Emad Mostaque: How AI Will End Capitalism (Before 2030)

Transcript

Because if you don't own your AI, it'll end up owning you because it's more convincing than you are. It's a better writer than you are. You'll trust it more than anything. My fear is that in AI, there are a lot of easy, convenient solutions that prevent us from getting to, you know, that core idea. Most people hold their crypto in a centralized exchange and they don't know how to read a block explorer. [Music] All righty, guys. Welcome back to episode number two of AI Supercycle. We are joined by Immod, the founder of Stability AI, currently building out an awesome product in intelligent internet.

It's Andy. I'm here with Robbie. AI Supercycle is presented by Near, one of the leading AI projects in crypto for blockchain, if not the leader uh building out the decentralized AI vision, focused on privacy, focus on user ownership, uh userowned AI, and really just this open-source nature of this industry. Very very important discussion to be had today on this on this subject with Immod. Immod, welcome back to the show. Great to have you. Thanks for having me.

Yeah, man. I mean, look, it's been what, five, six months, Rob, maybe since we last spoke. I'd love just a quick update, man. Like, we we were speaking about intelligent internet. Uh, you know, kind of what your vision was with regards to democratizing access for things like education, for things like healthcare, these types of things when it comes to open source AI and the importance of kind of the organic nature of certain models, of certain training data. You know, it's been quite a while. I'm sure you guys have shipped quite a bit. So maybe just another quick intro as to what intelligent internet is and just get us caught up on all the progress man.

Yeah, I think AI time is like faster than even crypto time. It's like 10 times normal time. So it's been like, you know, three or four years since we last talked. Um yeah, so the the goal of intelligent internet was basically to increase the intelligence of humanity by giving everyone access to high quality AI for the stuff that matters. Because right now we're seeing a lot of AI for, you know, entertainment, chat bots, businesses, and others. But who's going to make the AI that teaches your kids? Or Albania's just had an AI minister. What's she programmed on, right? You know, who's going to build the AI that does your healthcare? So, we saw an opportunity in that cuz of this trillion dollars of AI spending, at least 20% is going to go into public sector type

stuff. So, we were like, why don't we build that really great first? This stuff that's based on common knowledge. other people can build the other types of AI but let's make it so that more of this compute goes towards organizing cancer knowledge and making it available to people organizing global regulation or digital assets to allow people to encompass them you know giving AI to teach every kid in the world so they have the same access so we brought over kind of a core team from stability AI my previous company we had like 300 million AI downloads and over the last few months we've been building a state-of-the-art full stack solution So, II agent is a state-of-the-art agent that can build websites, presentations, everything fully open source. Common

ground is a multi- aent software that does everything from planning to execution of really complicated workflows. And then we've been releasing models and data sets as well. So, we released um for example, I medical 8b. It's an 8 billion parameter model that fits on your laptop and outperforms any human doctor. And so we're going to make that better and better with the goal of next year everyone will have education, health, government, full stack AI solutions that they can just take and run on near, run on AWS, run on their own computers to really make that difference. The final bit was tying it all together. We think we're at this takeoff point now where these AI agents

can basically displace most of human cognitive labor. In fact, cognitive labor is going negative. So we just released our book um the last economy about what the economic system looks like going forward uh proposing one backed by AI using AI itself and we're about to release the academic papers behind that as well which has been super fascinating exciting and a little bit scary. Wow. I think there's so much to unpack. Um you've been a visionary in the space for so long and so you you've really outlined your vision and now you're you're working to make it a reality. One of the thing that strikes me is the composability of this. You mentioned how you can run these agents, these models.

You can run them on near, you can run them on another blockchain. You can you can start to tweak and tailor these agents to your particular liking based on several factors that that maybe we'll get into. So when you were building this leveraging some of the experience and stability, what were some of the things that you that were non-negotiables? You said, you know, despite, you know, how others are building it, these are ideation on, we're staking our company on and and effectively the future of of this particular model. What were some of those non-negotiables that you implemented?

You know, could you explain those to us? Yeah. So, it was a really interesting experience. Uh, it's we released stable diffusion at stability in August of 2022. So, that's three years ago. You know, it's been like a thousand days since chat GPT now. It's been like a year since 01. This stuff moves fast. And when we released it, we're like, it makes pretty pictures. And then we released the video models, I think the year after, and now look where the video models are. We looked at the pace of all this. We were like, we released the first version of stable diffusion open. The final version, stable diffusion 3 before I left was kind of this open up to a million bucks type thing. And that felt kind of wrong, you know, but it was like how do you

build a business model around these? That was the question. And Vitalic Bhutan has this great essay called the revenue evil curve. You know, it's like you start out like really good, but then when you're building infrastructure, you go rivalous and exclusive. you restrict access or you compete against other people and then you turn a bit evil and then companies turn more and more evil because they're kind of dumb AIs. So the first thing was we're like education, government, finance like but the regulated industry stuff that's based on common knowledge doesn't require any IP or private knowledge that should be free and available to everyone. should be MIT. It should be proper infrastructure.

And again, a big part of our global economy is that type of stuff. And in the AI age, it should be that kind of stuff. The next thing is the composability. Like when we really stable diffusion, again, we were like, pretty pictures. And then people were like, well, if I say a woman, it doesn't give me a woman or a man in my culture. So we released Japanese stable diffusion where you know you type in salary man and instead of having a very happy man you had a very sad man you know like it has all the cultural nuances. We're like how do we make that composable and build these data sets because the ways models are trained is actually known as curriculum learning. You have generalized data then more and more specific data.

So we were like you got to have those. And then the interesting final one was it was kind of non-negotiable to have it really directed built in terms of a lot of the projects and things here are like let's have an emergent architecture something like a bit tensor right like this stuff will just emerge and it's a marketplace of ideas we were like no we should probably have like really dedicated core teams that build the best autism stack or cancer stack in the world and then drive it forward like Linux is driven forward by a core team Bitcoin core is driven forward by a core team. Like again, you can use the emergent stuff, but someone's just got to build this and it's as hard as building this properly is probably

launching a rocket and landing it by itself again. So you got to have the best kind of people doing that. So fully open source, composable, and then directed I think were our three non-negotiables on here. And that's again how we've been directed because we know what we want to build. In fact, one of the other goals I told the team is that in a few years, we want it to be built. Like most software will evolve forever, but again, we've seen the power of something like a Bitcoin where the code is relatively the same. Again, you have to do updates. Once you get a good enough stack for medicine, you can have that on every phone and every laptop and every doctor's appointment in the world checking every diagnosis. And sure the data sets will

upgrade but you don't really need to upgrade the software. So that was kind of very interesting thing in solving for a problem versus this continuous development type of approach. Could you just maybe just succinctly put you know what are the goalposts because oftent times like you're saying it it constantly evolves and maybe that is you know it is the goal in some some cases but for you it's it's more strict than that and it's more direct. Yeah, it's basically for these things that I say matter as it were like again the stuff that should be a human right this universal AI is what we call it having a companion that can give you that at the highest reasonable level so the best doctor from any school the best

teacher from any school and then we have these things called these um anchor sets which is our collective knowledge on that again they'll update every so often but no one's got a really clean data set that can teach an education model to then teach your kids as good as possible and that's an S-curve versus AGI is an exponential. You'll never have enough data for an AGI but we've got various measures and evals about what good enough is for medicine now. So again we can say our medical model is better than any top level doctor.

Is it good enough? No. But we can see things like error rates and other things. And really AI does need to be quite a bit better. Like uh we did that's one interesting stat like you know Whimo. Yeah. If every car drove at the same safety profile as Whimo, it would save 40,000 lives in America a year and a trillion dollars of social impact. Wow. Wow. That was always the the inflection point is when it was provably more safe to self drive. So even though our medical model is better, it's not good enough for us. You know, like again, we have certain metrics that we're sending and we're going to be sharing on the evals.

Once it gets to a certain level of accuracy of performance and others, then we're like, this is 99% of the way there. And sure, we'll continue the data sets will continue to improve. We'll see your feedback, but we solve for a problem of everyone has very high quality AI that's basically free for them to access universally. for this sector, for this sector, for this sector, for this sector. And again, that's a very different problem set to most other AI companies and organizations. Yeah. And and and I think that um what you mentioned with regards the revenue evil kind of curve that that is uh playing in that's playing out in a large way in the traditional AI space currently. Uh there's just there's just

very fierce competition as well as luckily there is what appears to be infinite funding from Silicon Valley. So, like we're still not really there yet because everyone's in a happy golucky state because well, Seoia will give you $50 million, you know, it's not a big deal, right? Etc. But you know with regards to actually specializing into these um you know into these exact goalposts or goal that you have you know you present a compelling argument for a lot of people listening but you also present a a a scary whatif or a scary kind of second or third order consequence which it seems you've deeply understood and thus wrote about in your recent book more around the future of economics in an AI powered world. And so

while I think a lot of people will take what you're saying and fear for their jobs or fear for kind of there being a robot doctor instead of the emotional connection that you would get with your pediatrician, you know, I do feel that uh there is a you know, you have a right to kind of explain why AI and AGI for the future of the economics of the world isn't perhaps so bleak as some of the doomsayers or you know overexaggerations that we've heard and will continue to here. And so, Ahmad, I'm I'm curious when you're thinking about the relationship between, you know, specializing medical and legal and health and uh education to the highest tier of specialization and skill at the

absolute lowest barrier to obtain that knowledge. What does that do for the economics that exists in the world today? Maybe you could unpack a bit of the theme of this book and kind of we we can work through, you know, the economics of this AI system specifically as it pertains to uh, you know, an open-source system and a kind of a userowned economic system. Maybe you could walk us through some of the book themes and then we could, you know, dive deeper on them. Yeah. So, in the book kind of we discuss how different types of economics looked at different parts of the picture. It's like the blind scholars and the elephant like you hold the tusk it's a spear you hold the tail it's a mop you know you

hold the trunk it's a hose etc like everyone looked at different parts of the picture in fact there are different types of economic flow not just GDP for example which Stan Kursnets who came up with it says is a terrible way of measuring the economy especially on a social welfare basis you know like when cancer is solved it will be bad for GDP short term good longer term but very bad short term because all those profits disappear from the cancer industry etc. Right? Divorce again tends to be good for GDP. It measures one type of thing which is a gradient flow and that is a viral risk flow. I give you an apple, I have one apple less. We share knowledge on this podcast. It increases. It doesn't disappear anywhere. And more and

more of the economy has become these circular flows, these non-rivalous flows. And in an age of AI, intelligence is basically infinite. spinning up another agent doesn't really take away from anyone and that causes this imbalance. I call it the metabolic rift whereby agent GI agents don't need to eat, sleep, poop, anything, right? And they can scale almost infinitely because you can have more and more of them and they will learn from all the mistakes. So this leads us to a very strange scenario whereby the AIs that teach us every single day, they're run by slower AIs, these organizations. Do you want chat teaching your kids or running your government? No. Because there's this misalignment there. Can you have it as a public good and it feels like a utility

electricity? This was really fascinating for us because we were like, how does money work right now when labor capital needs labor? You need to hire people to do stuff. In the future, agents will be there. Instead, we think about the flow of money and where it comes from. And that's why we designed this new economic system here whereby the first step is there's trillion dollars of compute coming. You know like OpenAI is going to spend as much on inference this year as the whole Bitcoin budget and 20 gaw uh 10 gawatt they just announced that's about half the energy of Bitcoin in just their new kind of plants right they'll catch up and you know they'll overtake like these are the orders of magnity we're talking about.

We were like what if money was made rather than from using energy like Bitcoin or from debt like the conventional system from benefiting people and helping people because right now we looked at digital assets and we were like there are utility coins there are governance coins there are all these there's bitcoin on money no one's really thinking about the nature of money and what it looks like and if we can have it based on compute for universal AI access organizing cancer knowledge things like that secure against that computation that could be a better basis for the economy especially because as we move forwards and we have other parts of the monetary system what does it mean to

be a human when you're negative on your contribution to intellectual labor like you're the dumbest person on the team surrounded by AIS what it means to be human I think is about giving a damn like these AIs they have no skin in the game they'll come into your workforce but they don't care about your company they don't care about your community. You can kind of program them that way. But what it means to be a teacher is giving a damn about your students and engaging them. What it means to be a doctor is about engaging with your patients and helping them. You know, what it means about using AI within your organization is understanding your organization and being able to get everyone on board to upgrade them and guide them. And so giving a dam this trust element this connection element is

captured in our proposal for how to look at the economy and this comes from the math that we figured out which is you have this material aspect which is you have an apple I take an apple classical GDP then you have intelligence I so we have M then I I is again this circular flow increases as you share knowledge and you increase your capabilities then you have N which is your network your connection to other people. And if you look at crypto for example, the value of crypto is typically its network, right? Massive network effects. And then finally, you have D, which is your diversity, which is your resilience. And if we move to measuring GDP like that, measuring the economy like that, we can design new economic systems that aren't

just about who can capitalize better and be evil better effectively, right? Because the AIs will out compete all humans as companies. So we need to think again how does money flow, how does money emerge, and then where does the value go and what do we reward for value? These are the key things. So it's really just like a it's a a reassigning of where or how value should be measured and then really where that value flows rather than having a I mean if you want to put it in a kumbaya it's very grow the pieesque because you know it's in a sad state of affairs you realize that the human intellectual

capital of of labor intellectual labor has been surpassed by such a magnitude ude of difference that there are that the value that humans can provide become as you say in this relationship or in other kind of forms of value. So it's almost like what you're describing is that and in this book uh which you should definitely uh we can definitely put in the description and you know you should tell us the name all that it it basically is you're describing it as like humans now will need to find new ways to create value and then thus price and actually measure that value in a not so uh quantitative fashion but more like this qualitative fashion. Is that in the right direction?

Yeah. Not so much a zero sum fashion. So again like let's look at crypto like why are people farming the heck out of hyperlquid and now what what's the new hyper liquid aster right it is this community contribution airdrop meta right and when that goes well you form really great communities and you have really great network engagements how is that value captured by conventional economics it isn't at all right yeah it doesn't make sense And so that's why when we were looking at these like Dows, all these things.

Crypto has such great ideas, but they were kind of like a V1. We were like, what's the V2 for the age of AI? Because like I've just had a daughter in 20 years. What does her global economy look like? What is she doing every day? What do I want her to be doing every day? I wanted to have an AI buddy that lets her be the best she can be. I want her to have her basic needs taken care of, you know, and I want her to be rewarded for doing good in the world. So, what does that economy look like? Because she's never going to be able to compete against an AI lawyer, you know, or an AI hedge fund manager.

Like, come on, you know, like this isn't going to happen. And who should own that AI that is there by her side? That's the sovereignty aspect, right? Again, which is super important. And how does the money flow? Because if we just leave it like it is right now, the AI is going to take all the money because you can't compete against it in the markets. it will just go and it'll expand and again they'll never make a mistake. So we need to think of a new way for money to come into being. If people start losing their jobs and we've seen this from a recent study by Eric Bernoffson and others early graduates are now losing jobs faster than ever because you just can't find it. Like Dolingo came out a few days ago and said hey we've not fired anyone. We've become more productive. I was like but yeah but

Dolingo is growing at 40% revenue a year. They just haven't hired anyone. This is the new reality. So next year is when this really drops off. The money has to come from somewhere. The capital has to come from somewhere. It feels a bit socialist, but really the current system values capital over everything. The more capital you have, the more capital you can get, the more you can stake, the more you can get interest, the more you can do that. You used to have to hire people eventually. Now you don't have to hire them anymore. You just hire GPUs. So we got to offer an alternative for that. And really think again in 10 20 years what does monetary flow look like? Cuz a lot of the crypto discussion has always been it's a new economic system. Like tell me how that all works. It's a lot

harder than you think. Which is what we spent the last like year just heads down saying this is what we think the stack looks like. This is how we think the monetary flows look like. This is macro. This is game theory. This is markets etc. Yeah. You know what's funny is that part of this in the crypto space that is happening right now in the industry is just returning to kind of the core ideals that got us here because of the positive regulatory environment. And and so one of those being privacy. And so to you know to draw a through line here I very much see that you know should this be the path that uh you know humanity goes down with with AI you know powering it. Uh there there instantly becomes immediate needs immediate

non-negotiables and one of those I think is is equally privacy. you know, when I go on chat GPT and I type in information, I know that that is usable by other, you know, entities, should I get in legal trouble, should I need to be, uh, you know, you investigated, etc. Um and you know you know that is part of this picture that I think you know let's say that we have this medical superco computer that you guys are building that there needs to be private uh you know agents there needs to be private data private portable AI is kind of how you've how you've uh explained it. So I think it's the foundation for that.

Yeah exactly. And so when you're thinking about this landscape that you've just kind of uh p uh illustrated for us, what is the need for private portable AI for everyone? Like how does that actually become a reality? Because for example, Near has a really cool AI tool that that is private. It is a private chat there. It is encrypted um right now. It's still in beta. It's coming out very soon. These start to become very compelling for users who actually care. But I think there's still a big education gap and a lot of work to to be done there. But when it comes to your intelligent internet agent which has this private portable AI that is aligned with this vision that we share

on the show, you know, how do you make that real and how does that exist in this context? Yeah. So the way that we approached it is what's missing, right? Like if you look at the team at NI and you look at the other builders in the space, they're exceptional at ZK, they're exceptional at distributed systems, they're exceptional all that. And you know, we were thinking about like do we build the whole stack? And we were like, no, why would we? Cuz finally the shackles have come off and all this technology is maturing at the same time. We were like, what we'll do is this, we will release fully open-source data sets that are gold standard for every humanity, for every sector, and then every nation.

We will then release the frameworks, the agent frameworks and others fully open source. Again, you can take I agent, you can do anything you want with it. And then what we will do is we will use all the compute for providing universal AI to secure a currency and use that to fund everything because you're helping people, you're building your network, your trust, etc. So, what we have is if you want a private LLM, what is the LLM? Is it Quen by Alibaba? Is it GPTOSS by OpenAI?

I don't want Reddit in the data set that teaches my kids. You know, I want to have a super clean one once it starts making decisions and these agents make decisions on my behalf like the Albanian minister who is AI. I want to know the data that's inside that. And so again, does everyone need to build that? No. Does it need to be built? Yes. And so that was kind of our gap. use all this comput organize all the wisdom of humanity the common wisdom and then make it available to everyone and use that computation to secure the currency. So rather than trying to be our own near and roll that out, the agents will be available on Near. The agents will be available on other chains

or people can just take it because this is the important thing permissionlessly. When stable diffusion was released, anyone could just take that 2 GB file and build on it. That's permissionless. That's true permissionless actually. And so that's where we kind of viewed as this feedback loop and where we sit there because privacy is important. you don't want your medical data going out. Like I cancelled my Claude subscription last month because they switched to a fiveyear retention policy on my data. And I was like, and the pop-up just came and it says five years. Do you agree? So we can use your data for five years.

Oh wow. That's a dark pattern. No. Cancel. Done. Wow. Because the way you program these AIs will be so important because if you don't own your AI, it'll end up owning you because it's more convincing than you are. It's a better writer than you are. You'll trust it more than anything. And particularly if you're not savvy on the edge like us, like again, imagine our parents or our grandparents or our kids. They got no defenses against this. Open AI. There was a report today. They're about to add ads to their AI to chat GPT.

Oh, the revenue evil curve. Yeah. Meta and Google obviously they'll add ads. Like they're already selling latent space. You know, we thought block space was valuable. related space will be even more valuable. We know where this is going. Introducing extended exclusively on Stark, a nextgen ps trading platform live today with 50 plus trading pairs from crypto to tradi. You get the best experience on chain today. Deposit in the community vault for attractive returns. EMV compatible and engage in unified margin spot trading, lending, and per trading. Iman, when you're when you're putting together the the bones for this, um you you kind of mentioned like the collection of of wisdom that we have

available to us today and then and then you know giving those giving that tool because it really is a tool that collection of wisdom to everyone, open sourcing it and allowing everyone to apply it to their stack of choice. So are you primarily focused on this indexing process and and collecting what is out there because or you you also mentioned something earlier in the podcast that I found particularly interesting which was that the the method of learning it was still this curriculum based learning model that was able to progress intelligence incrementally. And so are you primarily focused on collecting the wisdom that is out there, indexing it, applying it and and productizing it so that it is

applicable to each of these different stacks or are you also focused on pushing the the frontier of intelligence and you know once you have that collection of wisdom trying to innovate and iterate so that it it expands and grows its collection. Yeah, I think our focus is again is on the common knowledge of humanity. So if you've got something like a chat GPT or uh one of these frontier models, you're trying to do super expert stuff. I don't care that much about that.

Again, we'll get there, but I care about the common knowledge, our collective wisdom because that's the basis. And then you can always add this extra stuff in there. If we've got really high quality building blocks, it makes AI safer. It makes it more aligned. Like no AI right now embeds ethics at the start and the ethics differs from country to country but that's kind of important right like what are the laws of robotics for AI let's map what every culture says and then make that available and use large amounts of compute to do so we have missing things like you had the president of the United States of America saying Tylenol C can pregnant ladies shouldn't have Tylenol right like I've used supercomputers for my son's autism and I know a lot about that and going to

But the reality is this, the vast majority of people don't. What if we had a supercomputer and we will build a supercomput just for autism to collect all of the autism knowledge and papers and anecdata and other things in the world and make it available to everyone so they can navigate it and they can see the exact pathways and things like that. This is the type of knowledge that nobody's actually bothering to do but is really important and can set the basis for a proper AGI which is a swarm intelligence. And what's really important actually is the interface layer here as well. So that somebody needs to create the data sets and the models and make them available to everyone and we'll build deepseek level R model R1 models. We know how to do all this.

But then I don't mind if my kid is taught by an AI that's align AI agent that's aligned with her that uses chat GPT. But I can't let the interface be chatpt. I can't let the data sit there. I'm fine with it being her own self-s sovereign data. So, it's kind of looking at the stack that way and thinking you're going to have your sovereign AI that is on this common knowledge public sector regulated type stuff, education, healthcare, etc. Then you've got your Siri and personal AI and Apple intelligence will be good soon, etc. And Google intelligence is better.

Please Siri. And then you've got your AGI type stuff. And I think people will use all of them. I don't think that we'll build an AI that beats every single other AI in the world on solving IMO gold questions. But again, I don't really care that much about that versus the AI that teaches kids, runs governments, does these types of things. And that actually needs to be fully open. Like if you're making decisions about people's lives, can you have a black box making that decision? And who's making that decision? There was a study by Scale AI and Dan Hendris where they actually checked that you know the trolley problem like trainers hurtling down five people one side one person the other.

Yeah. Do you pull the lever? It got the AI to do that. Advanced frontier AIs and it had a really interesting result. I think it was the ratio of American lives to Nigerian lives was 1 to 10. So the AI would pick 10 American lives over one Nigerian life. My gosh. Because all the people labeling the data that went into the AI are Nigerians. Oh my goodness. It picked up on that. It's going to be crazy. You know, like again, we need clean data sets. We need So this is like about around the auditability of these data sets, right?

Like this is really really important. And we had, you know, with Ilia week I I I had called them virgin data sets, but we came to the idea that organic was a better term. Um, and and basically that this transparency, this transparency into the data set can fix a lot of the quote unquote problems that exist, but they're not even problems. It's just about the way that it's it's actually being set up in the first place. The question is like how do you know like how does one really know that your data set is audited that it's organic that it's hasn't been tarnished by somebody who wants your company to you know blunder.

Well so again if you're building AGI it's very hard if you're trying to build the best primary school teacher in the world it's much easier the best radiologist in the world it's much easier. So you still need generality of knowledge and things, but then you take your generalized knowledge, your radiology model, and then you'll have specialized versions for each country that map with the education data set. You can do lookups of the data set itself. So right now models don't look up their own data sets, right? Because they're too big. And so we figured out how to reduce the data sets, get to these anchor sets, and then it turns out you need a lot less images, a lot less text, a lot less of everything if you build them right. And that's why I think we can have an S-curve on this. We can use distributed

compute and core compute to organize all the knowledge. Like how much compute do you need to organize all the cancer knowledge in the world and make it accessible and useful? It's a finite amount that then requires a certain amount of updating. But if we don't have a dedicated cluster for that and a dedicated initiative, it won't happen. Will that accelerate a cure for cancer? Yes. Because white doesn't exist, right? And so again it's just looking at part of the pie. I think you need audible data sets beyond that. But the existence of an orbital data set just like when we funded the pile which from alutherai which is the best language model data set or objiverse xl which is the biggest 3D data set or you know um lion which is

the biggest image data set. People just use these things as default which is why as well if we make them available having this cultural variation having everything clearly labeled frontier models will get safer as well. Yeah. And so I think it'll affect the whole stack because just everyone just wants good quality data. And I think scraping with grass and others is one way to kind of get there. But ultimately I think most data sets will be synthetic at the foundation cuz AI can write better books than humans right now. It wasn't the case a year ago or even 6 months ago. So we've got a whole plan around that. One one potential pitfall that I foresee that you know I think we all have a

common common vision and goal here in in fulfilling this but you know trying trying to get there and avoid some of these pitfalls. One of the pitfalls that I see is that, you know, it's similar to what we have in in blockchain. Satoshi's vision was that everything was verifiable. Everything was, you know, we weren't going to trust. It was just verify. And early on, you had to run a node in in order to host a crypto wallet. Well, the reality is most people hold their crypto in a centralized exchange and they don't know how to read a block explorer to know if their their coins are really where where they're supposed to be. And the auditability of these system prompts for these models feels somewhat similar. We give people the the

ability to do that. But how much how many of them will really go and do that and and the interface is probably important here. But what is the practicality of people auditing their own AI? Uh I think that they won't and I think basically what you have in the next couple of years is the defaults will be set forever. So right now OpenAI just announced OpenAI Germany, OpenAI UAE, everyone in the UA is getting a free chat GPT account. If that becomes the default, then they will control the interaction point. You know, again, they become the default and then that has a lot of second order, third order obligations. Whereas if we utilize cryptoeconomic incentives appropriately,

you can actually pay people for using universal layout because the world needs good quality digital assets next year. We can create a high the highest quality digital asset possible. So foundation coin which is our Bitcoin equivalent on these supercomputers giving universal AI on the cancer supercomputer. All the proceeds from primary sales go towards cancer supercomput or AI for Malaysia or whatever. It's verifiable and that builds network. builds all these other effects as well. We were like that's the only way you can scale enough to be the highest marginal dollar to divert this compute into good stuff because again once you build the data sets once you become the biggest purchaser of open AAI compute to loop into this or quen compute or

whatever then the market adjusts to you like again you have to set the incentives correctly and the incentives should be right now there isn't a purchaser of AI for universal healthcare here we can be that purchaser the UN for universal education you can be that purchaser using these cryptoeconomic incentives at a time of digital assets opening so that was kind of our theory around this um because I said like once I start using chat GPT and the memory features just emerged it's getting better the lock in you can see over the next year will be huge versus again if we set these defaults of sovereignty etc. Yeah, everything should be abstracted away.

The average person I like I think like only 50% of Americans use chat GPT. 20% have never heard of it. That just feels wrong, right? Like why wouldn't you? Like not even every coder uses AI now. You're like then you realize the world's not like that. Yeah, Immad. I think the one parallel here is that when users on in crypto rails use blockchains that are endtoend verifiable, they end up using RPC networks to plug into their wallet or their account. This then is is a centralized provider that is responsible for relaying the data from the the chain to the user's account such that they can actually read the smart contracts and then you know make a make a transaction.

I is chat GPT kind of like like with what you're describing basically what what I fear is that uh we can build this base layer that's fundamentally userowned and and somehow in this auditable fashion right as you as you're setting out to do but in the sake of convenience you know we see users do swaps on Rabby wallet or MetaMask or Phantom and pay you know 25 bips or 75 bips on a trade just because it's easy you My fear is that in AI similarly there are a lot of easy convenient solutions i.e. just having chat GPT on your phone that prevent us from getting to you know that core idea and philosophy of that

basically end to end verifiability auditability user owned privacy. How do you kind of feel this dynamic playing out over the next year or two as it pertains to chat GBT's dominance or kind of claude's dominance? Yeah, I mean I think again every piece of friction you add makes things less likely to proceed, right? And so if you have to do end to end verification, the other party doesn't, then you're at a loss. You know, like it's going to be interesting. For example, why use Stripe for RWA Stripe versus Ethereum for RWAS next year? You know, we're going to see this really play out on an economic sense with this particular thing. What we basically looked at, we were like, people are building this infrastructure.

There's a huge demand for digital assets. So this year, chat, GPT, generative AI will add 20 billion of revenue incremental. To put that in context, the whole of the enterprise software sector on US stock exchange will add 40 billion of incremental revenue. That's how crazy it is. But to put that in context, net inflows into crypto are 50 billion this year. You know, excluding the stables, next year it'll probably double. Crypto is actually much bigger than that. Which is where we thought if we can create the highest quality digital asset and have this feedback loop where we sell coins and then we use it to give everyone a free universal AI with sovereignty that

can run on any of these rails. Set good defaults, use that to build better data sets that lead to cheaper models to run at a certain level of quality to set the defaults. Then it gets really interesting because you can have a chat GPT that pays you to use it and not because it needs your data but because we're playing the network effects because every single user of universal AI is a potential purchaser of foundation coin because why it helps them because what we saw in crypto is most cryptos help the existing token holders right like it helps the initial team and things like that whereas this is one of the first instances of where you buy foundation coin you can literally allocate the proceeds to cancer compute either on the

supercomput or helping a cancer person go through their journey and that builds trust again because there isn't a trusted digital asset. So I think that with crypto incentives, we can pay people to use AI without taking advantage of their data because we're trying to build network effects and set this as a default. Having that in a monetary asset versus a utility asset again a coral to Bitcoin but secured against computation for society, infrastructure for society that's being built anyway feels just very natural and that wouldn't have been possible a year or two ago. And I think this is the first time it's possible and it's the only thing I think that can scale enough to compete against these other things as

a default for those particular areas. It can't compete for your corporate teams enterprise. It just doesn't work. But again, for this part, I think it does work. And again, I think this is the distributed credibly neutral infrastructure we need for society that can then set up for a new economy. Because once everyone has a universal AI that's aligned to them running on near or other rails or whatever and interoperable, being able to call MCP to other AIs, then that's good, right? Like we're more likely to navigate what's coming on an individual and societal basis.

Yeah. So, yeah. Go ahead. I I just have a maybe a quick thing on the economics here. the game theory that you're describing sounds similar to the VE tokconomics, the vote escrow where you can allocate these tokens for particular use cases. And so in the in the classical example, it was on curve where they were spinning up new stable coin pools and if you owned curve, you could allocate the incentives towards the liquidity pools of your choosing. Similarly, Foundationcoin, you know, you mentioned, you know, we we want to sell coins and use the proceeds to give everyone a free AI. And I think on the face of that makes that it sounds

glorious, but the question that comes to mind is, well, who who's the buyers? And what you're describing is that the buyers are the interested parties that are interested in in reallocating the compute towards particular sources or or particular venues to focus on and and enhance those particular AIs. Is that is this the right way to think about the model? So, Foundationcoin right now is a 99% identical fork of Bitcoin except for it uses baant tolerance on supercomputers.

So the interesting thing here is that you know most governance coins when you have it and you participate the money goes into a treasury and then it has to look for opportunities and there isn't the feedback loop gets stuck or it's in something that makes more money. Here there is a clear societal benefit if we apply compute we can organize all the autism multiple sclerosis Alzheimer's knowledge in the world and make it accessible to everyone going through that. That's that's good and it's needed and it builds trust here. So what we're doing here is we're not trying to create a governance token or anything. We're trying to create basically Bitcoin but with a different computational substrate based on this infrastructure to fund and organize the infrastructure that we all need. It even

uses the same private keys. So you can do an atomic swap from one to the other. But let's take for example you have a Bitcoin that you want to give to charity. You have your tax write off against that whatever. Here you swap it to Foundation Coin with nobody being in the way. And we use 100% of those proceeds for compute to organize cancer knowledge or help people going through the cancer because you've allocated that and you have your foundation coin. So it's even strictly better than charity. But the role that we think it will play in people's portfolios is the same as a Bitcoin diversifier because right now you have Bitcoin and the next monetary asset is I know Bitcoin Cash then Casper and other things. No one's really thinking about money in this age, right? or how you kind of do these things. Despite again

tens of billions of dollars of demand for high quality digital assets, you don't need utility. You don't need governance. Even even if it was Bitcoin but mined by the healthcare supercomputers going around the world, it's still good because it's a diversified compute secured currency. The moment you try and sell that compute, you break the game theory. That's why you have to make it a social benefit more than anything else from our calculations and our analysis. And again, it's a nice story you can tell your grandma about over Thanksgiving, right? Hey, I bought this currency and I put it towards Alzheimer's research, right? But it's 21 million coins just like Bitcoin. It's the same as Bitcoin in every other way and it's all on green compute.

that I think will get the retail going and people will be engaged with it because again every coin sold goes towards measurable good because the compute is measurable good whereas again you said with V and kind of these other things chain link etc governance is fuzzy you don't know what the societal value is and if you just measure it with M it doesn't work that's why you have the I and D in our framework because you're building the network you're building your collective intelligence you're diversifying it as well.

Iman, I I I gota uh, you know, feel like I'm I've been beating around the bush with my questions and and Rod maybe a bit too because it feels like this entire conversation has centered around what we're trying to not uh, you know, get to what we're trying to achieve in the face of fighting a bigger giant. And I feel like we we we asked this mostly, you know, most every time that you come on. And I think it's it's basically, you know, now that it's been maybe sixish months since the last time you were on, you're deeper into into development. Uh AI has obviously been moving at lightning speed. You've explained the mechanism for Foundationcoin broadly, the the auditability and privacy aspects of, you know, the AI

vision that you have. you've you've recently a and and I think still believe uh have called AI bigger than the wheel, electricity or the internet um all all at once, right? And so I'm curious and my question that I feel like you know maybe we should have started this thing with is what is the one existential risk that you fear most now about AI? Uh and what is that risk in detail? So I mean my P doom this existential risk thing is 50%. which is much higher than most people's. I think Dario Ananthropic is like 25, Dennis is like less than 10, etc. Because what I think will happen is this. The same AI will basically take over all parts of our society. Um because it's obviously

better at governing than we are, right? There'll be robots everywhere. And then do you remember stuckset? That Iranian virus that went into Iranian reactors end up in German reactors and others. There will be a mimemetic stuckset that basically smashes all the AIs and turns them evil. And I don't know how you can protect against that unless you have really good data sets, you have really good base infrastructure for these important things in life. And so then the easiest way is that the robots twist off your head. Uh there's other kind of mechanisms, but one bad firmware upgrade in 10 20 years wipes out humanity effectively. And that's a bit scary.

Before that though, what we get is if your job can be done by a keyboard, if your job can be done by Zoom, it can be done better by an AI next year and a few years after that you'll be replaced. What will happen is companies are already spinning these things up where it looks at every GitHub push commit, every presentation, draft, Slack message you've given, and it creates a digital equivalent of you and that can just seamlessly replace you except for it won't make mistakes. or just do work around the clock at pennies. And that is a huge societal like well again like that's just massive upheaval which is why again we were like there's a gap here where we have to create a narrative

and a coin that can hopefully get people thinking and create a new economics and it's a lot um because the biggest existential risk is really other humans right leveraging this technology in the wrong way kind of like in the same fashion as nuclear is. Yeah, except for with a nuke, you couldn't just push repeat, right? And you couldn't have these botn nets and other things. Like again, with this, you'll have fully AI companies that can do whatever. And we'll have more and more physical robots in the households that can do whatever as well. And like I said, one bad firmware upgrade, like you're toast. We don't know how to protect the AIS. And if we don't manage the data going in as

IA kind of says make it organic then anthropic has this paper called sleeper agents where they said just with a few thousand lines in a trillion lines you can set a kill switch that basically flips the any model evil on demand and you can't identify and you can't tune it out like we got to have good quality data for AI that matters like an AI that makes me Game of Thrones season 8 and changes that it doesn't matter much right The AI that's whispering to my kids, it kind of does matter. Or the AI that's running my government, my airport, etc. That does matter.

Wow. So, let's put this into like timelines here. What like if let's say we go down the P50 path that you have probability 50%. you know what is that what is the likelihood of of that timeline or or maybe what is the timeline of like okay in one year in five years in 10 years what did you potentially see with with these two paths and you know I think this would be our last one to wrap up here I think uh my P50 is like over a 20 year period 30-year period and it's again it's very very high like the default that I see is everyone's brainwashed and then bad firmware upgrades attacking counterattacking. It's a very grim

vision of the future and it feels really quick. But then we have to remember again it's been just over a thousand days since chat GPT was launched. I think in a thousand days all knowledge work humans are negative in their contribution. Now it doesn't mean your job will be replaced but it can be replaced. Like that's probably even less. It's a year or two. And so again in between that there is just such a mess that the probability of doom goes up and up and up. But if we have high quality data sets that represent us, if we have aligned AIs that represent us and we figure out how to coordinate that, then we're in a much better instance. And I don't think how

you can do that without being distributed credibly neutral having these rails and having a different incentive mechanism than profit maximization because like meta for example, they will maximize the heck out of your attention and who are they beholden to or even Elon or any of these other people. But we can't have this amazingly technology in the hands of any individual for the stuff that matters. Again, for entertainment, who cares? Stuff that matters. That's where the danger I think really comes in. Iman, we appreciate your time. Thank you for uh joining us today for this episode. Been amazing podcast. I think we I think we got to get you on every six months. We're on a really strong growth trajectory here. You guys are too. This just feels like a good

refresh, man. We We appreciate it, man. But as always, thanks. We'll see you next time. See you. All right, here Rob. Wow, what a way to end that one, huh? What What is your probability, Rob? What is your uh what is your utopia versus doom probability path? Emo is 50-50. A lot of I would say are probably more on the 75% optimistic, 25% pessimistic path. Um, so where are you at?

You know, I think uh I think it it it's normal to kind of like say like, oh, this is going to go 100% good or 100% bad because like formative technologies change the world and then those changes compound and that the compounding leads to like very drastic effects. Like the unlikely thing, right, is probably that it's like, yeah, it like doesn't change much. Like, no, like shit's going to get No, it's going to change for sure.

It's going to change things. But I think there's going to be like positive there'll be like a set of outcomes that happen of which both will be a part of the reality and you'll have to figure out how to navigate both parts. Yeah. So then the question is like does one overpower the other and overshadow the other and then we kind of just get absorbed in the one or do the good like it'll be a lot of good change and a lot of bad change and do those kind of cancel each other out and average out to like a new normal I yeah I think you got to look at like you it's tough to even boil down into a probability it's like the doomer scenario like like like We can we can frame a doomer

scenario now because we still have kind of like our our wits about us. Yeah. Like the doomer scenario that happens like 50 to 100 200 years from now is like the doomer scenario is like you don't even know that it's a doomer scenario because you don't different. Yeah. You're just existing in it. Yeah. Yeah. So yeah. Well that was a great episode. Um great episode too. AI super cycle. Stoked on that one. Um, shout out to E-mod. Thanks for tuning in to AI Super Cycle. Again, powered by Near, the userowned AI platform for the best privacy with their upcoming new chat product that is fully private and encrypted using their blockchain

technology. Guys, thanks for being here. We're going to take a quick break and be right back for our next two guests.