore.

TOM BILYEU (01:38:49 – 01:38:49): Interesting.

Longevity Escape Velocity and a Two-Tier Society

EMAD MOSTAQUE (01:38:50 – 01:39:27): So again, we see this in sci-fi, Gattaca and a whole bunch of other things. You have multiple levels of society right now. A huge number of Americans don’t have one month of savings, but there is the promise that if they work hard enough, they can accumulate savings, they can move up the curve if they get a break, if they get a shot. That won’t be the case anymore. The owners of capital become wealthy. And even worse— not even worse than that, even crazier than that— I think in the next 5 to 10 years, they hit longevity escape velocity.

TOM BILYEU (01:39:28 – 01:39:29): Meaning we live forever, effectively.

EMAD MOSTAQUE (01:39:30 – 01:40:49): Longevity escape velocity is your total life expectancy per year increases faster than the year. So there is a drug trial that’s just been done by Insilico for something called idiopathic pulmonary fibrosis, basically lung things. I know a few people have them, so they’ve been on the trial. 4 weeks of that drug reduces all of your healthcare longevity markers by 4 years. Whoa. GLP-1s have shown massive cancer reversal and other things like that. We’re seeing the first of these drugs.

If you look at what’s coming through the pipeline, the rich will be able to live longer and healthier, whereas I don’t think things are great for the poor. And I think the ladders are going to be pulled up. This is the other thing. So the people who compete for the nice things are the rich, the owners of capital who can accumulate capital more. If you can’t exchange your labor for capital and you’re relying on government handouts, how are you ever going to have the progress to be able to buy that nice spot, to be able to buy those handbags, other things like that?

Again, the status games is what I was referring to. And you could have a 2-tier society, just like a food stamp society, and you have a VIP rich party socialite society, right? It could become even more diverse because how do you go from one group to another? The ladders are being pulled up.

TOM BILYEU (01:40:50 – 01:41:06): And again, interesting, that doesn’t strike me as the outcome when energy costs are dropping to zero. Intelligence cost is dropping to zero, labor is dropping to zero. When you get that, then it becomes— you can have things become extraordinarily abundant.

EMAD MOSTAQUE (01:41:06 – 01:41:06): Yeah.

TOM BILYEU (01:41:06 – 01:41:48): And the only thing that we’re fighting over is those things that can’t be made easily. So if there’s going to be a finite number of the cancer cures, then yes, now you’re back to— again, there’s going to be some sort of economic system. I don’t think it will calcify into the haves and have-nots because money is so transient in a situation where you can lose money.

If it became an ownership class and a permanent serf class, then yes, now you’re f***ed. But I would be shocked if we ever let ourselves get back into that situation, especially with a super persuader AI. He’s going to be like, “This is unjust.” We would convince us that we’re— that’s true because it’s destabilizing for society, if nothing else.

EMAD MOSTAQUE (01:41:49 – 01:41:52): Well, you can have relatively stable societies with that. I mean, sort of.

TOM BILYEU (01:41:52 – 01:41:53): It doesn’t feel—

EMAD MOSTAQUE (01:41:53 – 01:43:25): But again, let’s put the super persuader to the side. What’s coming right now is a series of decisions that we have to make as a society at different levels. And there are very few people, unfortunately, that I’ve seen who are looking very far out and really understanding the inevitables here. Yeah.

So one option is a serf society. One option is a Star Trek future. One option is freaking war and Mad Max, really. Another option is we all turn Amish. Again, you could see vast swathes of people just say, “Nope, I’m out. I’m not going to be ruled by an AI. I’m going to be the noble savage in 1984” or whatever, right?

And the level of human things, what happens if you live forever or if you live for 200 years? What happens if you can put a brain chip in and control your emotions? You can have all sorts of crazy sci-fi futures happen over the next 10 years. What happens when robots are walking among us? Do we have to give them rights? Do we have to give them the vote?

So what I’m saying is that these are decisions that we have to make, but there are some things again that are very tangible. As you said, people will want status and they’ll want a way to access that. They will need an identity that is individual and communal. There needs to be a way to progress some way or another, and you’ve got to provide for people. You can’t just let people starve. You’re freaking— this is America, right? We’re in the West. We should use the abundance to make sure that people have a baseline of living, and then we should have a system that through their own efforts they can excel. And the time for having those discussions is now, because like I said, it’s a sandpile that will start collapsing, I think, in 2 years.

Build Stronger Communities

TOM BILYEU (01:43:26 – 01:43:42): Okay, so to recap the things that people should be doing in the next 2-year time horizon: be the last man standing, make sure you understand your opportunities for entrepreneurship and ownership. And then what final thing should they have in their mind?

EMAD MOSTAQUE (01:43:42 – 01:44:06): I think it’s build stronger communities, right? The strongest communities will survive this no matter what because they can become self-sustaining, they can support each other, etc. Many of us live highly nuclearized societies where we don’t see our families, we don’t see our community anymore, right? We’ve moved to the big cities, and I think that support structure will be super important for what’s coming one way or another.

TOM BILYEU (01:44:07 – 01:44:11): Emad, this has been wide-ranging. Where can people follow along with you?

EMAD MOSTAQUE (01:44:11 – 01:44:16): You can see me @EMostaque on Twitter or ii.inc for Intelligent Internet.

TOM BILYEU (01:44:16 – 01:44:21): I love it. Well, thank you for continuing to try to solve these big problems, man. It is always a pleasure.

EMAD MOSTAQUE (01:44:21 – 01:44:24): Yeah, it’s the most exciting time ever, that is for sure.

TOM BILYEU (01:44:25 – 01:44:26): The good, the bad, and the ugly.

EMAD MOSTAQUE (01:44:26 – 01:44:26): Yeah.

TOM BILYEU (01:44:26 – 01:44:52): All right, boys and girls, if you have not already, be sure to subscribe. And until next time, my friends, be legendary. Take care. Peace.

Related Posts

• Transcript: If Society Collapses, Cities Become Unliveable In 48 Hours – Prof Lewis Dartnell  
• Full Transcript: Mark Cuban Interview on Diary Of A CEO – Oct 26, 2026
• Stephen Semler Interview: Poverty and Inflation Crush Americans – Tucker Carlson Show
• Transcript of President Zelenskyy Interview: Diary Of A CEO Podcast 
• Transcript of Megyn Kelly Interview: Cornell 7 Case on Tucker Carlson Show