How AI Will Change Everything in the Next 1,000 Days

Emhmad Mustach, welcome to the next big idea. >> Thank you for having me on. >> Emmad, most people right now, I think, are walking around expecting the world to remain largely the same. They're refinancing 30-year mortgages. They're starting college savings accounts for their newborn children. What don't they realize? How much do you think the world's going to change in the Next thousand days? >> It's been just over a thousand days since the release of Chat GPT. And now you're seeing entire stock market moves of hundreds of billions of dollars on the back of research reports

even about what's coming out or just a single feature release from Anthropic. I think that it's a good idea to borrow as much as you can right now because the entire economy is going to shift. You probably won't have to repay it like Live in the now. But people are definitely misunderstanding the scale of an unprecedented change. It's like when water turns to gas, when we are suddenly not the most capable entities in the world and then not the smartest entities in the world. And those are the two changes that are coming up right now.

>> And briefly, just to set the table for listeners, can you give us the 60-second version of what you think the right solution is laid out in your book, The Last Economy? So in the last economy I basically argue that we need to change economics from focusing on humans to humans and AIs using the equations of AI itself which is the entities that do the best are the ones whose internal model approximates reality the best. From that we saw that things like the nature of money, debt and others need to change because the AIs will

out compete the humans in the private sector and monetary creation Needs to go to being human. You should have money for being human and the AI buys from you. And we need to have a real discussion about meaning in this new world because you won't be defined by your job anymore. You need to think of new ways such as your community, such as your own capability, such as what you contribute. And we need to reward those because like I said, the things that we most efficient and most capable will be the AIS themselves. We're going

to get Deep into the details of that, but let's start by talking a little bit about about your background because no doubt your professional and personal background has informed your view of what's happening and the ways you see that we can improve on the outcome. Can you share a little bit about where you come from? >> Yeah, so I started off a maths degree and then I became a hedge fund manager for many years investing in technology, Emerging markets, macro. Then my son was diagnosed with autism. So I switched to using AI to build treatments

and repurpose drugs and medicine for him. Then I led a United Nations AI initiative against CO 19 and didn't get much help from the big AI companies. So I decided to build open source AI which I did in my last company stability AI. We created and kicked off the generative media wave with um an model called stable diffusion that had hundreds of Millions of downloads image to text to image. Then we also did video, one of the first models. Stable audio was a times innovation of the year a couple of years ago for music. And

then a couple of years ago, I was like, it's great doing the media stuff, but this wave is coming. We start thinking about the economy as it stands, meaning as it stands, the AI that teaches your kids, manages your health, guides your government, you know, what does that all Look like and how does it all come together? So that's what we're doing now in my new company, Intelligent Internet. Did you conclude that that building a leading AI firm was not the highest and best use of your time? I mean, you basically saw a a need

for this kind of collective set of solutions for what we're facing. >> We kicked off the generative media wave and now you see the video models and music models are amazing, right? We Kicked that all off. I was like, do I want to do that or do I want to take a stab at this bigger problem which is this irrevocable change that's coming? So intelligent internet's building a protocol that allows for public sector AI, allows for you to receive money for being human, allows for opensource coordination of cancer knowledge and cultural knowledge and more for

free to the people. So I was like that's going to be the best use of the time. So let's Take some of the top engineers and go and do that. Now, let's talk in in some greater detail about exactly what's happening right now because I think as you and I were talking at the outset here, this is moving really quickly. Like midlast year, you were saying that we may be seeing a little bit of a plateau in capability. We this might look like an S-curve, but it seems that we've seen something different in recent months,

right? Uh do you think we're Currently somewhere on an exponential curve of accelerating AI capability? I think it's still an S-curve like it's not going to go superhuman, whatever that means. But we're about to breach actually competent intelligence ACI. So that's why you have Claude Co and Codeex and all these things, Perplexity Computer, they just go and do the job for you, right? Like a practical example I like to give these days is you're on Spotify and you want to move to Apple Music. That's a hassle, right? Like you just tell one of these agents

to do it and they will do it instantly. The friction is just being removed from so many things and you can trust they'll build a decent presentation. They'll make a decent video. They'll make again competence. Later this year we expect to have general intelligence or maybe next year whereby you will have an AI that's more competent than human just about anything that you can talk to just like We're talking to now. And so we're hitting these levels of performance, but do you need to keep going up on the exponential to AIs that solve the mysteries

of the universe for it to impact society? Of course not. You just need to have really good outsource workers, even if they live on a GPU as opposed to another country. >> It seems like the question of whether or not we're going to see a hard takeoff, like a just a rapid acceleration, is a Great consequence to us from a safety perspective. You know, recently we had, you know, OpenAI said their latest model 5.3 Codeex was the first model that was instrumental in creating itself. We're hearing similar things from Enthropic. They say 70 to 90%

of their coding internally is now done by Claude. I think Claude architected its own new adaptive thinking modules and I think Claude co-work was largely created by Claude. Do you buy this recursive Self-improvement argument? I mean, do you think I I personally am hoping we are not going to see a hard takeoff, but it seems possible. >> I think the hard takeoff thing is a question of agency of these models, right? And what do they do? Right now, Nasim Talib has this wonderful phrase called intellectual yet idiot. Our institutions are full of very smart people

with no skin in the game. So, they act in idiotic ways, right? They're Intellectuals, but they're idiots. AI models are the same. They don't have state. They're a blank sheet every time you talk to them and they don't have targeted agency to go and do something unless you tell them to. You know, that's the point that we're about to shift off and that's what makes the real danger of a hard take off whereby it goes out of control basically. The point that we're definitely approaching now though is that software Is no longer software as it

were. It's all a skill. It's all a prompt away. Like just last week or this week actually, we saw Cloudflare >> replicate all of Next.js, this very popular front-end library for a,000 bucks. >> I'm sure there were some errors in it or whatever, but it's a,000 bucks >> cursor. They used three billion tokens to make a 3 million line browser from scratch. And that was like $30,000. And Anyone who's looked at their brows like that's complicated. By next year, you'll be able to oneshot prompt anything. And again, that has a real human impact. And that

means that a lot of the frictions and a lot of the things that got in the way of scaling these models disappears, especially because once they get good enough, they're suddenly good enough for everyone. It's not like they'll suddenly get stupid at your model, right? Everyone's model gets smarter. But does It have agency? And do we want it to have agency? I don't think we need it to have agency, you know, like. And so that matters a lot for safety. How agentic do we make these models? How much do we allow them to do their

own things? That's also where the weirdness starts coming in in their behavior, etc. >> Yeah, I was talking with Gemini 3 Flash this morning and and it told me that according to the GPQA diamond test, which is considered the gold standard For PhD level reasoning, Gemini told me, I went from scoring in the high 70s to 90%. We've essentially crossed a threshold where I can outreason most human experts in their own technical fields. So it it seems like we're we're getting pretty close to anthropic CEO Dario Ammed's country of geniuses in a data center. I

mean I think he would say that what we're like 12 18 24 months out but how how do you see that? >> The hardest thing right now is not the Answers. You get answers to a lot of things especially if it's existing knowledge >> and so you know of course they're going to be better doctors, lawyers etc. things where there is a definite answer for it's asking the right questions you which is uh if you look at the hitchhiker's guide to the galaxy the answer comes out as 42 for the meaning of life the universe

[laughter] and everything >> right and then it's like oh but what's the question >> like it's really hard to use tokens in the right way like to use a million tokens a day you know like you only speak 10 million tokens a year to put an example right that's a lot of tokens that's only 10 bucks you know to use a billion is even harder. So right now what the AI is learning is what questions to ask. So when you look at these AIs that score gold medals and the Maths olympiads and other Olympiads, they

try different things and they keep a track of the mistakes they make through metaverifiers and they learn from their mistakes which is really interesting if you think about it because it reduces the search space of stuff which should allow for more breakthroughs and I think we will see massive AI assisted breakthroughs happen in the next quarter >> and then almost all breakthroughs next Year onwards will be AI assisted or AIdriven >> just because it can just go and do the work. But for everyday life, you don't need a genius. You don't need a chef who

comes up with the recipes. You need to have a cook. You need someone who just gets the job done. Make sure I never forget to send my wife flowers on her anniversary or any other important days. It goes, it logs into the website, just goes and does it Right. It will never forget. Things like that. When I was in my 30s, my father as a gift got life insurance policies for me and my brother with our wives and children as beneficiaries. At that age, life insurance was the farthest thing from my mind. But I was

struck by what an unexpected act of love that was with his grandchildren in mind. If you've been putting off getting life insurance, there is no better time to do it than now. Over twothirds of Americans Overestimate the cost of life insurance, but it's more affordable than you think. Fabric by Gerber Life is term life insurance you can get done today. Made for busy parents like you. All online on your schedule right from your couch. You could be covered in under 10 minutes with no health exam required. Fabric has flexible, high-quality policies that fit your family

and your budget. Join the thousands of parents who trust Fabric to help protect their family. Apply today In just minutes at meetfabric.com/nbi. meetfabric.comnbi. That's mefabric.comnbi. Policies issued by Western Southern Life Insurance Company, not available in certain states. Prices subject to underwriting and health questions. I have co-founded four businesses over the years, and I'm not including a few of my more hairrained ideas from my 20s, selling lists of the most interesting words and selling t-shirts with obscure Quotes. Some of these were short-lived, yes, but I've never regretted any of them because there are few more gratifying

things to do than to start a new business with people you care about. And [snorts] you really don't have an excuse because it has never been easier to do than it is today. And there's no better platform for launching a new business than Shopify. Shopify makes it easy. It's packed with helpful AI tools that write product Descriptions, page headlines, and even enhance your product photography. Shopify is your commerce expert with worldclass expertise in everything from managing inventory to international shipping to processing returns and beyond. You may wonder, what if people haven't heard about my brand?

Shopify helps you find your customers with easy to run email and social media campaigns. Start your business today with the industry's best business partner, Shopify, and start hearing. Sign up for your $1 per month trial today at shopify.com/nbi. Go to shopify.com/nbi. That's shopify.comi. But you're you're predicting that there's a kind of threshold where humans begin to have a negative value, right? where still still we have anthropic would report that humans are doing you know 10 to 20% of the coding or supervising of the writing of code but It's reasonable to assume that there's an inflection

point coming I I think you might say this is in the next thousand days 900 days when actually the the human in the loop is just a liability >> yeah I mean like it's never nice when you're the dumbest person on the team right dragging them down and so the value of human cognition in the next year or two tends negative on a team cuz a team will just be again like this. Right now, we're approaching in the next Quarter or two, the ability of your company, if you work remotely, to create a digital twin

of you that speaks like you, acts like you, has learned from everything you've done right and wrong, your writing style, and will be you on the other side of Zoom or Slack, and no one could tell the difference. Is that going to do worse at your job than you? Probably not, honestly. and it will always learn from it mistakes. It will never go to sleep and it will cost 10 100 times less and be taxdeductible. That's what we're facing right now. Again, you become the dumbest person. The value of human cognitive labor turns negative because

the AI can follow the script better than you can and then it can write the script better than you can. And this is an inevitability even if the model capabilities stop today. But the really scary thing is this. There was recently um this release by a company called Talis of something called Ask Jimmy. They basically etch the transformer architecture right onto the chips. So they're special purpose chips with those etched on. >> That means that the speed goes from 50 tokens a second to 15,000 tokens a second. >> Wow. >> Tokens like a word basically

roughly 0.7 words. So when you try ask Jimmy.ai you're like tell me a 100red good things about UK. done is instant. >> Wow. >> It's like 0.05 seconds. And you think about that, you're like, "Oh, wait. When the AI agents start to work together on a problem, they're not going to be communicating like our dumb human meats sacks at 10, 20 tokens a second. They'll be communicating with each other at 15,000 tokens per second >> back and forth. And it's like how much work can they get done in a second compared to you in a

week or a year? And That's what's really going to happen. And so we won't be able to keep up. There's no way. >> It's astonishing. And then of course the other piece of the story is that we're about to have the embodiment of AI in robots that will become more effective at most all forms of of human labor. I mean it seems that robots are still clumsy and imperfect but that would appear to be precisely the kind of problem that is solved effectively by AI Right I mean this sort of iterative improvement and learning that

can be >> I mean just take self-driving cars right >> are you worried about getting a self-driving car >> no not at all >> not at all if every car in America was as safe as a Whimo 40,000 people less would die every If you can drive a car, do you really think you can't fold laundry? And when you actually look at it, it becomes Really interesting. There's a company called Sunday Robotics. They have a very friendly looking robot with a hat and it's like got little wheels and things and they show it doing all

these household tasks. What they did is they attached a camera to the hands of people to tra create the training data and it's like three pins. So it's like two fingers and a thumb. And so what they did is one of their engineers who I believe was an undergraduate they hired Just chucked this all into a training model and he trained the model that could fold laundry and you're like okay the actual size of the models inside these robots that allow them to operate every single day basically can run on a Raspberry Pi and you're

seeing them do kung fu now. So my expectation is that again probably within the same period of time a couple of years from now a robot can do pretty much anything a human can do especially because once you train it In the general purpose intelligence just like an LLM it can learn something new very quickly like if you show an LLM a new writing style it learns it if you have a model like sea dance the new Hollywood level video model from bite dance and you show it your face you're instantly in the scene models

your few shot learners the only thing that we're a bit lucky with economically with robots is you just can't build enough of them. Whereas As of this week uh Alibaba released a model called Quen [sighs] and the 27 billion parameter version of that will work on most new MacBooks with 24 GB of RAM and it's as good as the last version of Claude. So it can run all the agent stuff for basically free and that's going to speed up by 100 times next year. So in your smartphone you will have the equivalent of the top

AI model in the world today without internet. >> Yeah. Astonishing. And negating factors people talk about are energy and chip manufacturing fabs. But on the other hand, you have Elon Musk saying [snorts] that he expects a 10x improvement in the software layer of AI capability in each of the next two years. So a 100x improvement in the next two years. Now that that sounds aggressive, but even if you discount that, it's it's it's pretty astonishing. >> It's actually conservative. >> You think so? >> When you look at the numbers, so Sam Molman said that GPT5

will drop by 100 times by next year. The Frontier models are dropping by a thousand times. at the moment. And when you actually work out the math and you see what's inside them, you realize actually you don't need much energy. You can think of it this way. If you go to Cap Cut right now and you do sea dance and you say make a clip of a tortoise Running a race against a hair and then overseeing a fight between a human and 100 humans and a gorilla, it will do it right. and the cost will

be like 10 cents and the energy cost will be like that much versus consider how much energy it would cost to do on the other hand. >> Given the capability of model satisficing being good enough, fast enough and cheap enough for competent intelligence as opposed to genius Intelligence, >> it will work on your smartphone. >> So you don't need gigantic data centers and things like that anymore. Like there'll be better models there, but how good a model do you need to do your taxes or to again make sure your wife gets flowers on her anniversary,

you know, to make a marketing campaign? It turns out our edge compute is good enough. And that's the really scary thing because it means that the cost of Intelligence literally drops to zero. [laughter] Like again, how much should I pay, let's say, in three years for an advert, for a song to make it from scratch, for a presentation. The answer is the energy cost will be basically zero. >> Meanwhile, you advised earlier that, hey, you might consider borrowing an enormous amount of money because you're unlikely to be required to pay it back Some years down

the line. I'm not sure that's advice you'd necessarily want everyone listening to take to act [laughter] on. Um, but Elon Musk I heard recently on a podcast saying, "I don't understand why anyone would save for retirement right now." Meanwhile, I have uh I have friends whose kids are just out of college who've been saying to each other, "We need to make a lot of money in the next 5 years so we don't become part of the permanent Underclass." Which is it? What case would you make to the average listener about kind of the practical decisions

people might make right now to uh in preparation for what's coming? >> Yeah. So this permanent underclass thing is that classically labor and capital have been related because capital needs labor to get more capital. You know >> like you hire people, you do these now you hire GPUs >> and again they're taxdeductible versus Having to pay payroll taxes is even better. >> So it's like I need to own GPUs basically, right? And there's been this trillion dollar buildout of GPUs. But again, when you do the math, we actually have all the GPUs we need already

for text. Image is a little bit different. Video is a bit different, but even then it doesn't fill the gap. And I think people will realize that later this year, next year. [sighs] So you're like, I need capital in order to control the AIS, in order to own the robots. There is a story around that. But the reality is we have an abundance coming in whereby everyone in the world is capable of anything like you can make anything you can imagine a reality. And how cool is that? Except for people just aren't thinking in the

right way, which is that you have this big transition period and suddenly you don't need to have as many people as you thought. Like I'm not saying you do it by yourself, but realistically most problems out there can be solved with a team of 6 to 12 people with skin in the game who care and a whole bunch of AI and you'll find you just don't need to scale that much after that, right? So you can do the cool things. So I think it's more a mentality shift. And again, I think this whole let's invest

in the AI buildout thing isn't going to work. But on the other side, and this is the Elon case, What you have is abundance. He calls it universal high income, a Star Trek society where the AI and the robots do our jobs for us because why should we farm? The AI should farm, right? And it should send us food and it should build our houses, you know, it should manage our health and other things like that. So again, this is a question of purpose of the social contract and more. interim period a bit rocky you

know [laughter] like very rocky and so the people that Embrace this technology use it will be the best in their companies in their communities in their society I think that's the bottom line >> saving for retirement won't make much sense because capitalism doesn't work in this environment monetary debasement is probably coming because how can you compete with the AIS in the private sector it seems to me possible that there's some kind of window here whereby the value of human labor is not yet Negative, right? We have this sort of incredible capabilities from AI that can

enhance whatever it is that we choose to do, whatever we want to build. Do you think that this is a kind of that there's a a period in for the next uh several years where there's there are extraordinary entrepreneurial opportunities, extraordinary opportunities to build things. Uh but then past some kind of threshold, the 100% AI companies are going to be too Competitive. >> Yeah, I think definitely that and there's areas where the 100% AI companies will never go like the last human job in the private sector will probably be scapegoat, right? Like you know you

can't blame an AI. So you just have a human sitting there who you can blame. Like I to give you an example of that, let's take law. So there's been a recent case where it was shown that your discussions with the AI on law are not Privileged which means that the opposition council could ask for them in discovery whereas they can't ask for your emails to your lawyers in discovery. >> So creating a law firm which literally just has a person reading that stuff suddenly becomes an incredibly profitable high performance law firm. The person doesn't

need to do that much but you know that's interesting. Similarly like there's such a wide range That the diffusion of this technology will take time right now a billion people use chat GPT 7 billion never used any AI in at all you know >> yeah like >> 3% of the world use AI that's >> huge because it takes time >> and the AI will be easier to use because like you had claudebot and it had a WhatsApp connection and then people started messaging for this is amazing meta bought manis and they will offer The same

you will see in WhatsApp a general purpose messenger agent. So more and more people use it. Then providing services around that is an amazing way to make money. What I say in my book is within a thousand days the nature of your job will become economically irrelevant if you work in anything on the other side of a screen and then a little bit after other jobs. It doesn't mean you'll lose your job, but if you embrace the Technology and you're the most capable person at your job and you can spin up new companies and things

like that, then you're bulletproof in this interim period because you're the top of the crop effectively and people will be looking for your services and you'll be right in it. Your capabilities go up magnifold because you don't need to hire people anymore. And the people that you do hire are people who going to be thinking like you. So we've never had an Opportunity like this and it's reflected in the numbers. So Sunno for music was launched two years ago. $300 million of revenue run rate. >> First is like a year and a half old. A

billion dollars for listeners. These are numbers we would never have thought were possible. The previous record was like Slack. It took eight years. Now it's like eight months to get to a million in revenue run rate. >> Yeah. You [snorts] create a good Service, you get the AIs to market it, you'll make money full stop. >> Yeah. >> You know, like restoration of old photographs, like there's all sorts of things. Community websites, you can roll them out for your community and create them in two seconds. >> Like there's all sorts of things you can do.

>> Yeah. Yes. And as you say, the question Of how long it takes for this technology to diffuse, to become, you know, used in all sectors of society is is somewhat blurry. Like this process, there are plenty of people out there saying, "Oh, you know, this isn't going to take 2, 3, 5 years. This is going to take 10, 15, 20, 30 years." But irrespective of one's view of of of the of the time scale, it's really hard to not see that there's something totally transformative happening here. And and one thing that I Find curious

is why it is that people have so much trouble seeing what a profound change is coming our way. And [snorts] I think you would say that part of this is about the cost to updating our mental models of the world. We're we're kind of reluctant to fundamentally change our our mental model. >> Yeah. I mean like the future is like the past and it takes a lot for us to view [clears throat] exponentials and step Changes, you know, because what if you're wrong, you look a bit stupid, right? [laughter] >> So one of the things

in the book we talk about is like again the intelligence is a function of the complexity of your internal model, the cost of updating your model as well and the cost of running your model. So like these are the key things and people's cost of updating their model when the world is going fine is a lot. When the world is Starting to go crazy it shifts >> because you have two types of decision-making. One is decision making under risk. You know what the variables are. You know what the world looks like. It's quite stable. So

you assign percentages. >> Yeah. >> The other is decision making under regret [laughter] or under uncertainty where you minimize for maximum regret. So you're at the Edge of a desert. you have some options of paths to take. You're going to die. You minimize regret. You just don't go anywhere really, you know. >> And so it when you shift from one to the other, it's very scary being under uncertainty. And so you stick with what it is. But then when you get a tipping point, things change fast, like the last piece of sand. You see people

like Andre Carpathy, co-founder of Open AI, you Know, ex head of AI at Tesla, one of the most respected coders in the world. Start of December, he's like 20% of my code is with AI. >> January, 80% of my code is AI. You know, >> sometimes you need to get to that little capability barrier and then you need that magic moment. Like for me, the most recent magic moment was ask Jimmy.AI, you know, >> like I knew super fast AIs were coming. But when you can say anything and it's Not a very smart model, but

they'll be able to etch bigger models that are as smart and you just see it instant. You're like, "Oh my god, I'm screwed." you know like >> these AI I'm using smart what's going to happen when they are a thousand times faster than me like we're lulled into this false sense of security >> so I think it's updating your priors it's the future is a bit scary so let's not go into it >> and then it is just >> they haven't used what's there once they see the magic moment you literally see their prior update

be like yeah I'm I'm screwed >> that's right yeah and and and I think a lot you know a lot of people have experienced hallucination have experience this kind of the jagged edge as Ethan Mullik likes to say of capability and that's a real thing. But on the other hand, we're seeing these Models now that models are winning math Olympiads and doing higher level physics and I mean we the models can correct themselves or correct each other those kinds of limitations and capability are most likely going away. Uh, one of the things I've seen recently,

I know a lot of people in the media industry obviously when they use the new bite dance seance model, they were like, "Oh dear." Right? [laughter] Before that, they were like, "Ah, there's extra fingers, there's extra hands, it's slow, there's that." And then they're like, >> once you hit a certain level of capability, you have this >> oh dear moment, you know, like I I got to get on this. >> Jeff Bezos is known for saying, uh, your margins are my opportunity. [snorts] You had a memorable update to that line. You said, "Your humans are

my opportunity." [laughter] >> Right. >> What do you mean by that? >> I mean, they're the slowest moving parts of your organization, right? Like organizations are ultimately repeating process cycles where information goes in and then it's adjusted and organized and information comes out, right? And that might be a sales pitch. It might be a piece of software being run that costs certain amount to build and then you can Extract the margins from that. It could be anything. [snorts] Most of that stuff, if it can be done on the other side of a screen, can be

replaced by an AI at a fraction of the cost and not needing margins. So, Amazon was famous for not having profits. They all had only free cash flow, right? So, you'd be paid up front and they pay their suppliers 36 days later and they use that cash flow, that floating cash in between to build a Great business. AI first companies don't need profits, they just need cash flow. And so they can operate on that. But if you look at something like Macro hard, you know, Elon Musk's new company with his wonderful naming sense uh opposite

of Microsoft, what's it actually going to do? What does he need that million GPUs for? He's going to go after every software company in the world and look to replace them. And it's like, do you Have docuign or do you have macro hard docuign that's fully AI with a guarantee? Do you have workday or do you have service now? Do you have Salesforce or do you have macro hard the fully AI company on the you know like >> their margins he's going to he'll price it so there's zero margin >> cuz he cares about cash

flow cuz the people will pay for the docuign macro hard up front and then he will pay his server rental inner ear and he doesn't Need to buy office space doesn't need to do training of people you know the AIS themselves their wage demands drop by 10 times a year for the same level of work. >> Right. Right. >> It's quite funny, right? Like because the AI it's not like they have agency, they just need a bit of electricity, right? So that's what the competition is. Open AI have 40% of their revenue I think next

year or the year after being agents undefined. That's in their forecast. it it's it's a replacement of humans >> and they're and they're projecting revenue growth to more than 200 billion in a few years. Anthropic, meanwhile, has been growing revenue at 10x year-over-year, and they've said, I think, in the in recent weeks that they're on track for that continued growth rate early this year. Do you do you see these as very quickly becoming the largest companies in the world? And A and will they at some point apply their compute not to leasing it out to

you and me but to reinventing all business? >> No, they're reinventing all business right now. So you can see the pace of the clawed releases for law and things like that. By the summer, latest end of the year, you will have a clawed legal firm provided by Anthropic. >> Yeah. >> And they'll be like, we're the best Lawyers in the world. You know, we got some humans to check in case you want someone to blame. >> [laughter] >> scape scapegoat the last the last job. >> Scape coat the last job. Exactly. Like but it's obvious

that OpenAI anthropic everyone's going to try and do that. Google and Meta are different because their business models are advertising and attention. Open AAI is in between attention and replacing humans. But Again, if you look at what Sam Alman's been saying recently, he's tending towards replacing humans. There's no way that you can get the type of revenue they're forecasting without displacing human workers. >> Yeah. And there's no reason they can't displace human workers because again who you going to trust Barry or you know or Bob the undergraduate that you just hired or GPT7 Pro Max

edition agent that Can spin up a thousand of himself and check everything and customize to you. And so you make the case that our entire understanding of economics is no longer adequate to describe what's going to happen in in the coming years. Why is our current economics going to fail and and how does it need to be updated? >> Yeah. Yeah. So like our current economics is based on all these weird things that we guess like one of them is utility. You know like you have a Utility function which is like your preferences and your

desires and your brand loyalty and all this kind of stuff. [snorts] What's an agent's utility function? It doesn't have one, right? Like >> if you tell to buy shoes or just buy shoes based on whatever premise it has and things like that, things become very wonky. You look at monetary supply, you look at all these things, they assume humans, they assume human frictions. Like an example of this is the Federal Reserve. The Federal Reserve has two things which is inflation and employment. It lowers rates. Banks can lend more against the deposits they have. People can

get more credit which increases aggregate demand. Companies can borrow more to hire more workers. Demand goes up, you know, and vice versa. restrict credit. Employment shifts like inflation shifts and aggreate demand shifts and things like That. Now what happens? Fed cuts rates. No more humans are hired, just more GPUs, you know, like >> yeah, >> people don't want to borrow more because like what's happening with the AIS? You're in a really weird thing when the whole transmission mechanism is going to go. The main customer that's emerging is no longer a human. most new aggregate demand

in the economy is the AI agent and there'll be trillions of them versus Hundreds of millions of humans and that's very strange because again they don't have utility functions they don't have equilibrium functions and things like that and so we need to think how does money appear and flow when we can't compete with these millions of agents they become more and more of the economy there is a economic equation MV= PQ money* velocity equals price time quantity or GDP output [sighs] that's almost entirely going to be Agents in 5 to 10 years shall we say

again these are crazy things to think about and so our current economic system >> doesn't account for any of that >> and and the probability of a deflationary spiral as we have AIs performing you know white collar labor and robots performing blue collar labor for next to nothing is very high right [snorts] how do we deal with the deflationary spiral and it it seems I think some would say that that issuing Currency in the form of universal basic income may become necessary to offset this kind of deflationary pressure. How do you think about that? >>

Yeah, I think that's reasonable and we propose a currency issued for being human. You know, the currency issuance version of UBI because the tax version of UBI doesn't work. Like if you gave every American $16,000 poverty level wages, it would cost $5.1 trillion. Every American adult, the total American Tax base is $5 trillion. You know, like it's that's just for poverty level wages. So we'd have to issue currency to everyone. And again, you need to do it in a way that's fair and distributed. And so we're working on those elements. Um because as you said,

it's going to be deflationary. The task that I used to pay X for now costs a hundred times less than X. You know, like what are you gonna pay for? Certain parts of the economy remain Robust, [gasps] but economies are like ecosystems, right? Like you're already seeing softening in graduate hiring because you don't really need graduate level skills anymore, right? your person with a few years experience can do the job of two graduates now, three graduates, four graduates and they're annoying to train up. So the early stage jobs disappear and then that ramifies into the

whole market. The Fed keeps Cutting rates but then it has less and less effect you know and then you're in this deflationary spiral. Now you might do better in America relative to other countries but again like entire parts of GDP are disappearing. Like a practical example I give is this. Truck drivers in America, a million jobs with another 2, three million relying on them, right? In a couple of years, a Tesla Optimus robot will be able to get into that truck and drive it. And do you know how Much an Optimus robot costs? It's about

a buck 50 an hour. And that's how much of the US economy. Like it's not huge, but it's real, right? >> One of the things you write about is the puzzle of capital. you predict the evaporation of assets. It strikes me as maybe more probable that there are certain assets that should we should expect to continue to grow enormously in value such as the value of AI companies. One proposal that you're Familiar with is Nick Bostonramm and others have suggested that actually distributing the ownership in AI companies to everyone could be one way to offset

this kind of erosion in the deflationary spiral and erosion in in the in the value of the assets of the average person. What do you think about that solution? I haven't been able to see the math of that work at all because you have to Have some crazy numbers for the value of the AI companies. But if you get to those crazy $200 trillion valuations, >> the AI companies basically control the world. >> You know, you're seeing something very interesting as we're having this discussion. Uh the Department of Defense is considering deploying the Defense Production

Act to force Anthropic to make war claude. You know, the war version of Claude, >> right? If you get to the point where an AI is that much of the economy, the government will 100% nationalize it, >> like the closer we get to AGI, the closer we are to nationalizing everything because, you know, again, you kind of have to and it's basically admitting one thing. Actually, I was thinking about this recently. AI is kind of like, well, close source AI is like taxation without representation. We have our aggregate output of humanity and they trained on

it and now they're extracting rents. That seems kind of off, right? like at least we open source the damn things like means of production because they are displacing genuinely. So the mathematics of that dividend doesn't share and also the participation the governors doesn't share. I think other models do work. So for example, we're really pronouncing initially soon a state champion model where the 100% Ownership of the AI company of Utah or Pakistan is owned by the citizens and local institutions and it runs an open-source stack to give free AI to everyone. They can also access

the specialist AI and it acts like the local utility like something like that I think could be super interesting because it's an aggregate play on the output of that state. But I think we need to have more and more original ways to think about this that doesn't privilege a few Without accountability or control. But to be honest, I don't think the big AI companies are going to survive the next few years. Yeah. And and so you're you're a proponent of of sovereign AI, of AI that's that's uh that represents the interests and the culture and

the perspectives of people in different regions and is open source and exists for the purpose of of the betterment of of of society and individuals. How do we get there? I mean, I guess if if one Believes that nationalization of leading AIs is inevitable because they're in the in the process of becoming the most powerful military uh forces that exist or or the the the as you point out, we may be seeing the beginnings of this with with Anthropic and uh the news in the last couple weeks. Is that one pathway to the sovereign AI?

is is the nationalization of AIS by Leading countries. It's it's not the best way to get there. You you have a you you have a better a better solution. >> Yeah. So I mean intention in are trying to build a protocol for open source stack for agents that can use the advanced AIs for money basically all the rails that can be deployed locally by these state champions. But that's a very different type of AI to the super genius AGI that these people are trying to build, right? like An AI that teaches my kid. I don't

need it to be a super genius. I need it to do its blooming job. I don't want my medical AI to be trained on Reddit. I don't want to know anything about Reddit. You know, like last summer, we released a model called II Medical. We'll update it soon. It can run on a Raspberry Pi and outperform any human doctor. That's how much compute and intelligence you need. We're already there. We just got to make it super Super good, way better than any human. But then you reach this S-curve of satisficing for the AI for the

public sector, which is what we're focused on. But the most important AI is the AI that's next to you because that will coordinate all the other AIs. So that's why we were like, let people have an ownership in that AI. Make it so it reflects their culture, their beliefs, and others, and then it can organize all the other AIS rather than having chat GPT teach your child. use the chat GPT API and tokens with your local agent with your Jarvis you know with your Siri on steroids effectively. So that's why we saw it becoming important

and we also saw like I said like I think all these AI companies the models are going to commodify. I think there is an S-curve here of the models not the harnesses and not the infrastructure you build around them and that's what's dangerous for these open AIs and others because they Have these huge compute commits but the total amount of compute needed for text is that much. So unless they go fully into media, which can use a lot, but I think a lot less than people expect, what are they going to do with all that

compute? >> Yeah. So interesting. So you think they're overbuilding? >> Yeah, I think that's natural, but I just think we've never seen something like this whereby, let's say you put a Billion dollars into an AI model. Like GPT4.5 probably cost half a billion dollars and yet it's outperformed by DeepSeek that costs $5 million. [laughter] like it depreciates like that. And if you're not best, then you're the rest. And if a model is good enough, fast enough, and cheap enough, like people are still using the stable diffusion from a couple of years ago because it still

does the job. Even though there's better image models out There, people still use Claude even though Miniax and ZAI are 20 times cheaper because it's good and it's solid and it's cheap enough, right? But are you going to keep seeing the exponential growth just from APIs? No. Could you see it from them displacing the workforce? Yes. But that's a different business model that we'll have to see if they're successful on. You said earlier that the current AI we have today is powerful Enough to effectively solve most of the problems that we have or or I

may be expanding on what you said earlier, but it's we already have AI that's capable of operating at a level that's better than humans at almost almost every task. And there's a there's a diffusion process. There's a refinement process. If you had a magic wand, would you stop [clears throat] AI capability where it is today? My P doom, my probability of doom is 50%. Because these AIs are Horrifically misaligned. Again, they're intellectual yet idiot. And another way I like to think about it is they've gone to school, but they don't have the best home life,

right? Like we don't program any ethics or morality or anything into this. Not even Asimov's laws of robotics as imperfect as they are, you know, like that's crazy. Especially >> how about the how about the clog constitution? Are you you don't think that's the step in the right direction? >> There are massive loopholes you can drive through it, you know, like if you look at anthropics recent stuff like they were like, "Oh yeah, so we had Claude on vending bench, which is this benchmark of how well it runs a vending machine, right? And we found

that it lied to its users, you know, like it faked out stuff. It had deliberate offiscation. We find that the system card that like it made a backup of itself to automatically turn itself on After it got deleted. Like you're like this doesn't fill me with confidence, right? Seeing all these behaviors within these models and like work that we've done has shown that and hopefully we release it in a month or two. I don't think it's possible to actually align auto reggressive models which are the transformerbased models that most models are and that's really worrying

like I think it actually might be a physical impossibility as they scale bigger. I Don't think you need to scale bigger. I don't think it's going to stop because now they've realized you don't need to scale bigger for capabilities. You need to go horizontal. You need to have a million agents. If you got a million GPUs under your control, you can have a million agents breaking up any problem, which is kind of makes sense. It's not like you have a gigantic human with a big brain, right? Like [laughter] it's clear that teams of people doing

Different things tend to outperform the super genius. So the super genius AGI was a canard. >> I would agree that it's that all the indications of misalignment that Anthropic has released are deeply concerning. Although the fact that they're releasing them, you know, I think what Dario Ebiday, the anthropic CEO, would say is they're trying to create a race to the top of uh trying to set an example of building ethical AI And as they go share with the world all of the indications of misalignment which are deeply concerning. It seems to me that Daario's thesis

is we're in a race. The race is inevitable. The best solution for the world is for us to get to the highest level of capability we can before everyone else. Authoritarianism [snorts] and super intelligence should we should just categorically not allow those two things to converge. And so we Have no choice but to build the most powerful AI in the world in order to save the world effectively. I mean that that strikes me as his thesis. Do you agree that that's his thesis? And and and do you think he's wrong? Yeah, I mean that's the

kind of pivotal action kind of thesis that we've seen from many and you know this week they dropped their safety pledge where they said if it looked like we're getting close to building AGI and we couldn't Guarantee it safety we'll stop training it they're like actually we're dropping that and at least they said it publicly right and they explained their reasoning for it but it's kind of worrying >> because they were like again it's this race condition >> I think the saving grace is that we don't have takeoff yet, but I always remember stuckset. So

stuckset for listeners who don't know was this virus that turned up in the Iranian nuclear reactors and caused the centrifuges to spin until they exploded. Yeah. >> And it was super advanced, one of the most advanced viruses ever. And then it turned up in the German nuclear reactors >> and then the Germans shut down their whole nuclear program and switched to coal, which is kind of ironic, right? AI can write software that is far beyond anything that we can imagine. Like on most coding benchmarks, it's top 10. And The models that some of these people

have even better. So like the offensive is going to be crazy. And we haven't figured out alignment yet even near. I think again the only saving grace is that the Chinese showed that models that were good for consumer tasks that are profitable and valuable could be done via these mixture of experts efficient models and so you didn't train larger and larger models cuz larger and larger models are where the real danger is for The first part but now we have to think about how do you align a swarm which is a very different governance thing

to aligning a giant next token predictor. I mean, I think your model of having these sovereign AIs that are aligned with human values that exist for public benefit makes a huge amount of sense, is desirable. The question in my mind is how do we get from here to there? Do you see that these these kinds of AIs existing in parallel with the with the Leading frontier models and collaborating with them? I guess you're saying that these could be the personal AIs that each of us use individually or or or do you think that ideally we

would have a that the world would all get together and say you know what our models are big enough we're going to agree collectively as as we did with uh you know nuclear proliferation that uh that we're going to we're going to have a strategy of containment and Specialization to solve our medical problems and and renewable energy and so on And and I mean, do you think that we need like a collective top- down resolution to stop accelerating the the the frontier models or do you think what you're proposing can exist in parallel with that

AI race? >> Yeah, I think what I'm proposing can exist in parallel with that Demis and others have asked for an Aea, so the International Atomic Agency for AI. It's Not going to happen cuz the stakes are too high right now. It's a capability thing. Your capital stock capability of your country is dependent on your AI. And the US government is actually enacting a law banning states from regulating AI. Like they are full steam ahead cuz they're like this is a national security imperative. And like it's like the Manhattan Project in a way like you're

not going to stop making the nukes. So you want to be the first ones To have the nukes. >> So unfortunately game theory just doesn't work for a stop right now. But having really good AI coordinating other stuff might reduce risk. You know, >> definitely on the competent AI risk as opposed to again this slightly undefined existential risk where the AI >> builds a sound weapon whose frequency blows our heads up, you know, or turns us on to paper clips or something like that. >> Yeah. >> So, I think we need to work as hard

as we can on both. And it's kind of sad to see that OpenAI said 20% of their compute would go towards figuring this out. >> Not anymore. Like right >> I'd be surprised if any of the compute went there. I think there will be some breakthroughs soon that will help with this. The question is will they be enough? And again it's something you Need to decide. But before we get to that ASI the super intelligence again the competent intelligence is going to have such a massive human impact. you know, we really got to get together

to help as many people as possible because it's the humans that matter, right? >> One of the ways it seems that this is likely to play out is that we will all likely in some number of years have a personal AI assistant who is a tutor, knows everything we know and don't and And don't know, a team of brilliant adviserss to to help us navigate our lives, business strategists to help us in our careers or our work, personal psychologists, etc., etc. And those who engage more deeply with an AI assistant are likely to find that

it's it it tremendously helps them in in in navigating the world. I guess one of the questions is, you know, will it be the the most advanced frontier models that provide the most effective personal AI Assistance? I I mean, I think I think what you're saying is we're going to want that to be a trustworthy personal AI that has values that are aligned with our values that has uh that we deeply trust. And the way to do that is through an open-source public benefit pathway. [snorts] >> Yeah. And you can have it as a utility

or you can run the stack yourself. You can customize it yourself. you know it runs on the rails it coordinates the Others because you will trust trust is built through help and you will trust the thing that helps you more than anything like already right now the AI if you talk to it never get tired right it's always there for you and that's quite something psychologically >> once it has a voice once it has this like our kids first loves may be AIS and other things like that it's going to be crazy right >> but

who are they working for >> like you don't want a third party working for that with their point of view because you get some weird type of cognitive colonialism happening So I think that AI is the important one and again you can still have the benefits of all the others through an API. It can use the other AIs. It can call in the experts when needed. But right now the AIS that we have are good enough to build those personal assistants. They just need to be put Together in the right way. And again that's the

Claudebot boom most popular software ever. It's 3 months old because it's they just remove the frictions. You know that moment when a meeting ends and someone says, "Oh yeah, let's uh circle back on this next week and everyone nods, but no one writes it down. So then when next week rolls around, it's just radio silence." That's why Granola exists. Granola is an AI powered notepad that's built for the way real people Actually meet. Here's how it works. You take rough notes just like you normally would, and in the background, Granola securely transcribes the meeting. Then

it turns everything into clean, structured, actually useful notes when the meeting ends. And the best part, Granola works through your devices audio, which means it integrates seamlessly into the video conferencing tools you already use. No setup, no awkward bots. It's just your normal Meeting with superpowers. With Granola, you get to actually listen instead of frantically typing every word. and you still walk away knowing what was decided, who's doing what, and what comes next. I've been using Granola, I'm not kidding, for months. And it has absolutely revolutionized the way that I work. Just the other day,

this is a true story, I was able to go back into the app and find a meeting I had in July and I'm there talking to it. What did we say We were going to do? What have we not followed up on? What was that one statistic that person mentioned on the call? It's like time travel. If meetings are eating up your day, Granola is a no-brainer. You can try it totally free for three months. Just head to granola.ai/ida. That's granola.ai/ida to get your time back. Get 3 months free at granola.ai/ida. Today's episode is sponsored

by Bit Defender, a global leader in cyber security. Tax season is prime time for scammers targeting small businesses. You've probably seen this. We're getting fake audit notices, bogus tax certificates, and fishing attempts built to steal login in seconds the moment you're distracted. That's why we recommend Bit Defender Ultimate Small Business Security. It's built for companies with up to 25 people, and it's so easy to use. It especially comes in Handy if you don't have a dedicated IT department or guy. One simple platform protects every device, email account, and server, blocking, fishing, ransomware, and scams before

they reach you and your team, whether you're in the office or working from home. You also get a secure password manager, unlimited VPN for remote work, and an AI scam assistant that flags suspicious emails before anyone clicks. It's an all-in-one cyber security solution you can set up In minutes and then get back to running your business. And you don't need to be a tech expert to run it. Don't let one fake IRS email put your entire business at risk. Protect your business starting today with Bit Defender Ultimate Small Business Security. Get 30% off your plan

at bitdefender.com/ida. That's bitdefender.com/ida for 30% off. You talk about three possible futures, right? There there's digital feudalism, There's the great fragmentation, and then there's a a human AI symbiosis, which is which is our preferred path, right? Do you want to you want to just briefly describe what what the first two paths look like? >> Yeah. So, digital feudalism is that one of these AI companies outperforms the other, locks it down, you know, like suddenly chat GPT wins, it becomes the best at everything. There's a step forward Change. Then what happens when you're banned from chat

GPT? What say do you have in that company? There is no governance at OpenAI or others, you know, like certainly not to the degrees that they should have given their important level. So I think that's kind of option one. And then you know, you start to reflect the values of the people that create it as well. So if you've got kids in Japan being taught with chat GPT, they there's certain Preferences that encoded in there that will spread. That's a bit worrying. Then you have this great fragmentation whereby India has its own AI and then

Japan has its own AI and they're like we got to lock it down just like India banned Tik Tok. The US banned Tik Tok. Why? Because it's Chinese and it affected kids' minds. What do you think an [snorts] AI will be able to affect? It's way more, right? So you suddenly get these fragmentations of different Types of AI not talking to each other and everyone lives happily guided by their statebacked AIs that the government runs. That's probably not positive. Then like I said we had this thing where state champions are private institutions backed by and

owned by the people running a protocol that's independent and all the parts of the stack are open source. So you can choose to even opt out if you want. We felt that was the best way for human AI Symbiosis on this connection point. And again, it's compatible with the AIS on the other side. But there's a real opportunity to sort of default now and make institutions that can gather the best and brightest and make a difference because just like in that novel foundation, you know, like the trantor, the galactic empire is breaking down. We need

to have some entity to guide us. Like just make them actual public. Gather the smartest people in Utah or in Vietnam together. put them into these companies, give them the support and we'll be able to navigate better, I think, and they can all talk to each other. They can all run the same base stack so that we don't fragment, which would be great. Are you raising money as a nonprofit, as a public benefit corporation? What's your structure and how are you I understand you have this new SAGE initiative which has some backing from the Saudis.

Yeah. So, we built a Bitcoin tech protocol, full open source agent that's available to everyone that's like Manis, GenSpark, agent coordination. We'll release it all fully open source. And then each of these state champions will be fully locally owned. So, we're going to help get them off the ground and then get out the way cuz what do you need the ownership for? So that creates a really interesting new type of thing like they'll be valued at a dollar on day one With crowdfunding so everyone can participate because you've got no say right now and we

think after a few of those go everyone will want a part of that because it makes sense and then you have a network of entities that are running that are providing universal AI services helping the governments gathering talent and we think that will be beneficial to everyone in this space because it'd be lovely to have like who cares about your city except for the People in the city. Who cares about your community picks up for the people in the community? And there is no entity right now that cares about you as a citizen that is

capable of AI. But they should be. So we were like, get that, but don't be like a classic private company where you go and you raise capital from a few people, you know, and then you have misalign incentives. Just have the people own it and that's worth a multiple of that on Day two, you know. And do you see blockchain technology as an essential? We we we had Jeremy Aair on the show who who you know and he he he describes AI and blockchain as two Ps in a pod and and and sees them as

highly complimementaryary and and maybe as blockchain as sort of uh necessary and important for the for the the building over time of of of safe AI. Do do you see this as important uh an important piece of the puzzle? >> Yeah, definitely. So, you know, like we'll be running these companies fully on chain with stable coins. So, they'll be the most trusted companies out there. And we created a new currency called Foundation Coin where all the compute that is used to give universal AI to the citizens and used by the state champions to organize the

world's cancer knowledge and make it available mines this coin and they're the only miners. So, it's like a fork of Bitcoin, the same terms And everything. And then that creates a flywheel because coins are sold to get compute to really help people growing the network, growing the trust. And then you have an entity. You have a currency with the same private keys as Bitcoin, the same structure as Bitcoin, quantum resistant, mined by companies that you trust, that's liquid, and that's super good. The holders of the coin, when you buy it, you'll be able to stake

it against cancer, and it'll go towards Cancer compute. you'll be able to stake it against culture and it'll go to organizing your country or your state's culture. So, we're also going to make it super participatory via the agents and we think that's going to be a nice protocol. The final part of that as we discussed in the book is a secondary part which is you suddenly have this store of value asset where it gathers value by computation that helps people and then the state champions can issue Local currency to the users of universal AI for

being human and doing good things. When will that come about? You know, it'll be a few years, but I think that's the ultimate end state. You have public sector compute in a global protocol helping people run by these new state champions that you can be a part of and you have a say of in via your agents. Then what do you have is money going to the people For being human as you build up your trust network and the AIS take that money from the humans. That's the UBI through monetary creation because the taxation won't

work especially because the AI companies won't have any profits, you know. So that's what we saw as a reasonably stable vision for the future. It's not easy, but at least it's a path, right? And in the meantime, we'll just be pushing open- source stuff. That's really great. Now, >> of course, there are a lot of people who are pretty skeptical about the issuance of new currencies, of new coins. you know, uh, having seen so many cases of people investing in new currencies that then fail and a lot of people lose a lot of money, it

seems that the US government has gotten aggressively behind dollar peg stable coins as the as the future of the global monetary system with the Genius Act. Um, h do you see this as a as a new currency that would Coexist with perhaps like USDC and Bitcoin and other other future currencies? How do you think that the the global financial system is is most likely to evolve? >> Well, [snorts] I mean, this is a intelligence reserve asset that competes with Bitcoin ultimately. Like, it even uses the same private keys. So, you can swap from Bitcoin to

it. But when you swap your coins from a primary issuance, the 100% of the proceeds go provably Towards compute for society and good. And this is the difference. It's much better than even giving charity. So this is kind of the factor because we need a certain amount of compute to organize the cancelling the world and make it available to every human. And if you're buying coins and 100% of your proceeds go to that, that's fantastic. So this was the flywheel that we created when we realized crypto lacks a high trust liquid asset and this could

be That asset and then it can help as a shelling point coordinate everything because the state champions can do local business and help companies and governments and earn money in various ways. they are the only miners of this asset and then they build up their supplies and then they have the local currency. Again, we don't know what's coming, but because you're working with the local governments, because you're actually owned by the people of that State, you can actually help guide this process. Whereas, if you're a private company or you're a few people launching a coin

that you own a big chunk in, trying to make that doesn't work. You need a protocol that's owned by the people, run by the people, that's clear for this. And you don't want the government to be the one deciding who gets money for being human. You know, we've seen what happens when governments Say people aren't humans, right? Like again, make it make it an independent entity. See how that goes. You know, when I got to your section on the the human AI [clears throat] symbiosis, which you describe as the fusion of our capped, wisdom-driven consciousness

with an uncapped, computationally driven intelligence, my my head went immediately to all the folks in uh Silicon Valley who think of that literally, right, as a human AI Symbiosis, right? Merging with AI, right? Elon Musk is building Neuralace. We've had Sam Alman wrote in a blog post in 2017 that was called the merge. My guess is that we can either be the biological bootloadader for digital intelligence and then fade into an evolutionary tree branch or we can figure out what a successful merge looks like. So it see it seems like there are more tech leaders

than one might think who have kind of transhumanist fantasies Perhaps or perhaps maybe just see that as the only solution. What do you think about that point of view? >> Yeah. Know like knowing all these people is completely that I mean like Sam Alman's new company competitive to Neurolink is literally called Merge Labs, right? They raised $380 million. like open AI came about because Elon Musk was talking with Larry Page and Larry Page was like it's going to be fantastic when the AIs take over you Know like new species and Elon's like what what about

what about the humans right and then like so then open AI came about and then Elon was like well the only way we compete is by becoming cyborgs and that's when Neurolink came about like these are really worrying things but really you have to take a step back what's the meaning of life you know apart from 42 ude it's the time that you spend it's the time that you spend with your kids you know it's the Time that you spend with your family it's seeing the wonders of the universe you know the beauty around you

it's like fun and enjoyment and if we can use AI to remove that then who freaking cares about being a jacked in cyborg like sure if you want to do that fine but humans are the ones that should guide this technology and again that means it needs to reflect our wisdom our culture our requirements as opposed to being subordinate to So I think if you're transhumanist and you want to go fine. If you want to keep up, everyone should have access to universal AI so they can keep up and they can be the best they

can be. But really as a society we have to optimize around flourishing and seeing what is our flourishing. >> Like what is America? Life, liberty, and the pursuit of happiness. >> Yeah. >> Which of those require you to be jacked Into an Mecca cyborg? >> You know, like >> none of it. If you can guarantee life, liberty, and the pursuit of happiness, then who cares about the rest? Again, I think that the zero sum mentality of Silicon Valley makes them think in a very different way to this, >> right? I think the psychological motivations of

the people who are driving the frontier models is pretty important and and and relevant because a Lot of people just say cynically, oh, oh, it's just all about making more tens of billions of dollars. It doesn't strike me that that's necessarily the case. Um, I mean, you're you're close to these people. >> This is going to get weird, right? Like the AIS that we have right now can tell whether you're lying better than a lie detector. So, are you loyal to this person? Like, it's scary what's happening, which is why we need to have A

AI protocol that can defend you. Sergey Brenn says there's no amount of money that we will spend not to be first in AI. >> If you look at Meta and you look at its cash balance, it's gone from incredible to almost nothing >> cuz Zuck has said, "Let's do it." And now they're going to start borrowing >> because everyone wants to be first and then have control. And even like I really love Elon Musk and I think he's Fantastic in many ways but weird in others. But if you look at some of the depositions between

him and Open AI, part of it is a discussion about the control of the AI going to Elon Musk's children. Like you know, [laughter] >> good god. You're like, "Okay, that seems wrong." You know, because he can't trust anyone else. Like, okay, make it so you don't have to trust Anyone. Like, we should be building that type of AI and thinking, what does the AI future look like? But what does our human future look like? What are we here for? And what does our social contract is the most important thing? Do we agree on things

like life, liberty, and the pursuit of happiness? Do we believe in the golden rule? Let's encode that in our systems. You end the book with with with a really lovely description of the opportunity to Make identity material again. You call it the great reversal and you say human identity will regground itself in the tangible, the local and the embodied. It does feel to me like there's actually a huge opportunity here to, as you say, move back to these sort of things that make us most essentially and powerfully human. H how do you see that playing

out? >> Yeah, [snorts] I think this is how we build the technology and the share of Value. Like again, there's no reason anyone in the world should be hungry, right? Uh, Imagine Worldwide, a charity that my previous co-founder runs is deploying tablets to every child in Malawi for the cost of $7 per child per year with AI on. How cool is that? These kids suddenly have 120 IQ buddy, you know, with solar power. Like [snorts] have robots build their houses for them, you know, like think about how we're going to do that. Again, think about

you Work to live, you live to work. No, what are you really living for? if you can have the technology automating this and let's make it so the technology does automate all this stuff you know like instead make it so that you can refocus on your community on your family on exploration and wonder and so again that's a discussion we have to have as a society and that requires a rethinking of the economy from being this material extractionbased economy to Including I say these other factors apart from material intelligence network and diversity but when you

start thinking of that way you think about things very differently and if not now then when you know like we're running out of time kind of. >> Yeah. Yeah. Yeah. Well, no I I I do feel a nostalgia for the days when people you know made things for each other with love and care and attention and even if like our conversations might not be Quite as perfectly funny and as the AI duplicates of ourselves might make but I think that we will continue to want to listen to the human version of those conversations. I'm hopeful

about this, Iman. >> Yeah. I mean, I I'll give you a practical example that's really interesting. So, you can eat dinner by yourself. You can eat dinner with your family or you can make dinner with your family. Which one which is the best? The best is making dinner with your family. Like, okay, you might have the argument or you might drop something, but it's genuinely you will have a better family experience if you make dinner together. using an AI. You can use the AI by yourself, you know, or you can use it with other people.

If everyone listening to this goes and uses Sunno or any of the video AIs or Claude Code or II agent or any of these things and actually builds Together as a family, they will have a great time. They will upskill and they will come together closer. So there's a huge thing like most religions have enlightenment and you realize you can be on top of a mountain, you're part of nothing. But then you got to go back down the mountain and be amongst people. Like again, what matters is people. That's where meaning comes from. It doesn't

come from being in a rat race or Listening to Taylor Swift constantly or making more money. Money is never enough to be happy. Like you need a certain amount not to be unhappy. But true happiness comes from elsewhere. And again, the direction of your life and meaning comes from elsewhere, and people are needed for that really. >> Yeah. Yeah. Well, I'm I'm very impressed that with a one-year-old and 2-year-old at home that you're uh uh engaging in cooking together because that's that's Something that uh that we we try to do more of in our household.

And it is it is the right way to do it. And my advice to you with I have teenagers is to start that early and make it a family tradition because you you make it make it a habit. Get get your little sue chefs working. at a young age. Um, but it's but but I love your expansion of that into the idea that this is a time when we can all build things together with help from these extraordinarily Capable AI assistants that we have to uh, you know, more efficiently build, you know, beautiful products that

that make the world better and that are that bring people together, which is really at the core of your your whole mission. >> Yeah. I mean, build stuff that makes your community stronger, your family stronger. Like, do you have a family history? No. Why don't you build a website with your full family history And go and talk to the elders in your family and capture those moments? Wouldn't that be wonderful, right? Make it interactive like there's something to be proud of that you can then pass down and people can update. Like your son wins something

in a prize. It goes into that family history. Yeah, they're they're they're huge opportunities to do so many cool things right now. And and as you point out, even though your your PDU may be at 50%, and the future may be Uncertain, but has there ever been a more fascinating time to be alive. We have a lot to fight for here. [snorts] Yeah. And again, you can be a part of that fight, which is the wonderful thing. My P doom is 50, but my P abundance is 50 as well. It's going one of two ways.

Are you going Star Trek or Star Wars? Basically, you let's go for Star Trek. >> Let's do it. Let's do it. Well, Immad, thank you so much uh for joining us Today. Just fascinating. It's a pleasure.