Crypto's Answer to Centralized AI Monopolies | Anand & Niraj

let's say that you wanted to take out 3x levered exposure to Apple with DAI as collateral, and then at some price point you should stop out and close out the position to, you know, have some safety or something around it. Normally to do that in DeFi, it would take like 10 different smart contracts and clicking a bunch of interfaces. But I said that to you in one sentence, and I think the power is being able to take that one sentence and convert it into some sort of form of intent. And there's a bunch of companies building on the intent side and then being able to now interact with blockchains in a way that's actually usable by regular people and not just people that are very in it.

Scary stat for you. The odds of falling victim to online crimes are 1 in 4. Aura provides digital security protection to keep your personal information, passwords, online activity, and technology safe. Safe from online threats. For Empire listeners, Aura is offering 14 days free plus up to 55% off an Aura subscription when you visit Aura, Aura.com/blockworks. Terms and conditions apply, see site for details. You'll hear more about Aura later in the show. This episode is brought to you by Kinto, the safety-first layer 2 that features user-owned KYC and native account abstraction to solve the two biggest roadblocks to mainstream adoption: one, security, and two, user experience. Their founder Ramon sent me the onboarding flow. It's engen.kinto.xyz. I would highly recommend becoming a founding member of their launch program. You'll hear more about Kinto later in the show. Are you tired of high gas fees? I'm excited to let you know about SKALE, a zero gas fee modular blockchain that's become a perfect fit for gaming and AI apps because of their instant finality and lack of MEV. Explore the SKALE ecosystem today at scale.space/ecosystem and stay up to date with the gasless blockchain on X @skale_network. That's @skale_network. Big thanks to SKALE for sponsoring Empire. All right, everyone, welcome back to Empire. We've got Anand, managing partner at Canonical, venture partner Lightspeed Ventures. We also have Niraj, who spent 5 or 6 years as a at Polychain investing alongside Olaf and crew, and then is now co-founder of Ritual. So Niraj, Anand, welcome to the show, guys.

Thanks for having us.

Thank you.

Yeah. So I think the episode, the episode that I want to have here is something along the lines of there's all this AI and crypto stuff going on. We had Casey Crusoe on the podcast back in December. It's 5 months later. I'm seeing Martin Shkreli post about IONET and I'm seeing Akash and Render ripping and I'm seeing like I'm just seeing a lot on the timeline and I haven't really fully caught up and I want you guys to catch me up. So maybe the best place to start is, Niraj, I'll pick on you for this question of just like, why now? Why is all of this AI and crypto? Obviously there was the ChatGPT moment that kind of unlocked the like mainstream AI moment. But like, why is this all coming together right now?

The really the big reason for AI and crypto coming together at this stage is, is a couple of big reasons. And many of these reasons are what informed me to get into the space more fully as a builder. So if we separate out the two things, we have AI and we have crypto. If we zoom out and think about why AI, AI, like very similarly to crypto, has gone through these cyclical up and down moments where there's hyperinvestment and then it seems like it's dead for a few years and then constantly cycling through that, albeit in AI a bit more stretched out of a, of a time period. AI has been around for, you know, 40 or 50 years, whereas crypto is more like 10 or 15. Um, and so we've had these various innovations, you know, starting from the, the age of compute. Um, you know, even people like Nash were thinking about how we can have, you know, what, you know, artificial intelligence inside of computers, how can we replicate, you know, human intelligence and, and move past it. I'd say the most recent wave before the current one was in, you know, the 2010 to 2015 era, where we started seeing AI being used in things like self-driving cars. This is when a lot of the kind of bigger companies started coming out. Waymo and Uber had their ETG program. And then we started seeing AI in the SaaS part of the world. Ways to improve the lookup of data and entry and things like that. But I think the big fundamental challenge was that there was no interesting consumer use case for AI. Everything for AI was in the background. You refresh your Instagram feed and a recommendation system is getting you the best results, but you're not viscerally feeling artificial intelligence. And so fast forward to 2017, 2018, we start to get newer techniques in AI. Some I mentioned you've covered on the show before, but if not, we can go into it a bit later. Deep learning, one of the more recent fundamental improvements in the ML world, came about. We saw the transformer architecture, which is kind of the fundamental building block of the large language model. And so those things really brought us to a place where models are now really, really good at doing a lot of things. And that combined with the release of ChatGPT probably 2 or 3 years ago, took this highly capable model that is, you know, I'd argue is better than humans at most things, the vast majority of humans at most things. It, it beats, you know, most humans on SATs and math competitions and things like that. They're highly capable. And so OpenAI bringing ChatGPT out into the world showed that, hey, this is a way that you can use artificial intelligence and really feel the power of what it means to be able to get an answer to a unique or interesting question very, very fast, faster than you could with Google or asking someone. So with that backdrop in mind, something like ChatGPT has 200 to 300 million daily active users, which is a really incredible number in the span of 2 years. ChatGPT was the fastest growing product of all time. So that means that there's something very valuable and unique there. And I think AI has really cross the chasm in terms of people being able to use it and understand it. So if we go over to the crypto side, we think about AI is great, but what are some of the challenges that, you know, as crypto people, we foresee coming out? It's things around privacy. You know, most of these companies still take advantage of your data, being able to actually track how data, you know, the data that goes into the models and how they're being used. Being able to get confidence that a model that you want to use is actually being computed correctly and a whole bunch of other features. Basically, there's this increased centralization that if we don't solve with some other technology, like I think crypto, crypto being a tool for self-sovereignty, own your identity, own your data, own your money, we'll have these great AI systems that are the governance of them are an ownership of them are controlled by a few companies and governments, which is, which is a bad sign. We've seen, you know, money being taken advantage of when controlled by centralized entities. We've seen it in Weimar Germany and Argentina and so many other places. We've seen it in tech platforms. Things like Facebook and Twitter in the 2010s when they ripped away API access from all the devs. And so we had Bitcoin as a replacement for money and we had Ethereum as a replacement for technology platforms. And I really think we need a decentralized alternative to AI. And so that's really a couple of the underpinnings that we've thought about as to why now AI is ready. And we have the cryptography and cryptocurrencies that are technologically developed enough to be able to merge these two things together. And we can go into that, that more, I'm sure, later on.

Yeah, maybe we can dig. So there are features that we can talk about. There's agents and self-owned models and data and verifiability and crypto-incentivized AI behaviors. Local and peer-to-peer inference, there's decentralized GPUs. I think of these all as things that could be unlocked. But then there's this broader, like, the problem statement that I think we're trying to solve. And I think, Niraj, you hit the nail on the head. And it makes me think of this, the Peter Thiel quote, right, from maybe, I think he said it 5 or 6 years ago, where he said crypto is decentralizing, AI is centralizing. Or if you want to frame it more ideologically, right, crypto is libertarian, AI is communist. Anand, how do you think about this, this, like, like this overwhelming problem statement that we are trying to— we're trying to solve, where AI— what AI will ultimately do is make the Googles and Amazons and Facebooks of the world a lot more powerful than they are today.

Yeah, no, you're right. I think Niraj hit the nail on the head, right? Like, this is, um, when it comes to, um, well, I'll say two things, right? Uh, one of which is, you know, today with the state of LLMs, we're looking at what 200,000 years worth of data that's been used to train these LLMs. And to piggyback off of Niraj's point, the why now is also because we have all this data, right? Like I remember taking a programming class in 1999, which was technically about AI, but it was purely about just learning how to code in Lisp. And that was the— we were like, yeah, AI's here. This is 2000. This is when AI is going to happen. But it took us all this data and the ability for us to access this data But yes, now the second thing being that, to your point, a lot of our human behavior is being monetized, whether it's intents, it's searches, it's our personal behaviors. It's being monetized in ways in which maybe we didn't foresee, right? And these have become large monopolies. So, the real question in front of us is like, do we, when it comes to something like AI or it comes from future generations, and this is, you know, maybe to some extent, and there's some chatter about this today, that the current state of AI has sort of asymptoted based on human intelligence. But that's maybe the promise looking forward is that it will leapfrog human intelligence, which is sort of like, you know, the maybe part of the excitement and the fear at the same time. But if and when that, rather, when that leapfrogging happens, do we want to yield control back to centralized entities, especially if these are tools that we're going to continue to live and see into perpetuity? Do we want the governance to be controlled by large entities or do we want to have a sense of ownership around it? So agreed 100%. And I feel like when I first met Niraj, this was like the why now for him to move from what he's doing to what he's doing today was if it's not me, if it's not us, then who else is it going to be? And crypto with sort of the backdrop, AI is sort of the big problem statement and the repercussions of this is just too large for us to yield control back to the big power. So yeah, in agreement there.

What is the— so I was trying to explain this actually to someone yesterday. They're maybe 65 or 70 years old, and I was explaining how AI is the centralizing factor. And they said, great. Yeah, Amazon. No, they said, great. Amazon has done a pretty good job with my groceries so far. You know, Facebook's done a good job as a social marketplace. Like, why should I care about that? Why does it— why is it a bad thing that it's a centralizing force here? So I don't know who to ask that question to, but Would just be curious how you guys think about that.

I'm happy to start a quick stab and then Anand, I'm sure you have a bunch as well. I think when we talk about crypto AI, you know, Anand and I have had a bunch of discussions about this question. Basically, like, does crypto AI matter outside of crypto AI? Does it matter to AI people? And historically, the answer has been, you know, no, for the most part. I would venture out and say the average AI researcher or engineer, what they care about when they're going to join an AI company is they just want to create the best model possible. They are really fascinated and interested in how do we improve the artificial intelligence quotient. But as we've seen with a couple of big events, you know, it takes the breaking of a centralized system to, you know, pill people on why they are bad or why there are issues with them. And so two big events really happened that I think catalyzed, even within the AI realm, a category of people interested in decentralized AI. I mean, at least two that I can point to, but there's certainly more. One big one was the OpenAI fiasco in November, which everyone remembers. Where, you know, separately elected board tried to kick out, you know, kick out Sam for, you know, some reason that we haven't really found out yet. And, you know, for one or two weeks, every single AI friend of mine was not sure if they could even build on OpenAI anymore. And to have that, the, you know, the best model in the world, not, you know, you're not sure about the, the sanctity of the platform underneath is, is pretty bad. So that's one, is now being able to rip away access to models and make sure that through crypto you can more easily access distribution. The second one was earlier this year when we saw Bard, Google's Bard assistant, making highly sort of politicized responses to questions that people would ask the chatbot. And I think what that highlighted, more so than that the answers were, you know, quote unquote wrong, was that the answers were tilted to how the developers of the model wanted the answers to be. And I think what I care about in crypto is that we're pushing for a user-owned operated internet. And so it's ideal if users are able to instead pick how they want the political leanings of the model to be or how they want it to act and interact. And so seeing those events, I think, caused a lot of people in the AI community to start thinking about, hey, we should start doing open source models, we should start making it easier to do local development of models with GPUs in a box at your house. And it also catalyzed a bunch of people like Anand and myself and many others to go and pursue getting more involved in the space. And so those are a couple of reasons that come to mind for me as to that question.

Yeah, I mean, those are the big cataclysmic events over the last 6 months, and I have a feeling we'll see more of these as we look forward. But generally speaking, at steady state, I don't think the consumers should care about this. That's fair, right? Fully admitting that in the world of crypto, we have crypto speak and it feels like a very isolated community today. But yes, I don't think the average user should care about this. But if you look at, I guess, just two more points on this, one of which is when you start looking at specific nuanced fields, legal, finance, medical, et cetera, where your data is being used as basically AI is going to help you in some very codifiable, interesting way, it is going to start to matter. You're going to want to care about why this model is behaving the way it is. Why is it hallucinating? What are the factors that went in and whether the output is biased or not? And then you look at sort of the primitives that we've grown up with in the world of crypto, transparency, auditability, verifiability, all these things that we throw around, these feel like absolute necessary primitives if you want to put AI to work in very enterprise, in an enterprise-grade scale. So agree with you on the state of affairs today. Hopefully the end user doesn't care, but there's certain systems where this will start to matter and it's creeping up very quickly.

Niraj, I'm going to ask you a really smooth brain question here, which will show my lack of understanding of all this stuff, but that's all right. How is the, so I listened to the Dwarkesh podcast with with, uh, Zuckerberg and— yeah, talking about, uh, open sourcing Llama 3, I think it was Llama 3. And then there's, um, I was reading about the Stability AI, like, raise that they did. And, um, I mean, basically what's happening here is like you have these kind of closed source models, now there's these open source models. And so how is something like— like, when you think about the platform risk, very much, uh, reminds me of remember everyone built on Twitter and then— or everyone built on Facebook and then they all got booted. But when these models are open source like Llama, like that platform risk doesn't really exist anymore in my, again, smooth brain understanding. So like, how is this— how is that different than like what's happening with crypto models?

Are really only one part of the stack of building an ML application. The very simple way that, you know, that we kind of talk about a lot in some of our presentations externally is we think of the AI workflow as sort of 4 levels. The first level being the compute itself. So these are the compute that run the models. This is when you train a model or you do inference on a model, actually getting the results. The second layer is the models themselves. So as you reference Llama or Stability or Mistral or whatever, new kind of model that exists. On top of that is the data. So the data that goes in and actually feeds the models, which is also, I think, a very important component of building one of these systems. And then finally is the applications. So yes, I think it's great that Facebook, Stability have released a lot of, a lot of work, you know, open source and publicly. If you go and use some of those models, if you go and use Facebook's, for example, The way that Facebook was able to get it out the door, like their legal team allowed them to release it, was they had to do this thing called red teaming, which is basically going and removing answers to racy questions. So if you ask it about a gun or a political candidate or something like that, um, it's not going to give an answer. So this is, I think, a step in the right direction, but it's not a fully open model yet because who's to say what you should and should not say? And right now, you know, it makes sense. It's a government trying to, you know, trying to do CYA on their, on the products they release. And so thinking about building an AI application, sometimes you may want an understanding of what data went into the model and how it's actually being used. You may want to use a certain class of, answers that may not be available to you. So open source is like a very important piece. And I'm glad those companies are building what they're building. But all the other factors we still, I think, need for many different use cases to be decentralized. So let's say you do have Facebook's Llama, but then you still have to run it on 1 of 5 places, a couple of cloud providers or a couple of GPU-as-a-service providers. And you don't get various guarantees that you may want, whether you're building a crypto AI application or not, privacy or computational integrity, censorship resistance, and those sorts of things. Got it.

Anand, from the investor's point of view, maybe we can get into some of the like, yeah, I guess, how do you think about investing in this space? And there's the like whenever you've got the, I don't know, Niraj, I appreciate you simplifying it for me, like the compute, the models, the data, and the applications. Anand, from the investor's point of view, like, are all of those layers exciting? Is one of those layers more exciting than the rest? How do you think about this?

Yeah, there, I mean, the space is so nascent right now that, you know, there's utility at the topmost layer with applications that can really break out and have an impact here. And of course, then you're sort of judging it with the same lens that you would sort of an application, right? Where does the defensibility lie? And typical sort of network effects will drive traction, and maybe that becomes the way in which a company breaks out. And we start to see some of these today where they're either literally putting agentic platforms to work using transparent models and/or creating companion products and using some on-chain primitives like DeFi, et cetera. So it's like interesting sprinkling of things that haven't been experimented with before, but it's still very early because when we think about applications on chain, naturally we gravitate towards DeFi, NFTs, and memes. And so like that becomes sort of like the— and you know, we have some consumer breakouts today like Farcaster and some others, but for the most part it's like, you know, we can name these off the, you know, very, very quickly. So I think it's early there. To Niraj's point, it started with like the most acute use cases, I think last year, right? Like when we started to realize that the horsepower needed to train large foundational models is being throttled and being limited because you need to either commit to spending 7 figures a month to be able to train models or you don't get access to those centralized entities at all. So like, where would you go? So naturally the painkiller use case became that, can we have decentralized compute networks to be able to train or fine-tune models? So that, we started there at the compute layer as Niraj mentioned. And then we're like, you know, now we need data. There's many different ways in which we can access data, put blockchain incentivized systems to work to access data, data marketplaces, so on and so forth. So to really back and use sort of Niraj's mental model, I would say the investments so far have been sort of like following that stack sequence very, very systematically. And so I would say compute networks, you've seen a lot of those. You mentioned a few of those as we started this podcast. And then Data has been maybe an interesting area of focus. Like I've been thinking a lot about that. And then we're now starting to unpack various application areas and Ritual is definitely, you know, far ahead here in sort of opening up the ability for applications to get built, right? So that's going to be the next interesting wave that kind of comes up as we look into the second half of this year.

Um, Neeraj, why is— you could have gone and built, you could have built a bunch of stuff in the AI world. Why is Ritual what you decided to build?

To give a little bit of back history on what I used to do at Polychain, the way that I first, you know, went to Polychain was by way of working in a cryptography lab. So I was working in a cryptography lab that did work on a bunch of different privacy cryptocurrencies, Monero, Zcash, things like that. And when I joined Polychain, especially early on, part of my job was to just go find interesting research advancements and either try to convince the founders to go and productionize it or find companies building those new advancements. And part of the thesis was new, almost like, what will developers be doing in the future and how do we predict what'll happen then? With the technologies that are, that are developing today. And so went through the gauntlet of a bunch of ideas at Polychain, you know, as early on as new layer 1 networks in 2018 to ZK and layer 2s and privacy. And so in 2022, I started paying close attention to AI because after I read the transformer paper, and started playing around with some of the work in the space, I got very similarly excited as to when I first discovered crypto. So I knew that there was probably something interesting to look at and started, you know, naturally thinking about, is there a cross-section between the two spaces here? And after a couple of months of, of research alongside my co-founder and, and some members of our team and outside the team, um, it became very clear that there was. And that this was a brand new area. And I very much like new frontier markets. Those are the most exciting for me. I prefer, you know, 1 to 0 to 1 type of bets rather than, you know, going and competing in a crowded, I don't know, crowded Layer 2 space or something like that. We really wanted to find a brand new area that developers would be using in 2 to 3 years Not to mention the kind of backdrop again of the increased centralization alongside the increased usage of AI. And so largely all of those pieces made it to be a very exciting kind of problem to work on. Big market, very difficult technology, early, and a lot of smart people that I've learned from and are building in the space.

Nice. What are the— when you think of the, like, real— I guess you have to put timescales on this. There's like, what's exciting now, what's exciting in 5 years, maybe what's exciting, like, the endless possibilities in, you know, 10 years and beyond. But when you think about the— maybe, uh, let's put a 12 to 24 month time frame on this— like, what are the possibilities that excite you guys? Like, the things that I was jotting down before this was, um, like, yeah, decentralized GPUs, kind of like the Uber Again, these are things I've, I've heard and read about, but I don't know if they're actually real that I'd like your guys' like help understanding if they're real. There's the Uber and Airbnb for GPUs. There's like doing more local and P2P inference. There's using crypto to incentivize AI behaviors. There's probably some sort of verifiability to be like, this media asset is real. There's probably like self-owned models and data. Like I've got all my data that should be worth something. And then there's like agents using stablecoins. If you had to pick some of those, like, what excites you? What's fluff? What's real? Anand, maybe I'd throw this first one to you.

Yeah, 12 to 24 months. I think one of the things that you mentioned has probably been top of mind for at least maybe in our bubbles and our circuits has been fighting deepfakes, which is sort of like the reactive measure. And the proactive measure being how do we use ZK to actually codify the source or authenticity of certain media? Because I guess somewhat the blast radius of problems that have come up with AI have been, it's awesome, we can generate images that fit certain characteristics based on certain prompts, but then again, it's really hard to discern what's real and what's not, and that media's gone out of control. So that's definitely one that's been interesting and top of mind and has been, I guess a lot of research that's gone into this, everything from audio to images and some now coming up with video. And then the reactive piece is actually quite challenging to do, but I think there's some work that's probably very much in front of us here today. I definitely feel like, I think going back to the question that you started with, right? Like the, you know, why this, why now? Some of the things, you know, so a motivating factor for me, you know, I think someone like Raj, you know, and maybe some others. I've been sort of like peripherally involved in the crypto space the last 12 years, and there have been various events that have led to adoption of crypto networks over the years. And if I were to forecast out, I'd say that with a certain set of daily active wallets or monthly active wallets, we've gone to this level of transaction velocity and change, which substantiates the TAM that we have today, which is like as of today, maybe like $2.3 trillion or whatever that is, right? So again, maybe putting a more financial hat on, like, what's going to get us to that next order of magnitude, which means that all the work that we've been putting in is actually being used massively at scale. And you combine that with the fact that I've always had a tough time connecting the dots between, like, oh, we need to get the next million, 10 million users on chain. And I think blockchain networks are beautifully suited for agents to participate, right? So now kind of let me land the plane by saying your point about agents using stables, agents paying one another, agents, you know, I shared this paper with Niraj. I'm like, this is a fascinating concept of social science, agents putting humans to work and compensating humans for doing real-world work. You know, like, and this sounds like a game or a toy, but you know, to quote Chris Dixon, right? A lot of these things start off looking like toys, but could have practical use cases down the road. So again, to summarize, I'd say agents are a big part of this. I'd say more pressing or topical today is deepfakes, fighting misinformation, disinformation. Connected to that, it's like things like proof of humanity. And yeah, you know, there's gonna be some interesting game-like experiences that come out of this as well. So I think that'll spark some innovation quickly.

Can you say more about the— agents putting humans to work idea?

Yeah, I mean, uh, at a high level, this was like, you know, some like— so this, uh, the paper does a really good job of this, and, and I hope I don't butcher too much, but how can we generate and test like social hypothesis, social scientific hypotheses using large language models, right? So they call these things like structural causal models or SCMs. So the combination of these two effectively means agents are coming up with hypotheses and saying, hey, humans, go do this work. And when you do this work off-chain, you will get compensated for it. And then that information is kind of relayed back for agents to assimilate information from the real world that they can't otherwise get because everything they're going to be doing happens within the silo of online and on-chain. So it's a really interesting, like, you know, reinforcement learning concept that really hasn't been done yet. And by the way, it's gotten a fair amount of critique as well. Like, it's a, It's something that can really get out of control very quickly. But again, it's like you're, you know, at some point these systems are going to get smart enough to be able to put us to work. By the way, maybe not too dissimilar. I remember talking to Ilya from Near about this maybe about a year ago. And his sort of big quote was that at some point we'll all be working for AI or agentic systems, like we'll be doing the work. Whereas then I think his theory was that like maybe there's layers of management that get replaced by AI, right? And so Maybe this isn't too far out. Like, I think what I'm proposing, what this paper is proposing, is more of a science experiment, appears more game-like. But yeah, there are ways in which, yes, you know, there may be tools that tell us like, by the way, this is sort of what your day looks like, hit these goals and these milestones and you get compensated. Yeah.

The two folks who have talked about that on the Empire Podcast are Ilya is one of them. And then we had Rune from MakerDAO talking about that as well. So, Niraj, what are your thoughts on this whole AI agents, stablecoins? When are AI agents going to be built? Actually, can you just almost break down like what AI agents are? I have an idea of what they are, but can you, can you break it down? Sure.

Everyone has their own, their own answer to this question. But the way that I would describe it at a high level to like give a framework for what is agents are pieces of software or computer systems. It could be a DAO. It's basically a self-contained entity that is able to take actions on the behalf of someone or multiple someones or multiple people without the user explicitly programming them to do that task. So a very simple agent could be let's pretend that I run an office with 20 people here in New York, and an agent might look like, if the bot sees that I have meetings with 11 people today, maybe it'll get me, it'll order Starbucks for me in the morning for 11 people and order to the office before they all come in, something like that. And that's something you could do today with OpenAI and stringing together a couple of APIs. And so that I think can be very valuable in a crypto context. You're now able to apply it to various different actions in a crypto context. You know, one might be an agent might be able to tell you that, hey, your collateral in your MakerDAO position is getting low and, you know, you should top it up, or, you know, I'm happy to top it up for you. Or, hey, your 6 friends are buying this NFT, you might want to buy it. And a whole bunch of other things. It allows you to almost be predictive about the tasks you're going to do, and someone can execute it on your behalf. So that's how I would generally kind of slice and dice the agent's question, at least at a high level.

That's interesting. What are other— Niraj, what are others that you are— let's go to decentralized GPUs. Tell me, real, real exciting or not exciting?

I find it exciting. So, yeah, so this is the— I call it GPU as a service companies, or a lot of people call it Airbnb for GPUs.

Is this Akash?

Yeah, I would say Akash probably fits into this. This is maybe one of the bigger areas of crypto AI in terms of number of teams. Really, the people are here operating on different theses as to why this is an interesting category. Some people are saying the big GPU-as-a-service, centralized GPU-as-a-service companies companies like Lambda Labs or CoreWeave are good, but you can't get access to enough GPUs. So there's a supply crunch, or they're only offering certain classes of GPUs, things like A100s and H100s, which are the kind of upper end, you know, last generation NVIDIA enterprise chips that are widely used for both training and for inference. So they're saying, hey, there's a shortage and it's hard to get access. Maybe you should start using lower-grade GPUs, of which we can bridge into a bigger supply, whether it's because, you know, it's a crypto thing and a bunch of ETH miners used to use GPUs to mine ETH, or people are proving, you know, ZK circuits with GPUs or doing FHE workloads with GPUs and a whole bunch of other stuff. So GPUs are these like special-purpose computers that are really good executing very simple tasks in parallel. And so in a, in a ZK system, it's doing these, these multi-scaler multiplications or fast Fourier transforms. And in machine learning, they are doing basically matrix multiplications. And so you can reuse that hardware in the, you know, in the GPU-as-a-service context. The big thing to win here is that AI is growing very massively. So there's a very big market to win here. If you can cut out a piece of the AWS, GCP, they're building more of a hosted type of solution or Lambda Labs CoreWeave, which is, they say bare metal is what it's called. There's a pretty big market to win. We're very early. I think one of the challenges is that Certainly you can incentivize new supply to come aboard and bring new devices on. But bridging the Web2 demand to the Web3 supply, I think, is like one of the bigger questions. And so I think over the next year, we're going to see how that starts to play out. I don't want to make a prejudgment because it's very early. A lot of these networks didn't even exist 2 years ago. And this is a problem that we're trying to solve. We plug into GPU-as-a-service companies and are basically trying to generate new forms of demand for, you know, ML workloads basically on-chain and create crypto-native AI applications or AI-native crypto applications in whichever direction you want to take it.

This episode is brought to you by Aura. Online crime is growing at an alarming rate, and cybercriminals continue to find crypto investors like the Empire community lucrative targets. We've all heard of someone who's fallen victim to some sort of online attack. Aura helps prevent against that. For example, if your online accounts or your passwords were leaked online or someone tries to open a bank account in your name, you'll get notified in real time about that. With 24/7 US-based support, Aura's team will be there for you to resolve fraud issues, even if it means getting on a three-way call with your bank at midnight. Not saying I had to do that, but I might have had to do that. So aura.com/blockwork, sign up for the 14-day free trial. Terms and conditions apply. Site for details. Aura is the new standard in digital safety. Check them out, aura.com/blockworks, free 14-day trial. This episode is brought to you by Kinto. Kinto is a safety-first layer 2 that features user-owned KYC and native account abstraction to solve what are probably the two biggest problems in crypto: one, security, and two, user experience. I'm an investor in Kinto. I've known their founder Ramon for a long time. Ramon the other day sent me a message on Telegram and said, you got to try this onboarding. I'm super proud of it. Please just go ahead and try this out. Tell me what you think. It is the most magical, seamless onboarding experience onto an L2 that I've ever had. You can, you can try it out. It's engen.xyz. I would really recommend going through the whole onboarding flow just to see what the future of crypto can feel like and what it probably will look like in the future. Go onboard, engen.kinto.xyz. Let me know what you think. And if you want to chat with their founder, Ramon, happy to make the intro. Everyone, Jason Yanowitz here. I know you've been hearing a lot about the intersection of crypto and AI, and that's why I'm really excited to share our newest partner and sponsor at Empire, SKALE, a high-performance modular blockchain that's revolutionizing gaming and AI with zero fees and instant finality. SKALE's unique architecture allows for massive scalability and has already saved users over $6 billion in gas fees. The 5 main takeaways that you need to know for SKALE: 1, zero gas fees. 2, SKALE has instant finality and lack of MEV. 3, the zero gas fee model of SKALE is really important for blockchain gaming adoption as gamers don't have to pay transaction transaction costs or have the SKL token. 4, SKALE is multi-chain. This design allows for nodes to be combined to create chains and for individual subnodes to actually be removed and relocated, which creates greater security and collusion resistance. And 5, SKALE has become a really good fit for AI dApps because they're fast, automated, zero-fee transactions. Big thank you to SKALE for sponsoring Empire. We're excited to partner with you guys. Bridge over to SKALE if you're listening to this. @scale.space/ecosystem. Follow the journey along with SKALE on X @skalenetwork. That's @skalenetwork. Big thanks again to SKALE for sponsoring Empire. Remember in 2020 and DeFi summer and then starting in 2021, everyone's like, all right, great. Like DeFi, DeFi is booming. The institutions are coming, the institutions are going to play in DeFi and they never really did come. And that was fine though. That was fine. And I think there's always this, like, aspect of crypto where people love to say, like, that big group of people outside of crypto, they're coming in. But in reality, the best products that get built are for deeply crypto-native folks. And it's an industry that's built very bottoms up. I'd be curious to get both of your takes on, like, do we need— will crypto— I've heard one side of the argument that says, like, AI will be the thing that helps crypto cross the chasm. Because, yeah, Google Bard is like, I don't know, censoring X, Y, and Z, or I don't know, there's these— we can't tell if the proof of humanity or something like that. And like, Worldcoin is going to help with this. Or the other side of the argument is, look, it's all going to be crypto native. It's crypto native all the way up. How do you guys think about that?

Yeah, I'll take a stab at it and maybe, you know, having spent time with Niraj and team Ritual, but also folks like Jensen, for example, who also, they weren't necessarily crypto native, or the team at Sahara, which is like an academic team that focused on AI and NLP prior to this, or Synthify from Austin. I mean, these folks are all mostly AI native. And so I'd say that, I mean, just my mental framework for like what is going to will we get this utility happening outside of the crypto world? For me, I think it needs to start with talent that believes that the primitives that exist in crypto are the ones that really exist for us to build a beautiful decentralized AI network, right? So the fact that this talent's kind of coming in waves, which makes me feel really happy, is I think probably the first way to get started. We're seeing sort of like net new talent flowing in. Folks are like new to crypto. But have incredible AI backgrounds. I think that's like the first thing. And the second thing, you know, again, like the kind of tools that the likes of Ritual and others are building are really for to enable developers to build things on top of their stack, right? And at steady state, my hope is that the average Web2 developer, and again, this is yet to be proven, and, you know, Raj and I talk about this quite a bit, which is like, how do we get the average developer who needs to use, who needs to perform some kind of inference, how do they use a stack when they don't really care about the underlying mechanics, whether that's a crypto-incentivized, decentralized, deepened network or it's a centralized entity? Like they shouldn't have to care about this, right? So I think that's sort of like the end state we want to get to. Also, like lastly, I'll say this by going back to talent and maybe slightly more philosophical, but I'm going to butcher this because I think Mark Andreessen said this. Very well in a very concise way, but it was something like the best founders out there sort of will their perception into reality. And when I first started talking to Raj about what he was thinking behind Rituals, like, if I don't do this now, I mean, this is my calling. I need to go do this and I'm going to make this happen because this needs to exist. And if I don't do this, I'm sure someone else will, but I want to be a part of this wave that makes this happen. And I think with the kind of talent that's coming in, which is creating this beautiful network effect of more talent coming into crypto. I feel like we're gonna get this steady state where like this layer as to what, where, what stack this is running on is gonna get obfuscated. And but hasn't happened yet, but that's sort of the challenge that's in front of us. And I feel like we have the right people who will tackle this. So a more philosophical sentiment on why I'm bullish on the sector.

Yeah, I would agree with a lot of that. I think, The way that you framed it is exactly right. It's not often that you can just take crypto and fit it into a product and then new users will come. It does have to be a more bottoms-up, kind of within the community emergent approach. And if you go back and look at when Ethereum first came out, many people couldn't even begin to guess what the use cases would be. And 3, 4 years later, we had DeFi, we had NFTs, and even 4 years past that, or 8 years past that, it's such a wide market. And I think as we're laying down these very early primitives, connecting smart contracts to blockchains and decentralized training and decentralized verifiability, privacy, I think we're going to see some very exciting emergent behaviors start to come about. On the application side. And I think the crypto user base is a very good user base to build for first in many ways. It's a really good way if you're trying to build something important in AI. It's like a brand new tact rather than use the same distribution path that everyone in AI is using. Go and talk to a brand new set of users that tend to spend more money on products. Tend to be at the early days, maybe more critical of the products they use. So you can really test if you're building something valuable, as if people are using what you're building, and then ultimately seeing for the emergent properties of what you're building, if there's something valuable that users can take away. And so certainly that's like a very big focus for us, is making sure that we can again cross the chasm, have that ChatGPT moment for crypto, And maybe AI is part of that story for, for, you know, ultimate crypto adoption.

Hmm.

Who will— what do you think that moment will be? Well, you remember, remember 2018 and 2019 after ICOs and everyone basically said, what is ETH good for anymore? And the price fell from what, $1,700 or $1,800 all the way to the low of like $85. And everyone's just like, yeah, maybe ETH isn't good for anything. And then DeFi summer hit and we said, ah, ETH is good for something. What is the, what is the like DeFi summer moment going to be for, for what are we calling it, DEAI or something for decentralized AI?

I think the first, the first thing will just be being able to play with it. Crypto AI is unique in that DeFi summer happened after we had DeFi smart contracts for like 3 years. I think Dharma launched in 2017 and we had Ethland and Compound came out in 2018. So there's a DYDX, there's a whole bunch of things that started then. And I think it took the combination of the creation of basically liquidity mining from Compound and the launch of Uniswap, the growth of the long tail, to really carve out a market to where DeFi was a valuable asset. When we take the same lens that you used before around DeFi, institutions didn't really use— I think people originally, the thesis of DeFi was it's like a brand new, from bottoms up, correct by construction financial system. And so it should just be net better for the traditional financial system to use. But the traditional financial system is so full of, you know, outside of just the technology, there's lobbying in place and broken systems and relationships that DeFi is not going to overtake it overnight. I think the better way is the Trojan horse approach, which is make the crypto user base really, really wealthy, influential, using a lot of products. Basically, they are their own sort of force. And then now they're already using DeFi and they don't need to be converted into anything. They're already just a natural user of DeFi. And for something like AI, largely, I do think one of these early use cases will become quite big. Agents, to me, is a very exciting opportunity that we have in front of us. In the Web2 space, this is also what people are very excited about on the traditional venture side. We've talked about this example a few times, which is being able to interact with blockchains with natural language will become a very, very interesting and valuable crypto-native thing for users to be able to do. So an example that I usually give is, let's say that you wanted to take out 3x levered exposure to Apple, with DAI as collateral. And then at some price point, you should stop out and close out the position to have some safety or something around it. Normally, to do that in DeFi, it would take like 10 different smart contracts and clicking a bunch of interfaces. But I said that to you in one sentence, and I think the power is being able to take that one sentence and convert it into some sort of form of intent. And there's a bunch of companies building on the intent side and then being able to now interact with blockchains in a way that's actually usable by regular people and not just people that are very in it. And so I think being able to work along the side of agentic, sort of emergent applications, as well as being able to interact with blockchains in an entirely new way, are going to be the pathway towards finding the emergent, you know, DAI use cases.

Anand, you were nodding your head there. You want to?

Yeah, I mean, it's so well put. I mean, I think the— I don't think this is an or, like it's not DeFi or some decentralized AI application. I think it's an and, right? You have— I think what Niraj succinctly said is that these sorts of LLMs and LLMs could be the UX unlock we're looking for, right? You combine Intents, which is sort of a framework that exists today as a primitive on Ethereum. You combine that with natural language and you're able to create something beautiful. And then furthermore, if you have like agentic systems that can truly process who you are, your risk threshold, your ability to put certain amount of capital to work and make that happen for you, then you have this really interesting flywheel happening. But this is all sort of converging today. That could be, that's like I went narrow into one use case or like putting DeFi to work with LLMs, with agents, but then like there's other possibilities that could also substantiate this, right? We touched on this briefly, things like where private data and the ability to perform actions on it will start to matter, right? Things like privacy preserving machine learning, using encryption techniques like fully homomorphic encryption on such data and being able to use the blockchain for consensus. These are all sort of things that I think will augment and start to unlock utility in ways which, you know, they may not have the consumer feel to it, but I think we'll have some enterprise-grade feel that is going to unlock some net new opportunities. So these are also exciting areas that I've been looking at.

Yeah. Who's using— what non-crypto companies are using crypto AI companies' products right now?

I can name quite a few. So like, you know, there's a really interesting integration that happened with this company called BrevDev. Basically, they give you the ability to spin up like Jupyter Notebooks, which are essential forms to start to do machine learning training of data. So BrevDev is unlocking the ability for any developer to use Akasha's decentralized compute network to be able to put their GPUs to work. So this is kind of like multi-tiered in some ways, right? So developers using BrevDev, BrevDev's got the ability to spin up H100s that are running on the Akash network. I think that's number one. And I've seen all the way from that to various inference systems. Hey, by the way, I think Niraj alluded to this earlier, right? There are certain things that have been extracted away from certain models today because of which you can't do certain things, right? Whether it's not safe for work stuff all the way to specific nuanced utility of models for something that may be very, very hyper-specific. So I have seen examples of where like, you know, I want to create my specific companion or character that might not be, you know, centralized entity AI kosher today. How do I create or spin up something like that where I can have a companion, whether I want to share personal details with them? This is my, AI waifu. Those are the kind of scenarios we've seen crop up. So I just could— I give you a little bit of a spectrum there. On the latter, I'd say there's quite a few of these. The names sort of escape me, but at least one example where developers are putting, you know, some of the backend systems to work today.

What is this company mentioned in the notes?

NUS?

N-O-U-S? Nouse?

Oh yeah, yeah. Nouse Research is another interesting company. Uh, they are— yeah, they're a bunch of like AI experts who've been EHOPE, who spun up a Big Tensor subnet, but they're primarily in the field of synthetic data generation and being able to get LLMs to become smarter because of the work they're doing in creating synthetic data. But yeah, another example of talent that's come in to work on crypto in a very credible way. And let's say a research firm primarily today.

Can I run through some of the projects that I see on the timeline all the time? And I'd love to get your guys' takes on them. I don't know how much you guys can, can share. If you guys— I'm sure there's backstories with a lot of these companies, but, um, all right, I see Martin Shkreli posting all the time about IONET. What is IONET? Real, fake? Tell me about IONET.

So IONET is building a way to plug in. In effect, it's a GPU-as-a-service network. It allows you to plug into this network of GPUs that supposedly come from a number of different data centers and you can then use inside of your applications. So in some ways you can think of it as a drop-in replacement to Lambda or something like that. I think the controversy was around fraudulent GPUs on the network. And Martin, I don't know if he's building in AI crypto, but I know he's very interested in crypto and at least building adjacently to it. And I think he had issues setting it up and was kind of diving deep into the backstory of IONET and things like that. I haven't followed too closely, but that's what I've learned and read so far.

How about— go ahead, Anand.

Yeah, no, I was just going to say that this problem set, it's a really interesting topic that got unraveled in some ways, which is The beauty of sort of decentralized networks, there's a little bit of a double-edged sword, right? When you're joining a permissionless network, either on the supply side, or if you're on the demand side like Martin was trying to put some of these networks to work, and you need to have this level of transparency that exists, but it got a little out of control. All this to say that maybe a call to action is around sort of reputation systems or being able to really understand what's happening in these decentralized physical infrastructure networks. Is going to be probably going to start to get paramount, right? So that was sort of a big takeaway for me because when it comes to using a central, using AWS, there is an SLA, there's terms of service on like what you pay is what you get. With decentralized networks, we need more transparency here that needs to emerge. So that was sort of like, I'm glad we ran into this problem today because there's more work to be done to ensure that we're providing a level of transparency that doesn't exist yet.

How about BitTensor?

Yeah, I, I could start here. BitTensor is, uh, is fascinating. It's— I would say it's a marketplace for generalized compute. Um, that's sort of like maybe the high-level tagline. Um, what's become interesting is, you know, within BitTensor you have these various subnets of which there's like a limited set that you can have today. And each of these subnets has a very specific purpose, uh, for, for why it exists in the BitTensor ecosystem. It can be anything from file storage to decentralized compute, to very specific inference. Hey, provide a prompt and get audio back, something along those lines. And each of these subnets is very opinionated about the kind of work that exists in that realm. So I would say that, and I'm sure Niraj and the Polychain team have more background in sort of like the genesis story of Bittensor. But I would say again, maybe a little, it's been really fascinating because a lot of folks have gravitated towards it. Being able to spin up these subnets and actually start to put various resources to work. Noose is one of them. We've had prediction subnets that have come up. Now we've got companies like Virtual launching agenting systems on there. So yeah, it's a really interesting test harness for a lot of sort of AI crypto projects. And yeah, at the core of all this is sort of like their blockchain consensus system. So I hope I did it justice, but it's just I would think of it as like, you know, several different sort of nation states that have a consensus system behind them, which is the BitTensor consensus network.

I won't play 20 Questions asking you guys about every AI crypto company. But Niraj, as someone who's like deeply building in this space, and I'm over here just scrolling on Twitter about it, what feels very obvious to you about the intersection of crypto and AI that maybe doesn't feel as obvious to most of the rest of the world yet?

I think one of the big— I don't know if they're mistakes, but one of the big oversights that a lot of crypto VCs made over the past one year about crypto AI is they kind of— they basically looked at the previous attempts at crypto AI in 2017 and 2021 and basically wrote off the category. If you went to any competent crypto VC even 1.5 months ago, I mean, not all of them, but most of them hadn't looked at the space at all. If you were to ask them, what are you doing in crypto AI? They would have said, oh, there's nothing interesting there. That's just a meme. Those are just meme coins. And you fast forward a year later and every single fund is, is now they have a crypto AI strategy and they're writing articles about it. So I think, you know, in some ways almost the pride of the VCs made them overlook what was a very exciting opportunity to invest. I still think there's a lot of exciting things to come from now, but that again, my philosophy in crypto investing is that you know, you actually do want to be early. You know, it's not like traditional venture where you're seeing the present clearly is the most important thing. I think in crypto, given how cyclical it is and how herd-driven it is and narrative-driven it is, you being early is being on time. And so that, I think, is one of the bigger mistakes or oversights that I saw that people didn't get right.

Anand, you have anything for that question?

Yeah, I would say that, I mean, maybe this is a little bit further down in the weeds, right? Which is if you think about various topics and conflating various different things, you know, so like in the spirit of what you asked, I'd say I see open source and decentralization getting conflated for one another quite a bit, right? Which is like, I would say we really embody open source. And of course, let me, sorry, let me unpack that even more. Like when it comes to things like LLMs, there's the difference in open source and open weights. It says open source, open weights, decentralization. So I think we latch onto some narratives very quickly. So in terms of what we overlook, I would say that the core tenets of decentralization are almost essential to me for for AI to be functioning and for us to promote research and bring in diversity of thought and ensure that we're not sort of like tethered or tied down by the confines and guardrails of like big tech, right? So sorry to keep parroting this narrative, but I think that's important. So that's decentralization primitives. On the other hand, open source is truly like the tailwinds that are necessary for this innovation to happen, right? And the one thing that I'm starting to see a lot more of, you know, just coming back to talent, open source, you know, historically, for as long as, you know, you've been a programmer, you've understood open source, you've understood licensing behind it. But when these two forces come together, open source plus decentralization, it's actually like a very, very powerful tool. And so there's inherent questions today, you know, I've written a lot of open source code. And I've had questions like, hey, you know, is how does— or like Vads have contributed a bunch like Stack Overflow, which is now obviously, you know, that technology been superseded by LLMs. But how do you get compensated for that? Like what is the framework that you use to compensate really credible developers who are building things in open or answering questions, et cetera? But now we're having these primitives come together. You're being rewarded for your work. You can, You can launch a new fine-tuned model on chain and get compensated for it. And so these things are getting unlocked. So come back to your question, sir. A long-winded way of saying what we're overseeing, I think we think we latch on to narratives like what is the utility here? But to me, as we look at sort of paradigm shifts over the last 20 years, when these sorts of platforms get built, whether it's cloud, whether it's mobile, it's blockchains, when developers flock in and they start to see a sandbox and they're like, wow, we can do something creative here, we can monetize this. Unlike anything we've ever seen before, suddenly that just opens up an application space unlike anything else, right? So we saw this, obviously, you know, several examples of this over the last couple of decades. So this is, this gets me a lot of optimism. And to Niraj's point, and feels, it felt contrarian maybe a few years ago, it still feels contrarian now, I would say, Niraj, even to your point, I still see a lot of investors like, yeah, this is thematically, it feels like something we're just latching on to because, you know, of X, Y, and Z, but I just see an opportunity here to go build some really cool shit that hasn't been done before.

Love it. That's great. The data in these LLMs is like, I mean, it's like the two things that are working right now are the, the like NLP, which I guess NLP is, you know, now it's just these, like it used to be called NLP and now it's the big language models. So it's like it's text and then it's computer vision. It's the CV side of things. What do you think about and on the language side of on the, on these large language models, it's like, so it's a lot of the, a lot of the data is coming from either like social platforms or just like the internet, basically. Niraj, what do you think about leveraging the amount of data that the blockchains are spitting off? Like, the data stack today, I mean, like the L1s, like L1s and L2s, like, I mean, spit off boatloads of data. And then, you know, Alchemy and Infura and Quicknode and stuff clean it. And then you've got the API providers and then folks that sit on top of those and stuff like that. But yeah, how do you think about using like blockchain-native data in some of these AI systems and like what will come from that?

I certainly think there's a lot of interesting user applications for using blockchain data inside of an LLM. We talked about the example of the human language interface. And so that would take advantage of smart contract data and block data that you could then leverage inside of an application. I even think using crypto as a substrate to gather more data will become very valuable. There are companies building in the label-to-earn category you could call it, or contributors are basically rating if models are good or not, like the responses from the models are good or not, and then using that to basically reinforce the decision-making of that model. And so that is all to say, there's so much interesting surface area to build in. There's a company called VANA, V-A-N-A, that is trying to basically do an experiment around paying users for their data. So they're creating this subreddit called— I think it's called DataDAO, and they're experimenting with ways to pay people to submit their data. Because one of the challenges, you know, the New York Times is in a lawsuit with OpenAI over the use of, you know, their writings inside of the OpenAI GPT-4 system. And there's this big question around how do you do attribution and how do you pay someone, you know, if you do use their data in your model. And, you know, these, these systems are lacking access to our personal user data. So if you can bridge and, you know, allow people to get compensated for bringing their Facebook data or their Spotify data both Web2 data, but then also, you know, back to your point, Web3 data. You know, if I could use my Forecaster data or my Sona data or, you know, whatever else, new, you know, social system inside of a crypto LLM, there I think would be a lot of interesting use cases around recommendation and ultimately interfacing with these systems.

Raj, if you guys are wildly successful, like let's say everything goes right, for 5 years in a row. What is the— how is crypto different or how is the world different if Ritual just nails everything for the next 5 years?

I think if we nail everything, decentralized AI will be a real credible alternative to centralized AI because I think the backdrop today is that centralized AI is going to win. It's a monopoly business. They have all the resources and the talent. And the attention and the regulatory capture. So how could crypto AI win? And I think there's a lot of really interesting approaches happening in the crypto landscape. I like the direction that we're taking to go about on this pathway. And in 5 years, people are pulling models from decentralized AI systems. They're building on decentralized AI platforms. They're contributing data into decentralized AI systems and run the whole gamut of mix and match for building products. So ultimately, I think it's pushing towards world domination. We didn't get into it this podcast, really, but if AI continues to permeate the market as it has, we're going to have a billion people using AI in 3 to 4 years. Inevitably it will be used to manipulate people or to create propaganda. And those things will be used to tilt people's opinions of public favor and governments and corporations, how they spend their money, who they interact with, what they do. And so ultimately it's both like we're fighting it from like a market standpoint, like there's some better properties here, it's cheaper, whatever. But there's also a very fundamental values difference that I think is very important. And I hope that, you know, we can have, you know, help the world understand the kind of underlying problems that exist today.

Anand, on that note, I saw you put in the notes, agents could drive more transaction volume on-chain than humans ever have. Yeah, would love to just hear you expand on this.

Yeah, you know, I think this is the, you know, we were chatting about this a little bit earlier, right? We, in some ways, we have this like complex, you know, user experience, customer experience problem with blockchains as we know them today. You have to install a wallet, you need to know how to transfer fiat to crypto to get onto a chain to be able to interact with DeFi. It's like, you know, doing one of these things while hopping on one foot and like, you know, it's complicated, right? So, and Raj said this really well, right? How can we unlock the ability to use natural language plus intents and with LLMs to be able to do things on chain? And then, you know, maybe taking one step further from there, you know, can these agents act on our behalf and to do some credible work? But that's again, assisting humans in one way, because again, you know, as Niraj said, there's many different definitions of agents, but at steady state, there could be many, many such agents for a single human being that are doing various things for you. Whether it's activity on-chain or whether they're doing some work off-chain for which we just need verifiability, and therefore they need to run on decentralized networks. And so, and then of course, they could be tasking one another, they could be doing things, they could be putting humans to work, as we talked about on this podcast. And for all these reasons, the transaction velocity is going to go up. And so the metrics for me are, you know, how do we drive transaction velocity? And then commensurately, that increases crypto TAM. And I think in some ways, we're kind of born into this world of crypto and believing that number continues to go up. But obviously, the price going up is one way in which the TAM goes up, but transaction velocity is the second piece. And so if agents truly start to interact and put the blockchain to work in ways in which humans have not been able to do so, do so far, that substantiates the, you know, it realizes the true utility of this network. So that's what keeps me— again, there's a lot of work to do here, and I'm more optimistic than not that we can make that happen. So that's the steady state world that I see in front of us.

Maybe we can just end with a closing thought to, like, tie all this together, which is kind of what we started with in a sense. But I'd ask both you guys the same question to close this. So either one of you guys can start, but if someone, someone's listening to this, they're like, I think I get it, but give me the summary. Like, why does AI need a blockchain? How do you think about that?

So the summary that I would give is AI and crypto have these bidirectional benefits to which you get when you, when you combine them. And they're technologies that are both entering some degree of maturity to which it makes sense to combine them and creates these interesting emergent properties. Crypto provides you tools for self-sovereignty. It's ultimately a way for you to control your data, your assets, your identity, and use that in a portable way across a wider range of applications— DeFi, NFTs, gaming, etc. And we've seen many interesting benefits for doing that and users in many of those different categories. And then we have AI. AI is this tool today that provides you infinite abundance. This, you know, you know, it's intelligence. It's giving you the ability to ask a question about anything and expect a reasonably intelligent response back. Maybe you can't ask it a highly, you know, deep, you know, PhD-level physics question, but you can ask it, you know, really anything else. You can ask it for, you know, help with writing a letter or accounting or really anything else. And so being able to combine those things, you take this highly intelligent, highly powerful new capability and combine it with these guardrails that now allow you to give power back to the users and create AI systems that are owned by the users, you can incentivize different pieces of the stack to help grow the, the networks themselves, whether it's incentivizing people at the compute layer, or at the models layer, or at the data layer, or ultimately at the application layer. And You know, crypto can definitely help AI in providing these guardrails. And I think what's emerging is AI being able to help crypto. And we've talked about some of the use cases, natural language interfaces, Web3 agents, others are smart DeFi, autonomously generated NFTs, unique gaming experiences that are generated in real time, and a whole bunch of others. So that would be generally my thesis on AI crypto and what I find interesting about it at a high level.

Yeah, the first sort of instinctive reaction that comes to mind is obviously like, you kind of like, who are you talking to? Right? Like, instinctively, I was like, you shouldn't have to care. You don't want to have the average person thinking about what's the underlying system here that's driving my what is the stack built on, right? Like today we conduct queries on the internet and like whether it's Google, whatever the case might be, but we don't really think about the stack behind it, but it is going to start to matter. And that's where sort of the rubber meets the road, right? So I was thinking back, like I think there's a book called The Diamond Age, right? And I think in this book, humans are paired with an AI device from birth. It's a science fiction book. Now, you know, we don't have these problems that exist today, but I think the Asimov quote is something like, "Today's science fiction is tomorrow's science fact," right? So I think very soon we're going to have these sorts of devices. This is already happening. We have these hardware devices that are coming out every day, and we're going to start to increasingly rely on them. I mean, I use full self-driving on my Tesla, you know, almost invariably every single day. And so AI systems are going to permeate our lives very, very closely. So if that's sort of the backdrop and we're birthed with, you know, my daughter's 11, she's going to start, you know, she uses Perplexity or Sibyl and like these kinds of like augmenting search engines more than, you know, I think any other system today. So the real question is, you know, in front of us, like, do we want to let these AI systems be run by big centralized entities where the profit systems are monopolistic in nature and they're not clear as to how they're using our data? Like, do we want to continue to perpetuate the problems that have existed in the past? And I think that's sort of like the underlying issue we're trying to prevent with these systems. So I would say for the average user, probably doesn't care, but it will start to become important when we're talking about just artificial intelligence, right, which will start to permeate our lives more and more. So that is where I hope— again, I don't mean to sound like doom and gloom here because things can go really bad, but I think the opportunity for us to build things and create these guardrails, as Niraj mentioned, because these primitives is going to start to become more important. So let's start with those as the fundamental building blocks as opposed to, again, power aggregating in a centralized fashion. And I think we'll have, we'll kind of like create a future that we're all sort of proud to live in. So that's the call to action for me.

I love it. Good place to end. Anand, Niraj. Thank you guys, appreciate it.

Thank you so much.

Yeah, appreciate it.