Distributed Intelligence: How Crypto x AI Really Converge

Transcript

Thank you. Uh, thanks for making it uh, bright and early this morning. We've got 15 minutes and we're in Switzerland, so we're going to be on time. Um, where's my thing? Here we go. Oh, some people are late. This is not very Swiss of you. So, um, about a year ago, well, two years ago, I started this fund and about a year ago when I was starting to talk about our thesis, I kept getting asked this question. So, what's the intersection? What's the convergence? Tell me like where the gold is. And I was like, okay. And it was kind of a complicated thing to explain. Um, and over the last year, we've done uh we've done our now 10 investments in this fund. Um, it's becoming clearer to me how to explain this and also where uh

this convergence is really happening. Um, this is me. This is why I think I know what I'm talking about. Um, I did a PhD in AI in 97 at Cambridge. We actually were working on the same kind of technology that I use today. There's been a number of advancements in the field. Obviously most strong is like we actually have data now and uh obviously we have incredible compute um you know orders of magnitude more than we had before. Uh I've built companies, sold companies and distributed systems, started one of the first institutional funds in crypto. Um built my own crypto project. I'm trying to use the word company now to describe things we do in crypto rather than projects because it's kind of kind of sounds like high school.

Um otherwise um and then most recently started Nazare. Um in case you don't know what Nazare means, Nazare is a little town uh just northwest of Lisbon on the coast of Portugal where the largest surfable waves in the world exist. And the rest of the year there's no waves. It's not a surf spot. There are surf spots further north in Osiris, but Nazare once a year has the largest surfable waves in the world. So if you want to there's a great HBO documentary called the 100 foot wave which talks about this. So back in the beginning um we had uh crypto mining. So proof of work was everything everything was proof of work uh before proof of stake. Um Bitcoin was initially mined on well first on laptops and then GPUs and then eventually on AS6. Um Ethereum carried

on being mined on GPUs and this caused an amazing uh sort of surge in um amassing GPUs and really bankrolled a lot of Nvidia's early progress. Um and then all of a sudden, uh we switch to proof of stake in Ethereum and many other projects and there's just tons of GPU sitting around. So what happens next? Well, AI happens next. So at the same time as Ethereum is switching, it's like some kind of like uh simulation, right? At the same time as Ethereum switches down to proof of stake and all the GPUs are effectively sitting around idle. um the AI space starts coming up and in 2016 uh one of the big innovations was the idea that you could use GPUs to train these these models

very effectively. Um and that has led to now uh a very large demand for GPU space. Um you may not know this but Coreweave is one of the most successful crypto companies ever. So Kore started as a as a crypto miner, initially focused on Ethereum and then pivoted around about the time of 2019 and also as they realized Ethereum was demand was going to go down um and got a massive injection of capital from Nvidia to focus on um AI, right? So Corwe is now like just went public um I think at 19 billion and is one of the dominant players in the GPU cloud space.

Um so a lot of the initial sort of drive from crypto ended up financing AI infrastructure. So it's almost like the the the sort of uh the the roads the pipelines were laid for what happened next. And then at the same time and FTX was going down AI was coming up. So around about that same year 2022 when everything just got really bad. At the end of 2022 is when chat GPT 3.5 was released. And um so a lot of excitement happened around this period in AI. At the same time as a lot of really really bad news happened in crypto like I was wonder I was wondering if we were going to make it. Um and after being in the space for a very long time I like many

people many people thought about leaving. Um I stuck around. Um I'm I'm still in crypto. I'm on the board of various companies and uh still building things but I've also kind of pivoted to AI. So Jason Kalcanis uh famously posted this um and he you know it's been uh I think there's actually like another repost of this. He he did it twice but um I wrote a recent article which is like if you're in view in crypto pivot to AI again um because in my mind there's only two buckets left. There's only two product market fits happening currently and I'm very focused on product market fit after um what is it now 12 almost 13 years in crypto. We've

definitely got it in finance. If if you if you've been asleep, finance is happening. Finance and I'm not talking DeFi. I'm talking finance. Finance is absorbing DeFi. DeFi is becoming the rails for finance. Um, we've got spaxs, we've got ETFs, we've got all these different opportunities, real world assets, but it's finance. In between these two areas is like this sort of valley of despair. And I'm talking like NFTTS, identity, like gaming, like it's really sad like, and there's lots of potentially very interesting things to invest in at this point because, you know, you should be counteryclical. Um, and then AI. So, what the hell's going on AI? Um, so DeFi is basically finance engineering. It's just like, you know, trying to find ways to just make what, if you've ever dug into it or

ever tried to do a wire, you realize it's just like a bunch of paper moving hands. And this is one of the the first ideas in in crypto is like, okay, you've got Bitcoin, the store of value, but then what else could you do with blockchain? And even before Ethereum or anybody built smart contracts, we were all thinking, okay, well, maybe we can like try and uh smooth out this stuff and we can find like much more efficient ways to do things. And we absolutely have product that could fit right now. And so this is where like tons of the dollars are going into all the investments going into all the talents going into. Um and this is what we thought we were building, right? Like back when I was investing in companies and hanging out in Berlin and like sort of 2017 18, we thought we were going to

change the world. It's like we're building this world computer. We're like building these decentralized things. People had a mission. People were looking at this saying I'm here to change the world. Um but what we actually ended up building was some very interesting things some very interesting technologies like forget all the tokens forget all the price go up everything else. So a lot of these components here so first of all wallets and coordination um privacy and verifiability which is like we'll get into a second and then this kind of like frictionless on frictionless markets people call this like internet capital markets today and then of course programmable composal value. Now, what these give you are the components that you need for a decentralized AI framework and and many other things whether it's decentralized or otherwise.

And you know, back in 2013 when I first got into the space, all the weirdos were in Bitcoin. All the weirdos were in like I was just Bitcoin back then and forget about, you know, crypto. We didn't have that name yet. We didn't know we didn't know what to call it. So, it was just Bitcoin and other things. And now all the weirdos are in AI. Like sure there's like you know the big companies and so on and so forth but there are kids in their 20s sorry sorry everyone in their 20s but like kids in their 20s who believe they're building a super intelligence like collectively they they think that they're bringing something into the world and they're doing it not because you know number go up that's not their primary thing. sure that they're they're economic, they're capitalists, they're pulling startups, they want to

like, you know, do something else with their time in the future, but they're doing it because they really believe in something and it's very powerful, right? So, this is like this is the meta currently and but there's a number of bottlenecks here. So um there's a famous thing called the scaling law um which uh it's a bit like Mo's law for AI for intelligence which is basically like take the transformer architecture add data add compute and intelligence goes goes up except it does and then it kind of doesn't. So if you've been noticing using CHTP4 and then CHTP 4.5 and you're kind of like yeah it's kind of better but like I was I was kind of expecting more by now and the things are incredible right but it's like it's got no memory. It doesn't know

it forgets who I am. doesn't know what's going on and they're trying to fix all these different things. And so just like with Moors law, so Moors law also followed a very nonlinear pattern of of growth. You know, everyone understands Moors law. It's like the the law which predicts um the amount of computation you're going to have over time and how much you can fit onto a single chip. Now that was originally done just by improvements in lithography. In other words, like just sort of like obvious obvious approaches, but it required then innovations which were sort of outside of the the kind of the the standard research track in order to continue that sort of like self-fulfilling property of of doubling compute power. Um same kind of things happening in AI. So we're now starting to see now like um different

agent architectures that are coordinating uh different kinds of um I'm actually going off topic here but there's many many kind of like nonlinear innovations that are happening in that space that are going to drive through this this increase in the scaling law. Um so a number of the things that we've been building over the last I say we in the crypto world in the last sort of 10 12 years um are like very important now for the next wave of innovation in AI we believe and one of those waves innovation we I believe is is decentralization. There are a number of other areas that we're focusing on but the decentralization space we think is very interesting. So, I'm not going to go through all of these examples because I'm on Swiss time. So, I got to make sure I I stick to the stick to the

script here. But, um, we'll dig into these in terms of performance in the wild. Uh, so one of these is a Nazare portfolio company who's ion. I wish we were in Stripe, but we got in would have been too late. Um, let's just talk about Stripe for a second. So last year and I think it was October, Stripe released an agent API, an agent SDK which allows agents to communicate across the Stripe network and um you know basically do payments and all these things. Sounds familiar? Well, this is the script that everyone was running a year ago in crypto. They're like you can't do agents without crypto. Well, it turns out you can um as long as everyone's on Stripe.

And then you're like, well, you know, not everyone's going to be on Stripe. What about the people who aren't on Stripe in different countries? And what about the edge cases? Like, okay, great. Well, guess what? The next thing they released is a stable coin API. So right now if you ask me what the convergence in crypto and AI is, I would say Stripe because they are and these are dominant, right? These are dominant in terms of how people pay for things in the web 2, which I hate that word, but the real world, right? The the world outside of the very very small number of developers that we have in crypto. So if our stuff's going to like hit mainstream, it's going to happen through stable coins. It's not necessarily going to happen through like incentivized architectures and deepens and so on.

It's gonna happen in boring old stable coins. And right now, if you're in crypto, people look at you and go, "You're in crypto?" Like, I I like I like stable coins. I don't like crypto. This reminds me of like 2013 when people were like, "I I like the technology behind Bitcoin, but I don't like the the Bitcoin." And this went on for years and years and years. And now the current meta is like a stable coin is not crypto. Um, more like funky stuff. News Research came out of the Museum of Crypto Art. It was a Discord channel. And these guys are now they just did a um a round at um I think a billion dollar valuation with paradigm leading um they're going to launch a decentralized training and inference network and reinforcement learning network on Salana and these guys have built some of the best open source uh

training models Hermes and and other approaches they built out um that are like very respected in the conventional AI world. Um, ION is building a uh decentralized compute cloud combined with Northern Data which is a large uh GPU cloud in uh based out of Germany. Um, Tether is like partnering with them to build out these things and stable coins allow them to do both a hybrid of centralized and decentralized stuff. Uh, we've got another couple of examples here. Prime Intellect, we're an investor in this from the earliest round. Um, these guys are building out decentralized training algorithms. So this is an example where in order to do decentralized AI you actually need to change the algorithms by which AI happens. So you have to have decentralized training algorithms in

algorithms you need in algorithms you need communication mechanisms such that you can compress and reduce the amount of data that's trans transferred between nodes in the network. Um vast AI is another one of investments. These like have built a decentralized GPU platform. is actually not crypto incentivized, but two of the the two co-founders are both um from uh my former crypto company, Orchid. And then Helas is building out essentially like a decentralized network that allows you to prove whether or not a uh a particular model was executed. So one of the problems in decentralized systems is you actually don't know whether things are done the way you thought they were. Like blockchain doesn't fix everything. Um these guys have figured out that mechanism too. um we're kind of in this moment where

everyone's like okay well now what this is like in in the enterprises we're actually in a trough for disillusionment it's a very like weird mixed bag world and this is in AI in in in general um but at the same time consumer feedback and consumer expectations are skyrocketing so people just want to use it more and more people are using I think like 10% of the world is now using open was the the funny statistic they had recently um and then a lot of the scale laws are broken so we're kind of this weird sort of mixture. So how this is getting fixed and this is again talking about not crypto but other ways of investing in the space is changing the way that AI works. So looking beyond the algorithmic space, looking beyond this tower of babble that's been built of like a particular stack of of AI and

keeping everything on the transformer architecture, add data GPUs, people looking at different ways to innovate. So there's featherless um and also mixt of course from uh from Paris and like a number of other companies, Deepseek that have focused on different ways to change the algorithm in order to make the stuff scale better. Uh we call this matcher. Um it's not political. Uh so make AI cheap again. Um so some people asked me whether it was ever cheap and I was like yeah it was fact back when I was using it was free. I was at Cambridge like you know free free compute. Um and that's what we do. So we invest in uh essentially scalable AI of which decentralized AI is part of it and then some of it is crypto incentivized and I'm early so I'm on time. Good. Okay. Thanks very

much. Oh thank you. One one more thing, not to be Steve Jo, but one more thing. I do have like uh some decks on the thesis etc.