AI Unchained: Building Intelligence on Open Networks - TOKEN2049 Dubai 2025

It is such a privilege to have this combination of speakers together. Some people I've considered longtime friends and colleagues. We have a a range of views on AI and also AI and crypto. Some of us having begun in the traditional AI and perhaps centralized AI space long before. But all of us very committed to decentralization in crypto.

So, I'll start by introducing myself and then I'd love for each of you to give a brief introduction. My name is Irene. I was previously head of strategy at Layer Zero, most recently one of the venture partners at Bang Capital's crypto fund and now primarily angel investing and in this capacity advising a project called Chaos Labs who's also building AI products for crypto and finance use cases. Ilia. Um, yeah, I mean my background is in AI.

I was one of the authors of the transformer uh technology that is T in GPT and then uh started near protocol which is blockchain for AI. I'm Mad. Uh I used to be a hedge fund manager then I went into AI leading United Nations AI initiative and created stability AI where we had about 300 million downloads of state-of-the-art models across image, video, protein folding and more. Uh now I run intelligent internet. We're building AI for education, government, healthcare, you know, stuff that matters.

Um hi everybody I'm Shiram uh founder at Agon layer. Uh before that I used to be a professor at the University of Washington Seattle where I ran the blockchain research lab. Before that I used to work on computational genomics and AI. Uh had the opportunity to work with like some of the best scientists including last year's Nobel laurate David Baker. Amazing.

So why don't we start by teeing this up and laying the framework for why we're even meeting right now. Over the last decade, there have been enormous advancements though much more I guess hyperbolic recently around centralized AI. Uh but over the last year, two years in particular, we've been seeing more of a shift towards decentralized open networks. Could you guys share a bit more about what exactly this is and the implications for builders and also consumers of of crypto in the audience? Yeah, I mean the high level is kind of the theis AI from my perspective is going to be the last technology wave and it effectively will be the way we're interfacing with computing with each other with services and businesses around the world and uh expecting that the single company right will control effectively how we are navigating this digital space is you know similarly how AOL was trying to control the internet back in '90s.

It's a very kind of um limiting position on one side but also very scary position if that actually happens and so uh coming from a you know open networks from decentralization from user ownership uh you know I think we all believe that it's really important that we actually make sure that these models are in like on the side of the user they are not in a single kind of uh control framework but actually are decentralized, accessible to everyone uh and uh benefiting everyone in in equal way. So that's kind of the foundation thesis and then how exactly we're doing this this is all the different approaches and and implementations but I think we all agree with that as a as a main framework. Yeah, I think that if you look at the models last year the IQ was 90 and now it's 110. So they're smarter than the average person or they're only going to get smarter. There was a Reddit study done where they set loose a whole bunch of personas and showed that they were the 98th percentile of persuasiveness as well.

We've got this like new continent AI Atlantis with like a trillion AI agents coming and a billion robots and we can't have them in the centralized control. But part of that was because you needed centralized compute to build these models. Like we just chucked the whole internet in there and like Seuss Vid pressure cooked it and out came these gigantic models. Then a few of us kind of started building open source models and it led to massive optimizations like when stable diffusion first came out it took 48 gigabytes of VRAM within a few weeks it required four and that's the power of open innovation. Now you're seeing more and more open models catch up with the closed ones and get to that level of performance.

It means you can go to what you really need to be which is AI that is personal to you and aligned with you and not a black box that determines things you know and gets a bit weird when it starts glazing you over the weekend and other stuff. And I think that now we move to smaller models that can work on the edge. You can move to distributed paradigm particularly with this test time thinking compute because most compute classically has been sequential not parallelized on these giant clusters and we're seeing that now with these distributed RL frameworks and others that we can have millions thousands millions of agents coming. your point about AI that's aligned with you, we will definitely dig into later uh when discussing perhaps why why should open source win and also why might it win which are two distinct questions. I I I do recall it was about a year ago when at least for mass usage of uh crypto AI suddenly felt very proximate and that was primarily through onchain agents and uh this exploded in the form of Twitter bots uh and perhaps very friendly accessible but maybe less like rigorous uh onchain agents but now there are teams like yours at near Ilia who are really promoting ways to build agents onchain.

Could you explain precisely what does it mean for an AI agent to function on a chain? Yeah, so the effectively what properties do do you want, right? You want a verifiability. You want to know that this code, this process was executed and this is the outcome. This is what we're getting from the blockchain.

This is, you know, what we kind of really enjoy with smart contracts because now you can trust with your assets, with, you know, coordination, with identity, with all those things. There's also another aspect which we haven't had as much on the blockchain historically which is confidentiality right especially as you're starting to use this agents for personal use personal information organizational information you want to make sure that uh you actually you know don't leak and tell all of this data to everyone uh if you're doing trading you don't want to leak your trading algorithm to everyone so uh for for us effectively the onchain agents are a verifiable confidential compute that's paired with ability to execute actions across financial infrastructure like blockchain as well as you know traditional web two infrastructure uh you know APIs Twitter etc and so there's a framework for that there's infrastructure for that and this enables you know everything from kind of a very intelligent for example index funds that are you know reading Twitter analyzing sentiment and rebalancing the portfolio to you know obviously like autonomous trading to something that can get you, you know, a personalized news summary in your uh, you know, uh, inbox every morning or it can go and, you know, do your grocery shopping uh, kind of knowing all your preferences and feedback and maybe the, you know, you took photos of food that you've liked, right? So all of that effectively becomes a framework for this verifiable confidential agents. The euphoric promise we saw about a year ago that went viral on Twitter was around autonomous trading agents. Shiram, you've written about this framework of three different types of agents.

Could you walk us through exactly what those three are and why you even articulated this in the first place? Yeah. Um, when we think about the um when we think about what what onchain agents mean the first the laid out some of the properties and I think right now you you know many of the agents that exist uh onchain don't satisfy these properties, right? So they're not transparent, they're not verifiable, they're not unstoppable. And at that point, it's not clear what blockchains bring to AI.

So our thesis is like this these two properties, the fact that it's unstoppable. You know, a system continues to run, you know, without anybody needing to keep it up. And then uh verifiability, these two properties are the kind of like core falcrum on top of which you can build very powerful agents. And you know speaking to one of the earlier points about uh you know should crypto be open or not one or should AI be open source or not or should they be built on open networks one one way to think about this is you know when you look at the existing models and the competition you see mostly it's like this you know huge models competing with each other and you know as uh Emod pointed out we see an emergence of smaller models which can be competitive on certain dimensions. and for you know customized and for certain use cases.

So instead of thinking of this King Kong versus Godzilla, you can actually have a swarm of beast like many small agents, many small intelligences which can coordinate with each other once they're verifiable on chain. So that's the kind of like outlay of like what what our thesis here is in terms of what are the different kinds of agents and what levels of verifiability that that you want. One way to categorize them is by uh the utility. So agents could be personal, right? So you think of your own agent like you know I I want an agent that summarizes news for me or does some action on behalf of me.

You know it could be for for an individual or for a company but those those all we call personal agents and these personal agents not clear that they need to be onchain and verifiable. They just need to be o open source and owned by the user. And then you can think of the next category of agents which is agents of the commons right or collective. So you know a protocol like IGEN layer is an example right like it's it's a protocol which is onchain. It belongs to everybody.

It behaves in a certain predictable way because it it is onchain code. You can think of agents the collective agents like that. They're basically enhancing what smart contracts can do. If you think of what a smart contract is, it's right now not really smart. is just robust and rigid, right?

So, it's a contract that actually does what it says it does. But there are many many reasons to expand the scope of smart contracts to include onchain intelligence. You know, whether it's simple things like trading to more complex things like adjudication, right? So, when you think of a smart contract, why is a smart contract written in an obstruuse solidity language or any other language? Why don't we write contracts in English that can be adjudicated by an onchain AI?

So as you think about the spectrum u you know I covered personal agents agents of the collective and the third category of agents which I'm very excited about is totally sovereign agents right these are not agents that either work for any of our any of us individuals or even for like a collective they operate on their own interest and so I think this is something uniquely enabled by crypto it's not possible to do this anywhere else think of a kind of like an agent that just lives on chain what does it mean by live onchain like a a Smart contract is basically a piece of code that can own a piece of property, right? That's what a smart contract is. If you can take an agent and just put it on chain and that onchain agent can own property and do its own, you know, have its own objectives and intentions, you know, it could be really interesting because instead of having one megalith, you have lots of these small, you know, AI almost living things that that have their own intentions and objectives. So these are the different kinds of agents that we think about and you know the last two categories you know agents of the collective and sovereign agents clearly need blockchains to actually uh you know thrive. What I appreciate about this framework is you clearly articulate where based off of the utility or aspired utility of an agent crypto or crypto rails actually has meaningful purpose rather than needlessly injecting perhaps crypto or decentralization.

I'm curious from the point of view as as entrepreneurs and building different elements of the infrastructure stack. Do you feel that all elements of infrastructure has been built or is already live right now for builders to realize that third category of agent? And as you answer this question, help position say like near uh intelligence systems agen layer into the exact infrastructure stack for the audience to understand. Um so I think you know uh we just had that Ethereum panel and there's this concept of credible neutrality. AI agents will be smarter than humans and they will make less mistakes and the AI economy will be better than the human economy.

I think that's reasonable and the credible neutral layer that's verifiable is what they're going to build on because agents will come from all different social sizes. I think all the elements of that are nearly here. like we're seeing breakthroughs in ZK, we're seeing breakthroughs in speed, we're seeing breakthroughs in identity. And so, you know, humans build on top of the DeFi stack. There's going to be another layer of that whereby the agents will build themselves.

And I think this is the move from the synchronous agent, you know, the chat, GPT, goldfish, memory, stochastic parrot, stable diffusion, image comes in, other text goes out to just go and do your job and then it will just go and do it asynchronously. I think all the elements are nearly there for that. And the example I give of that is actually vibe coding. You know, like my nephews and nieces now make computer games and they're just like five, six, seven years old. I think every school should have a 1-hour vibe coding session, but the AIS will be able to vibe code themselves as well.

And you look at the stack and the different layers. All the pieces are there. And I think in the next year or so is when we see the explosion of asynchronous agents of these various types occur because they're good enough, fast enough, and cheap enough to run. and they'll start accumulating assets because they're persuasive and more and I think they'll fill in any of the gaps that they don't need to because classically you needed like APIs you needed very like writing solidity is hard you know and AI is very very good at it you don't need to audit the contracts in a couple of years for that so I think they will just build the parts of the stack that they don't need and pull from the best of the stacks for us at intelligent internet what we're doing is we're building the base brains and models so we're going Today we released one of the best medical models that's a bit later at my talk. Um, healthcare, education, government.

And so we're just focused on that part. But when we look at the verifiability, we look at the traceability, the ability of the agents to interact. We just released one of the best research agents. We're about to release a manis type agent as well. Again, we're very infused by that and that's also why we didn't want to build it ourselves because people like Ilia Shriram and others have been building this for years and I think it's finally coming to fruition.

I think I agree with you on the statement that all the elements are in place for an explosion of of creation using different AI tools like uh lovable bolt vibe coding tools cursor as well. However, I'm more skeptical. We spoke about this before the panel around all the elements being in place for decentralized and open network building. Could you elaborate more on why you think we are in that place now? I think the ingredients are there.

It needs to be all pulled together. And you know this is again the promise of blockchain interoperability. It's a promise of bridging. And I think this is where the credible neutrality and the ability of AI to be the bridge comes in. Like when you look at most of the hacks and most of the rugs in AI it's like in crypto it would be great if you had a credible bridging AI.

You know it'd be great if you had AI adjudicators. Like Dows don't have much intelligence. What happens when we have that added to a DAO? So you have humans and AIs there. But again, I think that the AI assistance over the next few years will help pull these things together at the same time that these technologies we've talked about for a decade now are maturing.

And that's why I'm enthused. So I think ingredients versus like, you know, it's not ready yet. We can't just like freeze it, you know, code freeze, let's go home. Perhaps this is the right segue into the meat of the discussion, which is why should open source win? And I think this is something that in in spirit we we all agree with but perhaps there's a little bit of disagreement around the second question which is why might open source actually win which is a more practical one.

Let's start with the former question. Why should open source win? I know each of you had made very bold statements throughout your careers around uh the urgency perhaps around open source and development right now and also take near for example the ethical considerations around privacy. Could each of you share a bit about why you think it should win? Well, my position open source always wins.

The question is the time frame. uh and I think I mean we we see this at at for any single thing right in especially in software but even in hardware uh effectively information wants to be free there's no way to like capture information and protect it so hard that it never comes out uh at least when we're talking about something like software where it's used by many people and so uh I think there's like a fundamental framework right and and we see this even with ideas in open in AI right even with closed source companies the ideas even when they don't publish the ideas are still get invented because people see what works you know and like okay we like we can reverse engineer from what worked and figure out how it how it worked or people change jobs right or whatever like the information just keeps propagating so that's kind of one side the other side is obviously we're getting better and better at training this models right like if you think you know uh 5 years ago 7 years ago right when we're training them versus 2 years ago versus now we're just getting way better right the models of you know smaller and smaller size becoming smarter and smarter um and so fundamentally the the some of the restrictions and requirements you know where you needed a you know massive pressure cooker right you can soon be able to do it on a skillet in your kitchen right so uh so I think that uh but from a kind of higher level perspective uh I kind of I frame this says right now kind of we built on a I call it corporate owned AI not actually even open source or closed source that's actually a like to me that's a not as interesting of a dichotomy it's who which meta game is effectively governing governing this model right so you know when I work at Google my goal is to make more money for Google and so the the decisions I make the data I put in in the model the way we AB test the launch of the model all All of this is based on how much revenue is this going to generate. And whenever you maximize your target audience, right? There's literally no more people with devices who can access this. You're going to start trying to extract more value from this people who are already users, right?

And we know this is happening with Google. There's been uh publications about it, right? Where you know effectively like how do we make people to search more? Uh like the ads team is effectively asking, right? So, so this is like a natural state of capitalism and and kind of all these companies and given this technology is fundamental to how we're going to interact with the world going forward, right?

Again, like the sciences are now developed by AI, right? The, you know, the way we do marketing is done by AI now, right? All of this is, you know, how we're going to influence people, how we're going to talk to people is going to be done through AI. it's fundamental that it's the meta game is not to make money for some company but really to make more individuals successful and so that's kind of where user only uh position is coming from yeah I think that the world runs on Linux servers not Windows servers and AI is infrastructure of the future but then the nature of these models you know if you actually look at the models it's like an MP4 it's not like a logic tree it's like a filter where information goes in information comes out and Google is already selling ad space in Gemma they're openour ource model meta is as well. So you say a beer, it says a Bud Light.

As these models make more and more decisions for us and we outsource more and more of our minds, you know, like not your keys, not your crypto, not your models, not your mind is what I say. This becomes important because the inherent biases of these models, if we don't have verifiability and traceability for the stuff that matters, is very important. Dan Hris, who's one of the advisers at Scale AI and XAI, did a study where he did the trolley problem for top models. You know, you pull the thing and does it kill five people or one person? And it turned out because he did it by country that one Nigerian life is equal to 10 American lives because all the labelers were Nigerian and he kind of did this and that shows you the inherent biases within the models.

You look at the open AI glazing thing that happened this weekend when it suddenly became so sick and a bit weird. Again, our children are going to be using these models. You know, our grandmas are going to be using these models and the way it affects our brains is going to be so much that I think we need open models for things that matter. So that's what kind we're focused on, education, healthcare, cancer, all that. But then also decision-m models.

I don't think you can have any decision-m model that impacts human lives without open verifiable data. You can't have black boxes deciding these things. And so I think that's one of the reasons it will win. The other reason is open source infrastructure wins. And the final reason is just innovation.

Again, we had 300 million downloads and there was a whole innovation ecosystem that built around this because it was permissionless innovation. This was the original beauty of Bitcoin, you know, that kicked off whole crypto movement. You can download the nodes and you can run it yourself. So, we had people coming from all over. It's so much easier and less mentally damaging to build an AI than crypto, you know, because you can just get to it.

I think a third of our coders only had one year experience in AI. And they just downloaded the model, they tried it, and they were like, I love this. And they went in and all of a sudden they were building state-of-the-art models. And I think we'll see that continue. So the infrastructure will be verifiable, coordinated and robust building on top of this stack that as I said is coming together now and I think that will definitely out compete the closed source ones for the things that matter making our lives better increasing our agency.

AGI is a different question ASI you know we don't have time to talk about that today. Sham do you have something to add? Yeah. Uh the the main thing that you want when an intelligence is kind of making decisions for you and is in your life in various ways. For example, if you have an Android or a robot at home and it's completely closed source, who knows what it can do, right?

So I think we will see a whole wave of you know not just out of the free market and the open innovation and the exponential cascading of permissionless innovation that open source you know wins we will also see regulatory requirements that some of these things actually be open source in order to be deployed right imagine I import like you know uh in in the US or in India you know import 100 million robots from some other country and god knows how how it's running. You know, it's a sovereignity risk for a country to actually have robots running around which are not verifiable, not trustworthy. So, there is going to be all these different dimensions why you need these models to be trusted. The more central position that you put these models at, the more trustworthy that they need to be and that's going to make sure that they're open source. Not just open source I think many of these models may even need to be open state like you know adjudication models these kinds of things open state means not just that the source code is open they're running on data that is replicable and verifiable and I think that is going to be really important and I think this is you know people think about this you know even myself like a few years back when I was thinking about this AI overlord thesis like oh you know there's going to be this AI overlord and I think there is a very interesting thing that's happening with intelligence that is replicable.

This is the first time in the history of like you know planet earth our evolution that you have something that is both intelligent and replicable right so when you when you look at what a leader is like you see oh you know this guy is trustworthy or this lady is trustworthy and you know I want to follow them because they're intelligent and when you when you have this for the first time models that are intelligent and replicable which means you verifiable and trustworthy way more than any human system has till now because you can just re-execute this model and get exactly the same result. It is going to create a shelling point for our justice system moves onchain. leaders emerge who you can trust because you know how they're going to behave and instead of like this you know somebody enforcing that will on people it'll be people voluntarily follow these AI systems because they are both more intelligent and more trustworthy only if they're open source and open state so that's why one of our one of my thesis is that you know maybe slightly counterposition to the panel members here is that fully sovereign agents which actually have these properties they're both autonomous and verifiable will command a large portion of the GDP in like uh two decades from now and so these are fully sovereign they just are acting on you know completely sovereign nature but they're verifiable so that induces trust in them for people to follow them along so if you predict that the majority of the world's GDP in the future will be run and commanded by sovereign agents let's talk a bit more about the why might open source actually win from a very practical business outcome point of view. For me, as an investor in both crypto and in traditional AI, I've long been skeptical of the superior business outcomes of a decentralized, yes, you know, supportive of open-source approaches, but truly decentralized open AI network. Uh, Ilia, you've said that you think that there is a possibility for superior business outcomes.

And then, Immad, you've also said before you think there is a trillion dollar opportunity opportunity here, but it's also the largest bubble ever. a direct quote. And so I'm curious to hear more. What is the case for a superior business outcome against hundred million dollar billion dollar uh funded closed source AI companies? Yeah.

I mean effectively there's few few sides of this, right? One is just a pure monetization. So let's say I'm I'm an AI company and I'm building a a product. uh I have effectively two two routes. Either I open source it and uh like I make very little money because people just take it and use it, right?

And uh uh and usually people do this effectively to get you know reputation and then actually keep things closed source, right? one way or another or I keep it closed source and now I need to actually build my own serving infrastructure HIPPA compliance GDPR like I need to build a whole ecosystem of services infrastructure processes compliance teams etc to a protect the model right models have been leaked and people have actually lost their economic engine uh you need to protect user data right user data have been leaked or been reused in in you know uh in other places uh or shows up in you know somebody else's data shows up in your chat. uh um so you have all these problems you need to deal with right so you need you know a really large team or you can use a this kind of infrastructure that gives you verifiability confidentiality that allows you as a developer to upload your model in such a way that the users can use it but you don't see their data right so you can now monetize without really needing to um to take on all that risk you can also give this now model to other people with their own hardware because Right now you actually need to trust the hardware provider that they not going to leak your model again which has happened. uh you need to trust the hardware provider with the user data right for example European requirements is that data doesn't travel outside of Europe but all the hardware is here in UAE and in Saudi right and so now there needs to be data packs if you want to move the data European data to Saudi to run inference and move it back versus you have verifiability and confidentiality that indeed only this model on this data was run and nobody have seen it right so it's really like an infrastructure that enables a a lot of like business properties allows you to have way smaller team to deliver the product to the users and users getting better properties because they also you know if I'm a company like I'm a Swiss you know foundation I have problems because I need to protect my kind of private data because it was given to me by some other third party right so I cannot feed it into open AAI right now and so I need to use something like this as well so you have effectively now B2B path a lot clearer so that's kind of I would say like from a business perspective, from a development perspective, again, open source just allows to like move faster, right? Uh you can do so much more things, right?

You know, uh with this models, with agents on top of it that uh you know, OpenAI is offering you like some of the like okay you can fine-tune with this API and you can maybe distill in this way, but it obviously locked in and you cannot well I actually want Laura on the image thing that is done you know I want like five stacked Laura. it's like okay now you cannot do that versus open source you can and you can like really quickly launch a product that you know customizes images for whatever generation. So those are really like kind of the you know immediate things that are allows developers to use this and then things that are coming down the line are inter agent payments adjudication like all of this other services that right now people not even thinking because I would say I call it like current AI like centralized AI is a single player mode right everybody kind of thinks from like perspective like I have AI and it does stuff for me they don't think about like well the other party also has AI and now this AI need to figure out how to do business how to you know decide when something goes wrong and so as we building out more of the systems and like it's all now AI right like you the call in restaurant the restaurant AI picks up right like how are they negotiating is becoming question like all of this is where again the tech we've been building in blockchain is really applicable and really solves this problems and we have the mindset of you know also just like different incentives and attacks etc. that again AI space right now is really lagging behind. Yeah, I I think um you know thinking about what Shri just said, Bitcoin was almost the original AI sovereign.

It provisioned lots of humans and captured 1.5 trillion of value by creating this credibly neutral thing. It has many of the same characteristics. Maybe as CZ said, you know, Satoshi is an AI from the future come back. If you look at this AI economy that's coming now, I remember I gave a grant to Midjourney to get it going and it got to 200 million in revenue thanks to the hard work of David and his team and they built on our open source models.

You see things like curses now hit 300 million in revenue. You're seeing record revenue numbers building on top of these models. The reality is nobody should be building models. Building models is a bad idea. Like once we build the world's best doctor model, everyone can just use it and then you don't need more because you have a satisficing level.

You have a VHS versus Betamax moment. Doesn't matter that another model is 5% better if it does the job. But looking at the other layers of the stack, that's where the value really comes in. For our own thing, we've got a white paper coming out in a few weeks, II. Where we talk about how we're building models, but everyone else should be building on top of models.

And that massively reduces the cost of scaling your team because what it's ultimately about is are you providing value to the end user and now you don't need to worry about hiring people which is the most difficult thing of all. You hire AI agents instead. So once you know how the models work, you can interact with them and people wrap them for these services. You provide an enduser service that's valuable and that's what people pay for in any kind of way. And so I think that base layer though is that infrastructure foundational layer and it's important to have open models but then once we have a series of open models that are doctors, lawyers, accountants, customer service etc.

just use those services as piped infrastructure like electricity and then scale above that stack. I think we're certainly aligned in building on on top of models and there being more perhaps value cruel there. You've mentioned that your team has been building a best-in-class medical AI model. I think the fine-tuning and applied rag that you can do is now where the extra magic is made. Like how can you make something domain specific?

Uh so that's as for me as an investor something I find particularly exciting. I as we get closer towards the end here I I want to round out by asking two different things. One is around the urgency the why now. I I think we would be foolish if we were to assume we have 10 years to make advancements in open source AI. I there's a very small window of time for open source to perhaps get a lead here which also means that the execution and talent has to effectively congregate around these missions in perhaps the next year.

So Shiron this is something I know you have spoken about. It's the time horizon. Could you speak a bit more to this and as a call to action perhaps for the group here? What does that look like? like how do we tactically actually make advancement there?

Yeah. Uh I think like one of the biggest things that crypto is great for is incentivization, right? Imman is talking about Bitcoin as this autonomous system that emerges because of what? Because of incentives that are built into the system. Can we how can we build the right incentive structure for anybody to come in and contribute to improve improve these models whether it is by acquiring data right for example if you had a system which is attributable data.

So anybody can contribute data to improve the model and as the model gets used if you know the people who contributed the data get paid. That's an example of what crypto can do. Um so in terms of incentivization to actually create these open source models I think we have a very short window of opportunity for crypto to play any role in this uh in this space. But there's also one thing I want to add here which is the ex the powerful AI models when they are made verifiable and brought on chain it fundamentally changes the substrate of what crypto itself is. You know when you think of like you know dealing with blockchains today you're signing off hashes of like machine code.

When you're writing a smart contract you're writing code in some bite code in solidity. when you're uh when you think of something like decentralized science, we are thinking of like you know onchain uh coalition of capital which is actually going and funding things in the real world. When you bring AI into the equation here as a call to action for what we can all go out and do you actually see that all of these layers get fundamentally transformed you know intents automatically taken in by the blockchain and it it says like if this is your intent I'm going to figure out and do it by myself. uh when you think about contracts that are automatically adjudicated because you can state them in English, you can think of scientists which are AI agents which live on chain and can collect money from people verifiably being spent on actually advancing knowledge as opposed to you don't know where where it went non-transparently. So all of these things you know just inside the crypto space can be augmented massively using AI.

So that that gives us like an in into the space. Yeah, I think you can scale your dev teams without them being verifiably North Korean as well. Also be positive. Um I think basically we're at the edge of a complete reordering of the economic system. The are smarter than us.

They don't make mistakes and you'll be able to zoom call or chat with them by next year. And so I think what people should be building for is ensuring that they are verifiable, trustworthy, helping people, increasing their agency rather than trying to replace them. And again, just thinking about what the economy of the future is and what the things you could build that you could never build before were because you didn't have access to these resources. Now you only need a team of five, six, seven, eight people to build stuff that can change the world. And how awesome is that?

Create the right incentives and they can proliferate globally. How great is that? And you have to do that now because we have to build the positive because like I said, I don't know what my kids will do for jobs, you know, because how can they compete with an AI? Maybe they'll be farmers. I don't know good for them.

The robots will be farmers. Yeah. Yeah. Yeah, I mean I think like as you said we are at the end of kind of labor capitalism and we're moving into post labor capitalism and it really depends on like effectively the structure of control of the system. Is this going to be like a single or a few few places of control or actually a decentralized network of effectively every that where everyone is benefiting and everyone is kind of balancing the system right I mean one way to think about it is you know one person with really smart AI can do a lot of damage and uh you know like one of the things you know everybody's like hey you know AI dumerism etc and my point always is like well you actually need to be afraid of people with AI not like AI doing bad things itself right and so the question is like the at least from my perspective just similar how we have in blockchain right we need you know 51% of people following the protocol we have we need to have more people with AI being on the good side trying to uh you know advance humanity and and maintain kind of wellness for everyone so that even if people who who do have access to AI I in a malicious way, everybody else's AI kind of captures and and you know uh limits that.

So you kind of need to build this network in a in a decentralized way otherwise again somebody hacking into one of the centralized companies stealing their weights can then go and leverage it and you know they're going to be effectively on the same parody uh with another company. So yeah, I think like it's it's a pretty short window of time for us to really build this up like probably 5 years and then we are kind of in a liftoff. I think perhaps less. Uh what I what I appreciate about each of your statements is the emphasis on human agency. Human agency is something I think the four of us deeply care about.

And uh for everyone who's here right now and and clearly cares about AI and uh AI systems that will will benefit humanity, now is the time to get involved. We need the best possible talent contributing to these systems. For me as an investor, that's something I I certainly want to see even more of. Uh and to start using all of these tools, whether it's chaos AAI or or building with with near intelligent systems and and iigen. I thank you all so much for the contribution of your time and and talent to this space.

Thank you.