Unchained - When AI and Blockchain Meet, How Can Each Technology Benefit? - Ep 516

Episode Date: July 11, 2023

In recent months, ChatGPT, Bard and other new artificial intelligence (AI) products have revolutionized numerous industries. But how this nascent technology can work with crypto and blockchain technology is still relatively unexplored. In this episode, Illia Polosukhin and Jason Warner delve into the transformative potential of AI and crypto coming together.  They explore a wide array of applications, from AIs managing Decentralized Autonomous Organizations (DAOs), to DAOs being used to help open source AIs, to blockchain technology making the inner workings of AIs more transparent.  Tune in to this episode for a deep dive into the fusion of AI and crypto. Show highlights: Illia's and Jason’s backgrounds, including how Illia’s work in artificial intelligence (AI) got him into crypto what AI actually encompasses and how to define it why Jason says that "it works right now, it's machine learning and if you're raising money, it's AI" how blockchains can facilitate the provision of resources for training data  the problem of attribution in training AI models how OpenAI and Midjourney hold a lot of power at the moment, according to Illia why open source code helps build better AI systems how LLMs are being used to audit code in the blockchain ecosystem up to what point AI can be used for crypto trading why misinformation is a human problem, not an “AI problem,” and how blockchain technology can help solve this issue how blockchains can be used to verify facts to combat misinformation whether AI can help mitigate attacks on DeFi protocols how DAOS and AI can work together and whether AI can coordinate a company what’s missing in terms of infrastructure for DAOs to thrive with AI how to structure regulation with this rapidly evolving technology Thank you to our sponsors! Crypto.com Arbitrum Foundation TOKEN2049 OKX Guests: Illia Polosukhin, cofounder of NEAR. Previous appearance on The Chopping Block: The Chopping Block: Why AI Will Change the Course of History in Crypto Jason Warner, founder at Poolside.  Links Op-ed by Illia: Blockchain can save the media - Blockworks CoinDesk:  AI Can Generate a Trading Edge in Crypto Markets 10 Ways Crypto and AI Can Make Each Other Better (or Maybe Worse) Blockchain May Offer Answers for AI Challenges Former FTX.US President Brett Harrison's New Venture Is Seeking to Harness AI for Crypto Trading Blockchain Developer Platform Alchemy Releases AI-Powered Tools for Web3 Builders When AI and Blockchain Merge, Expect the Mundane at First Imagining the Future AI and Web3 Can Build World Economic Forum: Blockchain can help combat the threat of deepfakes. Here's how Cointelegraph: DAOs: Where humans may fail, AI could succeed Forbes: Convergence Of Web3, AI And Metaverse: Navigating The Great Reset For Investors,  Learn more about your ad choices. Visit megaphone.fm/adchoices

Hi everyone, welcome to Unchained, your No High Presource for All Things Crypto. I'm your host, Laura Shin, author of The Cryptopians. I started covering crypto eight years ago, and as the senior editor at Forbes, was the first Maitre meter reporter to cover cryptocurrency full-time. This is the July 11th, 2020, episode of Unchained. Asia's buzzing, and everyone's going to token 2049 Singapore on September 13th to 14th. Paulusie Srinivossin, Mike Novagraths, Arthur Hayes, and 200. others will hit the stage, joining over 10,000 attendees.

Visit token249.com for 65% off with the code Unchained. Link in the description. You know it, we know it, they know it. The system doesn't just need an update and needs a complete rewrite. Web3 offers that rewrite. It allows us to take control back and to truly own what's ours. Visit okayX.com slash rewrite the system to learn how. Arbitrines leading layer two scaling solutions can provide you with

lightning-fast transactions at a fraction of the cost, all while ensuring security rooted on Ethereum. Arbitrim's newest edition, Orbit, enables you to build your own tailor-made layer three. Visit Arbitrim.io today. Buy, trade, and spend crypto on the crypto.com app. New users can enjoy zero credit card fees on crypto purchases in the first seven days. Download the crypto.com app and get $25 with the code Laura. Link in the description. Today's topic is the intersection of crypto and AI. Here to discuss are Elia Paula Suhyn, co-founder of NIR, and Jason Warner, founder at Poulside.

Welcome, Jason and Elia. Thanks, Laura. Great to be here. Thank you. Let's start with you, Ilya. You have an extensive history in AI from before you got into crypto. And in a way, it's actually how you got into crypto. So why don't you tell us your background in AI? For sure, yeah. So before kind of doing a startup journey, I was for about 10 years working in machine learning and I actually started my career there. And my now claim to fame is I worked at Google Research where I led a team work on natural language. We've done kind of a lot of work across the spectrum, which has launched actually on Google.com and a few other places, Google Play. And as part of it, I

worked on TensorFlow, which is machine learning framework that Google built back in the day that kind of started a lot of the open source innovation in machine running, as well as I worked on attention as all you need, the transformer model that now power solves the chat GPT, image journey, stability, and other kind of innovation language models and image generation. I then left to actually do GitHub co-pilot style startup in 2017, trying to use transformers kind of in 17 when we just build them as a startup called Near AI. And because we did not have Microsoft money, we ended up trying to be smarter and using a lot of kind of students around the world to do crowdsourcing to build a better data set for language decode, kind of descriptions to kind of functions. And we faced this

challenge where we needed to pay people, right? This was students and students don't have bank accounts frequently. They were from countries like China, Russia, Ukraine, which have some monetary control, again, 2017. And like, it's been a challenging to just like pay them for the work they were doing. And so we started looking at blockchain as a way to facilitate global payments, which what blockchain has been touted as a solution for. And so kind of in 2018, as we were trying to explore this, we realized there's no solution that's really trying to address kind of the simplicity of usage as well as kind of would power it at the scale even that we need it. And so that's kind of what gave birth to near protocol because we realized we can solve that problem and then hopefully build. a bunch of things on top. And so that's, you know, that started a journey since 2018 on protocol on blockchain side, but obviously been keeping track on the I side and now

involved as advisor and some other efforts on the I side. Yeah, it's fascinating actually because if I feel like it's actually a very similar story to Brian Armstrong's start in crypto, which is that he had been doing payments at Airbnb and he realized, oh, wow, you know, making payments across 200 different countries, the world, especially because, you know, sometimes, like, somebody who's visiting, like, Bolivia, but they're, like, from Iceland or whatever. He realized it's so challenging. And then that, you know, he learned about Bitcoin. And, you know, now obviously he's the founder of Coinbase. So, Jason, what about you? What's your background in AI?

So I'm mostly growing up through what I'll call building platforms. So before what I'm doing now, which is co-founder at Poolside, I was a venture capitalist. We'll ignore that. little phase of my life for a little while here. And before that, I was a CTO at GitHub, before that Heroku, before that canonical Be macabundal Linux. So it's basically been devs and platforms. But while at GitHub, inside the office of the CTO, while we were doing a bunch of incubation projects,

I had a small team that incubated co-pilot. As Ilya mentioned, you know, his first foray into this. They did have Microsoft money. We had Microsoft money at that point. And Open AI Microsoft. Wait, you guys just, can you define copilot? Because I don't know what that means. Sure.

GitHub co-pilot, which is the AI coding assistant that's pretty popular. And also the very first one that came out with the name, co-pilot now that there's a co-pilot for everything out there, including doing your factors or looking at the sports scores. And so, you know, that was in 20, I want to say 2020s when we started that work. And 2021 when we launched on beta. But we, I've been trying to build that when I, first got to GitHub, I wanted to build two things. I wanted to build what will become GitHub

co-pilot. So it's kind of like the ending of a long journey while I gethub. And now I'm doing something which we call Poolside, which is the way I describe it is simply open AI, but exclusively focused on software. And that'll go into a couple different directions here. And then I got connected. When do we meet, I was it 2019 or 2020 as well? I think it was 2020, something like that. Yeah. And, you know, we've been talking about the overlap between the two blockchains and AI as well in recent years as well. So it's been fun with that. But yeah, AI primarily blockchain peripherally. And when you say that Pulside will be focused on software, you mean it's an AI that's focused on coding as opposed to like giving you all the best tips for, you know,

where to go get drinks in New York City on a Friday night or something? Yeah, we're not going to tune it for any of those sorts of things. And definitely, if you know me as a person and what I would actually care about, too, definitely don't care about where I'm getting drinks at New York on a Saturday night. But I definitely will care about how efficient the Python or R or when is running. All right. So this is not an AI podcast. I am not an AI expert.

So I just need to have us set some ground rules for the discussion. How would you guys, or at least for this discussion, how would you define AI? And the reason that I'm asking is just like as kind of a consumer watching this, it feels like all the buzzes around the AIs that are built on these large language models, which is like very languagey. But then in my head, like blockchain stuff is very numbers based. So then I was like, hmm, yeah, how do those come together? And I obviously have read a little bit about the different things. But to me, it didn't seem obvious at first. Some of the stuff around like Dow's and whatever, that made sense. But, you know, when you use the term AI in this show, like, what do you mean or what are the different things you could mean? Yeah, I mean, I think like just to set the stage, right, the AI, I think people always call the AI something that feels magical. And then when they stopped feeling magical, they like figured out the proper term for it.

So machine learning has been like a proper science that, like, for example, I was doing everything from credit score to when you type things on Google. dot com, when you type things on your phone to when you slide up and you see suggested apps, everything using machine learning. Like literally every single action you do right now is using machine learning in some shape of form. I'm pretty sure like right now with like this recording, there's multiple sound processors that are running, there's video processes that are running. There's like package optimizers that are running which are using machine learning.

And so, so AI really has been this thing that's like kind of the next. thing that machine learning or whatever cannot do or we wanted to do. And so until we like get there and then it becomes a proper term. And that's why this large language models now have a term, large language models, is now becoming its own, you know, kind of a field, subfield. And so then people, you know, now dreaming about what's the next thing about AI will be. But I think AI is just this term that like whatever, whatever is like magical that computers can do, that is getting them closer to how either humans or like superhuman kind of ability we call that AI.

Really that. And then everything else, like everything we can really explain becomes its own term. There's the information theory. There's all kinds of things actually that became, that came out of AI over past almost like, I think, 60, 70 years of AI from existing. That's the joke I use is that if it works right now, it's machine learning, and if you're raising money, it's AI. And so everyone in the world already experiences this, as Ilya pointed out in various ways.

But the large language models are something that kind of emerged, you know, in this is called in recent years. And that is what we, that next wave of AI is kind of riding on. And, you know, we've gone through versions of cycles with AI already in like the 2010-ish era. 2015, 16, 17-ish era, and then this one. And so each one had distinct phases, just like even in crypto, they went through distinct phases. And this one is largely going to be defined by the large language model breakthroughs that

particularly OpenAI had pushed with GPD3 and 4, and we saw a scale. All right. Yeah, I was going to ask you something, but I think you answered it because I have a friend who actually was a writer for Siri. And so I was going to ask you if you think. thought of serious and AI. However, and I know Apple's secretive, so I feel like even weird, just talking about the fact that this friend of mine worked there. But, and I won't name the gender of the person, they started in 2011. So maybe like when I was going to ask you, do you think of

serious and AI? You're talking about like an earlier iteration where there was a human or many humans, you know, kind of feeding the AI. But yeah, would you consider that an AI? I mean, it's definitely a machine learning system. And to be clear, like, there was in 90s, there was expert systems, which were also, like, language-based expert systems that back then were called AI, which now nobody would, you know, touch with AI. But they were actually powering, like, medical systems for decision-making and things like that. And actually, their failure led to the big AI winter where nobody was finding AI because, like, everybody was too excited about it and kind of it didn't. live up to expectations. So very similar to crypto leaves and been running way longer too. It started in 60s. All right. So if you're just being as broad as defining it as machine learning, then that

means like you can apply so many different types of machine learning to crypto. So of all the different types, which ones are you most excited about when it comes to the intersection of AI and crypto? So I think like worst kind of stopping on the large language models. because they are kind of this recent innovation. And the interesting thing is, like, it's for the first time that computers went from needing a human to interpret what the output to being able to communicate directly, right? It went from somebody kind of needed to be, like, professionally trained

to understand the output to anybody can talk to. And that's why it's really exciting and really powerful is because it has this transition. And so now going to, back to kind of Web 3, broadly speaking, there's a number of areas where this intersects. So I would start with generic marketplaces of resources, right? So I mentioned we started because we wanted to use blockchain to pay people for our, like, data crowdsourcing task.

And so this kind of example, so there's data, there's compute, and there's kind of model, and there's kind of model architectures and ways to train things. those are kind of quote-to-quote resources that they use in machine learning, that blockchain is a really good marketplace kind of facilitator that can open up and really create better liquidity. To just give you an example, right now, if you want to get like a GPU cluster, you need to like commit, you know, $10 million at least to one of the cloud service. And you need to negotiate with, you know, some representative.

it's not a very efficient market, right? And let's say you did get this, but you're not using it. There's no way to facilitate renting it out for someone for like an hour, right? So things like that is where blockchain can already become a really easy facilitator to kind of really provide this resources across data, model, like model architectures and compute. And I'm sorry, you were saying like paying people, but also it felt like you were saying crowdsourcing as well? To do crowdsourcing, you need to pay people, right? So crowdsourcing is a way to kind of facilitate, like, I need this work to be done,

and I need to find somebody who is, like, skilled and willing to do this, and I'll pay them for doing this work. And Jason, what about you? Well, I think, like, going back to some of the central premise of, I'm going to say blockchain here, too. So let's say it's like, you know, it's permissionless and trustless. Like, just go back to those two things there. So theoretically, as everyone has pointed out on the internet,

whenever you're talking about a blockchain, someone to say, well, this could have just been a database at some point, too, and et cetera, et cetera. Yes, like technically, I guess that's true in many situations, but that doesn't adhere to those two principles of permissionless and trustless plus some other central things. Now, I think when we're talking about AI in particular, particularly this wave of AI, there's a lot of things where those two elements could play a really interesting role. So Ili is talking about marketplaces, but there's providence of data and there's ability to get information in and out of these things where, The permissionless and trustless is important because otherwise we're talking about how these things have to be centrally administered. So if you're talking about data as an example, somebody somewhere always has access to these things can manipulate and whatnot. So it depends on your views on what those things could, the potential challenges with those sorts of things. So right now as an example, we don't have any idea of providence of any of the sort of things that are going into any of the open AI language models. We have no idea what type of data went into this. Now, if that's an important aspect of what we want as a society, for how we're going to evolve in the future.

Well, how do we do that in such a way that it's auditable and traceable and we can understand how these things evolve? Again, we can go back to like, okay, well, we can show you the books and all that sort of stuff, but it turns out that we already have a system in place where we can say, if you just put it here, we can all look at it in some sort of way. But that's just one example of the way in which these things could work. So it could be embeddings, you can be weights. It could be all these different things that we talk about along the way.

Then, as Ilya was talking about, you know, resource management, which is a particular issue for right now, because I'm literally going to the most central of all central authorities trying to get GPU access at the moment because it's the most precious commodity in AI is how this compute is going. So I'm going to the invidians and the Googles and the Amazon's. And there's 30 people in the world that have access to the type of compute that I need at the moment to go build these things. And, you know, what does that really look like in a world that's a much more efficient and you could basically have GPU spot marketplaces. where we can just kind of come in and use it if we need to and then rent it out, if you bought access to it or whatnot. It's a very different world, you know, and if you build these things around that. Now, again, people talk about, well, that could just be a database.

It can, but then you're putting a lot of trust into people and a central authority. And, you know, depending on your views on that, what does that mean? And so, like where you were going with, it's kind of like you were saying transparency is something that blockchain tech could help bring to AI. What came to my mind was, you know, you've seen those news articles where the AI reflects the biases and prejudices in our society. Is that what you're talking about? Like, helping to... I'm going to use an artist's example here because I think like when we talk about provenance of data and what the training data is, it can get into a really nuanced area really quickly.

But right now, the simplest way to think about this might be just mid-jurney style image generations and what could have gone into that. And there's a gray area right now. A lot of this stuff will be tested in legal arena here. And I don't know what that means. Mid-Journey-style generations. What is that? A mid-Journey is an image-generation, like generative AI, image-generation model that will, if you tell it, hey, make me a pirate sitting on the moon, holding a banana.

Literally, it will understand that and generate several versions of this image that you can manipulate and say, okay, now make it more of like 60s retro style. and it knows how to do that sort of thing. But it only knows how to do that because of how much data is ingested, other images and things of that nature. So in some ways, when we talk about attribution, like, how does it know that? From that particular image, what inspiration is it drawing from? Because it is drawing from other images.

Some of those might be even copyright, some of those might be whatever. It would be interesting to understand how it knows which one of those things are. And no, no one's doing this at the moment because it's actually paying the ass to go, do. But if that's important in the future, there's an easy way for us to go achieve that as well. I would turn that around on the other side as well. Right now, Open AI and Mid-Journey, all these companies, they also need to decide what are things that are good and bad. They are actually making decisions for things that this model should not do. And so we're getting into a single company, pretty much deciding for 100 million of people what's good and bad. And we know how that

went before with social networks, for example. And so this is kind of extension of this problem. And this is going back to what we've been building in blockchain space, right, with DAWS, governance and really trying to figure out how to create an open, inclusive way of deciding on things. We can actually have this DAO's to be driving kind of training data collection that specifically tries to debize and find the consensus on what is communities. things kind of good and bad or safe or not safe aligned with humanitarian values or not, right? And use UN principle with all for that. And like, do that in an open and transparent way. So everybody sees exactly what are the rules versus right now. We don't even know where like

the edges of this models are. I was going to say this is actually something that I felt for a long time because I've worked in the open source community for, you know, going on 20 years at this point, but professionally probably the last 15. And one of the things. that you do when you open source your code but not your decision-making process is people can look at your code and they can say, okay, X, I understand that, I understand why, et cetera, but they might not understand how you made decisions to come to these things. So in an interesting roundabout way, open-sourcing your code actually can lead to a lot more scrutiny because people will then start questioning the decision-making process or who, who made

the decisions and when and why and all that sort of stuff. And, you know, five years on working on Ubuntu with the Devi and Linux community, it was a constant thing. And so there's a world in which you could then take some of those things, as Ili was pointing out, and put them in a system. And it's open for everyone to go inspect those sorts of things. Like, hey, this is the rationale, or this is how we're going to do this, or this is the rule system that's going to be in place for this. And again, like, you know, no one's actually done this fully to an extent. I do think that there's a world in which somebody somewhere is going to see that opportunity and say, hey, I don't want to have to be in a position where I'm going to go defend each one of

these things. Just find a way to go put that in there and let people inspect and understand what that looks like. Just to draw this out for me, because my question was like, what areas are you most excited about where crypto and AI come together? So what you're saying is if you have a self-funded open source project, like you can fund it with crypto and then that will allow all these decisions to be transparent? Is that? That's not specifically what I'm talking about, though, that would exist. I think that where I would be most excited about it is being able to show more provenance, I think, at the moment. I would like to be able to show where some data came from so you can attribute it back to folks in the future. Oh, and somehow you would use the blockchain to do

that? Yeah, you can. Yeah, you can say, hey, these are, stick all the data there. And you're going to say, this is in the series of data blocks that allowed me to, that we're using this, generation as an example, like saying like or inspired this or helped on this. And, you know, at the end of the day, it's all just data. Oh, no, I got it. Okay. Well, so I wanted to now ask, probably this is a more pullside question, but I'm sure Ilya has ideas because I was reading that AI's help coders do their work. But I didn't know, like, is that just, you know, like JavaScript or something or could they even do like solidity smart contracts or any kind of smart contracts? What are you seeing in terms of how well AIs can do the types of coding needed to run blockchains?

Yeah, all those are, anything is available. Any type of code is available in most of these systems. I haven't specifically checked on solidity, but I'm assuming it is. I've seen two, so I've been to a lot of hackathons recently. Two of the winners of two different hackathons were using OpenEI's API to do solidity auditing. So they were fun of use in solidity code, like using the API, pretty much conditioning it with specific things, as well as feeding it with some other static analyzers, but also like describing like what kind of our abilities this would generate. On our side, like I've seen people actually using it to write rust smart contracts for near as well, at least like somewhat common set of problems. So when it's doing that, when it's auditing and it's catching bugs, that is just based on what's already known, right?

Or if it's machine learning, could it ever, like with the chess thing or the go game, like, could it go beyond what humans have already figured out? Or will it only ever, because it's only like being fed, you know, what people have already figured out. What's the limit? Yes, it knows about those things, but from the past. But the idea here is, and this is where I think everyone's kind of racing to, is that it will go beyond what people have known. now like how we're ever going to be able to achieve that, you know, depends on your views. I think we're going to be able to do that and go beyond. But, you know, I do think that what Ili is describing here at the moment is most of what we're

seeing today is generation, like functional generation and inspection. And in reality, what it does is it knows more about the broad swath of like all the solidity decode that has ever been done and the docs so it can catch these things. And at some point, it is going to be able to test this and say, hey, this is where, you know, This is specific to what we're trying to do at pool side because we're focused on code and software in general. But we'll be able to test the outputs. You know, we'll be able to say this is good code or this is runable or this is insecure. And it's a future version of a thing.

So we've not seen it before, but we know that it's insecure because it breaks down on these rules or the testing environment that we have shows that it's probably vulnerable to these things. I also wanted to ask because some of the like news. articles that I was reading just made me wonder, you know, we've often heard about, for instance, crypto trading bots. And then I was just like, wait, so is AI, is that just a new word for what people have already been doing? Or is this somehow, like, I didn't really know what's the difference between a crypto trading bot and then using AI for crypto trading? Because like, suddenly there's all these articles on using AI for crypto trading. I'm like, is that different from before? Or are they just using a different word now? So I think, I mean, machine not even

used for trading probably like as soon as somebody did machine learning. It's actually what's real applications because you can make money super quick on this. They're like, hey, awesome, and we can make money. Yeah, machine learning always there. I think what people are doing now is being able to ingest more signals from language. So for example, you can read news articles or analyze transcript of this podcast and realize, like, you know, catch some of the project names or whatever and potentially make some decision based on that. All of that has been done, like Wall Street has been doing all of this, like just with less sophisticated models, right? Like, I mean, I've been doing natural language from 2010. This was, you know, early, early years

of natural language. So generally speaking, it's just like more sophisticated models. They understand better the language. You can kind of inquire it. And the other thing is bringing that use case as well up because this is the first time you can talk to computer right in the way. This allows to have two things. One is, you know, blockchain is a lot of code and numbers in the way and trading as well. It's just a lot of, you know, numbers and things. And so what it allows to do is it lowers a barrier for kind of people being able to come up with more sophisticated models without needing to really build all the code or strategies or structures. and actually have this conversation with kind of the bot, for example, to describe the strategy, right?

So imagine we're discussing, you know, it would make sense to buy low when, you know, some bigger whales are buying. Like, this sounds simple, but like implementing that and making that actionable, you know, before would require, like, building out a team of 10 high-frequency traders. Right now, you actually can have this conversation with, for example, like one of the systems and potentially it will build a strategy and start executing for you. And so expanding that, you know, now to DeFi. Defi right now is pretty complicated for a lot of things, right, beyond just simple swaps. Imagine you can describe like, hey, I won't, you know,

take a position in whatever the latest high yield pools is, right? And it can, like, iterate over all of them, find whatever's a high yield, take the position and stuff like this. It kind of provides a more natural interface to this somewhat complex still an ecosystem. And I think like that barrier will actually, like AI will actually help, like language models specifically will help to lower, barrier to access some of these things, both not just on trading side, but also on using side and kind of interfacing with blockchain applications.

Yeah, I just, I do want to insert a caveat, which is nothing we say here as financial advice. I don't want people to like whip these up and be like, I'm going to make a killing. And then they like lose a bunch of money or anything. Anyway, Jason. No, no, I don't know. You're going to lose more money than you're going to make in some of these cases. But for what it's worth out, whenever I hear bot, I always think that there's machine learning behind it. I just assume at this point when someone says bot, there's some sort of intelligence and automation that makes it that way.

As an example, there's bots emerge in all these different systems. And really what they're doing is they're picking past datasets or blah, blah, blah, is they trying to understand what they're supposed to do in the moment? So trading bots are not anything dissimilar. But as Ilya mentioned, large language models are going to change those again. But also in the crypto space, what do they have is that they have access to, to open access to the data sets as well as the other systems where good example here is if you want to use non-crypto systems and you're trading go to go to trading view as an example I'm just

purely as an example here by the way obviously but you could use gpt4 or chat gpt to say hey I know nothing about pine script I think is what they use but I do want to create some sort of trading algorithms help me write these one it can write them and you can put them into trading view as an example but then what you can then do is extend that out and start to do this. Now, the difference between that and crypto, which is like you're using this in the stock market stuff, is you don't have the same access to all the underlying data sets. So if someone could do the entirety of this in the crypto space, and then you could also find a niche much easier too because you could be off on some far-phone platform that doesn't have a lot of XYZs and you could really make you killing if you

understand how to wedge yourself into a certain system. All right. So in a moment we're going to talk about the intersection between misinformation and AIs and crypto, but first a quick word from the sponsors who can make this show possible. Join over 80 million people using crypto.com, one of the easiest places to buy, trade, and spend over 250 cryptocurrencies. With the crypto.com visa card, you can spend your crypto anywhere and get rewarded at every step. Up to 5% cashback instantly, plus 100% rebates for your Netflix and Spotify subscriptions, and zero annual fees. New users enjoy zero credit card fees on crypto purchases in their first seven days. Download the crypto.com app and get $25 with the code Laura.

Link in the description. Arbitrum stands at the forefront of innovation as the premier suite of layer two scaling solutions, bringing you lightning-fast transactions at a fraction of the cost, all with security rooted on Ethereum. From defy to gaming, Arbitrum 1 plus Nova is home to over 5,000. And with the recent launch of orbit, Arbitrum welcomes you to build your very own tailor-made layer 3, or, as the Arbitrum ecosystem calls it, an orbit chain, directly on the Arbitrum tech stack.

Designed with you in mind, Arbitrum empowers you to explore and build without compromise. Propel your project and community forward by visiting Arbitrum.io today. We've sat at once and we'll say it again. The system doesn't just need an update. It needs a complete rewrite. Web3 offers that complete rewrite because it extends beyond just money. Staking, NFTs, Defi, Earn, Web3 has become a world of its own. But who has time to juggle between five different crypto apps?

OKX Wallet is one of the best apps for everything Web3. It allows you to store, trade, earn, and manage your crypto and your NFTs across 60-plus blockchains, all in one place. And now it's one of the only self-custody wallets that doesn't require a seed phrase. There's just no other wallet that says powerful yet so simple. Give it a try at okayX.com slash web three. Join over 10,000 attendees for this year's biggest crypto event at Token 2049 Singapore on September 13th to 14th. Sandeep from Polygon, Eric Wall, Chris Berniske, and over 200 others will hit the stage,

joining the industry's most influential for an unforgettable experience ahead of the Formula One Grand Prix race weekend. Singapore will transform into a crypto hub for a week from September 11th to 17th, with over 300 side events that will make for unparalleled networking opportunities. Builders and investors at the bleeding edge of innovation will drive an agenda that covers the ever-evolving regulatory landscape, the convergence of crypto and AI, Web3 gaming, NFTs in the Metaverse, Defi, Scalability, Interoperability, and many more. Visit Token249.com for 65% off regular tickets, with the code Unchained. Link in the description.

Back to my conversation with Jason and Elia. So I have seen a lot of thoughts about how blockchain technology can be used to address misinformation from AI's. But I'm curious, yeah, for your thoughts on like how exactly that would work. Yeah. So I would start with that misinformation is not even the AI problem, right? This thing has been existing, you know, even the basis of crypto, right? the Byzantine fault tolerance is based on the misinformation.

So generally like the Byzantine general's problem, right? And so like generally speaking, like this is a human problem, not the AI problem, that AI is just becoming a tool to really kind of efficiently create a lot of as a misinformation or like very pointed content or create generated content that looks very real, but has, for example, wrong facts. Yeah, and in case anybody doesn't know the Byzantine generals problem, it's basically when there's different people, they don't necessarily all trust each other.

And it's something about, like, you want them to act together in concert, even though they don't trust each other, is that? Yeah, like, there's potentially, like, up to one third of them that are malicious and trying to, like, propagate some information that's not true. And you still want, you know, your army to kind of execute correct orders. The consensus of blockchain is based on that where like up to one third of validators may be malicious, trying to like maliciously attack the system. And you still want the whole blockchain to come up to a consensus like on, you know,

order of transactions and money spent on kind of decisions made on chain. I'm just saying like the source of the misinformation is, you know, Spanish like thousands of years. It's like as soon as people started talking, they started spreading misinformation. And so I think like the, you know, the even with generated like images now the reality is a Photoshop, you know, Photoshopped image existed before and was like cases when it was presented in court like altered images. So I just like want to set up the case because a lot of people say like, oh, AI is like, you know, creating misinformation. Like that's not true. Like humans creating misinformation. They're using this language model and image in like generative AI as a tool to create misinformation.

And so the question is, like, how to address this challenge with our, I call it society operating system. And our society operating system runs on language. We communicate with language. We have like all the government kind of systems run on language. And so this is where Web 3 comes in. Web 3 has cryptography, has reputation, has all of those things that kind of tied together. And they provide the tooling for us to actually authenticate the content, the link it to its creator or authorizers or kind of create reputation around the content source,

right? So for example, let's say this podcast was completely AI generated. How do you know that? How would the listener know that it was AI generated or real people? We three actually said, yes, we actually participated. And so the way to do that is for us, all three, after it's recorded, to sign the kind of hash of the whole podcast and say, like, yes, we participate in this and this is indeed content that we've been members of. And potentially it can be actually like edited with AI to like make it sound nicer, to have better images, et cetera. That doesn't matter. What matters is that I say, yes, I authorize this content and I put my reputation in this. And so to do that, we need cryptographic signatures. We need kind of identity that ties all of the things that I signed with.

So you can see like what I've signed. And we need a set of tooling that when you do listen it on, for example, podcasts or Spotify or whatever, it says, yes, it's verified and indeed signed by the right people. So that is kind of a system that needs to exist. And like not just, you know, for podcast, it is a good example, but, you know, this needs to exist on the government level because there will be like artificial generated filings, you know, companies that don't exist that are run by AI. You can have politicians that are completely artificially created that are talking to people through instant messenger, right, and describing their policy exactly adjusted to. to that person and they would not even know that. And that's like all possible now just like to scale

the level and created a lot more personalized. And so using cryptography and reputation kind of all the pieces that blockchain is, you know, powering would allow us to really kind of create a more trusted environment and reputation based environment that we don't have right now. This is so fascinating to me. And obviously because I'm a journalist and I really care about facts and just really hate misinformation and think it's corroding democracy and all things like that. I couldn't help but think, okay, so let's say that this system comes into place and then really reputable publications like the New York Times, the Wall Street Journal, the Washington Post, the LA Times, whatever, they start entering things in the blockchain. They're using their

reputation. I mean, I just thought of these really famous incidents where reporters were making stuff up. Like, there's a famous case of a guy named Jason Blair. There's another one, Stephen Glass. And there was also the case of the New York Times reporter who got the weapons of mass destruction in Iraq wrong, Judith Miller. So then what do you do when like a reputable source is stamping misinformation or I guess misinformation, I feel like is more intentional? And then, well, so in some of those cases, yes, the reporter was doing intentionally, but in some of them it was just a mistake. So, So then how does that work? What do you do in that kind of situation? So important part, now that it's on chain, right, you can inspect, right? So you can go back and see for this reporter, for example, have they said something already that was marked as incorrect or you cannot get out of this and say like, oh, I didn't say it or whatever, go and edit it post facto.

So you have now reputation attached to each kind of source. The reality is everything's subjective, right? And so although, like, yes, we want facts, but like at the end, it's subjective. And so reputation is subjective to some extent. But you can study, like now there's objective facts that you can inspect that cannot be like retroactively changed. Then the other thing is, let's say, you know, as a reporter, you probably have sources. And so now if you have this cryptographic and reputation system, you can actually have a zero knowledge proof of your sources without revealing who they are. Right. Let's say you have a white house course. You can prove that, hey, you actually had somebody in

White House that told you this without revealing who they are, but like revealing some fact about them and things like that. Right. Like now you can have kind of a lot more information being plugged in into this reputation. And then at the end, it still would be a subjective. Like, you know, some people watching Fox, it doesn't matter for them if it's not true all the time. I mean, I don't know. I'm just as an example. And like they've kind of continue. you're watching Fox, right? That's what I'm saying. In some ways, reputation is subjective, but at least you can have the facts and you can study them and you can observe them and you can have systems and browsers and apps that actually show you this kind of summary about

reputation is like, hey, there's journalists, you know, like out of the last three articles, two of them were flagged with some not correct facts, for example, by the community that you're following. Let's say you follow one of the communities of fact checkers and they're like, hey, we flag, like multiple facts are not correct and stuff like this, right? Like you can have this composability now of different pieces and communities kind of coming together and informing things. Oh, wow. I love so much about this. And I think you kind of addressed a question that I had because I'm sure you're very well aware. And we see this. It's not just news organizations, but for instance, presidential candidates, even when one source like tends to be repeating misinformation more frequently than

another one. People still, they just believe whatever they want to believe and they kind of don't care if one is like proven over and over again to be spouting misinformation. So a couple of things. First of all, my thought was, oh, but what if the people who just want to believe whatever they want to believe are start attacking the reputation of the factual ones? Is there a resolution for that? Or is it simply that because you have the sourcing, you can see like, okay, actually all of this does check out, where's the other, like, it doesn't have as many things that check out, or what? I mean, I don't know if we would be able to fully answer the question, but I would say this, think of it as a hierarchy. At the bottom, like, what blockchains would be incredibly good

for, useful for is, like, facts, like straight up, like identity as a good example. Is Ilya a human or not? Like, can he prove himself to be a human, et cetera, et cetera? Then going all the way, all the way to the top would be things that require some amount of interpretation as an example. Does that make sense? Like, hey, you've got to be able to juggle a couple of different things together and whatnot. You're not going to be able to do as much. And with that, you have to actually build some other systems on top of the base level systems to be able to do some of those things. But the point being that you could build those systems on top of each other. And again, going back to like even as I already said, some of the start with the Byzantine general's problem

and thinking about this, going back all the way to when everyone would start first start, talk about blockchains with Bitcoin and others, what it was was trying to take trust from a central authority and swap it out to a publicly transparent something or other. And if you think about just that one motion, going back to identity here is I'm a U.S. citizen. So how do you verify that I'm a U.S. citizen? You have to go ask central governmental authority. But what if that was instead over here in a publicly verifiable transparent system? That's an interesting, amazing ability to go do. And you could do that for other things as well. So verify that Jason was a CTO at GitHub. You have to go look at my LinkedIn. Then you have to go call GitHub. But what if that was not the way in which we had to do this?

Now, just think this through for a second. We think, why is Sam Altman looking at WorldCoyne and getting this off the ground for what does that mean? What is that supposed to be about all that sort of stuff? And I guess, you know, playing these things through, this is why I think that there's interesting overlap between these two types of technological movements, let's call them. But each of them has become shrouded in their own weird, I don't know what to call them, but idiocy might be the best way, which is AI is going to be used to generate all this bad spam on the internet and this is terrible content, all that sort of stuff. But let's just be honest. Like a lot of the stuff that had to go on the crypto for the last, or Web 3 for the last three years is just like straight up idiocy scams

and whatever this of the world. And it was not what the obviously, the influence. the base of the system was supposed to be intended for, and we're throwing out baby with the bathwater in a lot of that situation there. Do you see there being some kind of remedy once misinformation does get? I feel like so much will need to be built out for this vision to happen. Like, I almost feel like what we're talking about is minimum 20 years from now, if not like 30. So few things. One is, so we've been building, we call a blockchain operating system, which is a fully

cryptographically, kind of reputation-based, everything from, you know, blockchain itself and using all the blockchains in the Webtoe ecosystem to front-ends to kind of authenticated the content you see. Not just us, Lens and others who are building decentralized social. Like, we are all building the substrate for all of this. Like, it's not that far away. I think it's more about, like, how do we, how do we bootstrap the adoption? And I think this is where we just need to show more examples where if it's not used, it's going to be not great. Right. I mean, I think getting it used on a society-wide scale, that's what I'm talking about, 20 years. And that's what I'm saying. I think actually it will happen faster because of like

people will start using AI for misinformation for creating spam and kind of generated content that is like malicious and intent. Like this will become a motivator to actually start using. Like we already see right, Twitter is doing rate limiting because the kind of scrapers are running wild. Right. Like this is this is already beginning of this thing, right? Like, same. If you search for MNA language model on Twitter, you'll see all of the bots for your minds that are posting that, like, hit this, you know,

the rules that, like, not allowed. But, like, this is just things that we see that are, like, pretty much errors of that bot. But, like, all of the other rest of the Twitter is being generated already. That's a reality. The content explosion is just starting, but, like, as more GPUs become available,

as more of this thing, open source language models become powerful, this is just going to escalate, right? Again, like the podcast's fully generated. This is like next year. This is one where I think about this quite a bit, which is, Billy and I have this conversation in private a bunch,

but part of this conversation, which is how weird the future is going to get, how fast things will happen, how weird it will be, how many things will be generated, even if it's 10% real, 90% generated, or 50% real, 50% generate, there's going to be a lot of things

that just get generated, positive and negative. And then going back to what we just experienced in the last wave of the crypto, now that we're particularly in the winter here, I feel like the focus now, particularly like this is one of the reasons why I like what Ilya is building with Pagoda and NIR, which is block chain operating system, because it's a focus on the fundamental problems and solutions that would, back to the original premise of the technologies,

as opposed to going into the very speculative nature, effectively financial manipulations. and stuff with all these different systems, that felt like a really weird side path where a whole bunch of people can make a whole bunch of money, but it wasn't focused on what this is going to do for society. And this is, again, what Ilya is focusing on is how this could be useful for society

and how we could build applications on top of it. And that focus now with what is going to happen should emerge out with some really entrepreneurial people saying, like, well, this is a real problem, this is a real solution, let's figure out how they overlap together. It'll get interesting. And my hope, my hope, my hope is that coming out of this crypto winner,

people are using these technologies for these reasons as opposed to just, again, speculating with VC money so that everyone can walk away and buy a boat and we're building something real. And name it much wow. I had to ask also, you know, we talked about how you can use these right now to audit solidity code and things like that. Obviously, we've seen so many hacks

in Defi, I think for 2022, it's like more than $3 billion worth. And I just released an episode where someone had reposed circuit breakers for Defi to help prevent them. But I wondered, could you create an AI that would in real time do something to either help prevent these attacks or, you know, whatever? And it's not even just the act of withdrawing money, but even like the economic attacks where they're manipulating prices and stuff like that. So, I mean, like folks like Gondelet and others, they're using machine learning to do this stuff, right?

I mean, it's not always like language models per se because a lot of it is not based on language. A lot of it is like understanding time series of prices, understanding the movement of money. Like a lot of folks are actually using like what's happening on test net to detect if there's a hack going to happen on mainnet. So all of this is like predictive analytics type things already happening. There's a lot of smart people doing this kind of work. It's just maybe not as like hyped up as whatever new plug-in to chat GPT that can be explained in seven ways, right? If you want to think about every time you hear word bot, just think that it is actually already using some of this technology, just maybe not the specific large language model portion of the technology.

You're not, you're directionally correct. You may not be in specifics, but this is this is why people are. doing these things already. And, you know, there are going to be, again, going back to it's going to get weird, you're going to see people start building these things out quicker, I think, though, too. You're going to start seeing people who are manipulating systems more. And you're going to see both protective measures, but also adversarial measures emerge more as well. So one of the things that I do think about on a regular basis is what happens when all of these systems are available to anybody and any random person could generate a body army. And right now, we know that what's happened on the

internet with a lot of that has been nation state level for manipulations for various reasons. But, you know, it's also gone down to lower levels where it's like well-funded organization has created a bot armies to go do and manipulate and et cetera. And then, you know, spam creation, et cetera. So it's still funding. But imagine when one person could theoretically create these things on the fly and super fast. So this has nothing to do with specifically the overlap between the two, but the point being that like it's going to be possible for the barrier to entry for a singular person. to do these things is going to drop towards zero on an infinite timeline, but sooner rather than later, in my opinion, which means there's going to be negative things from that, not nearly just

positive things, but very negative things that we're going to all have to kind of watch out for and hopefully build protections in place for. Yeah, maybe to bring back to the defy example. So that bot you can use using Open AI API to find. vulnerabilities in solidity code, you can use it now to find a lot of the common vulnerabilities that do exist in existing defy deployed projects. So what this means is the kind of challenge being that the circuit breakers, and like I've been talking about having circuit breakers in defy as well for like over a year, the reality is like this needs to be deployed first before

we deploy defy, right? And those circuit breakers also needs to be ideally economically guaranteed, where somebody's underwriting and actually monitoring this, like, for the example, some of the audit firms need to actually take kind of sounds at risk and become an insurers of these protocols. And like that all needs to have, like that's the adoption part to kind of really start preventing some of the bad things happening. As Jason mentioning that like it's becoming let like you don't actually need to even know solidity potentially to start like generating attacks on smart contracts soon, right? Like imagine that. It will soon be happening. So to your question of like how fast this adoption will happen,

And I mean, we're getting to a level where we have all the tooling. It's now more about like whoever launch your things need to be thinking long term and kind of creating the right structures to protect things versus kind of throwing things out. And this is true about content, right? Like this big, big publications need to start thinking about it as well and kind of protecting, like, you know, signing and protecting the content they're producing in such a way that you can build things around that and build this reputation systems around that. All right. So this actually brings us full circle back to like what we were sort of discussing kind of at the beginning, but we didn't go into depth. And this is like when I was doing my research, I felt like this is where I was kind of finding the most interesting stuff. But I feel like DAOs and AI are sort of the place, at least where I thought like, oh, there's like a lot of potential here.

So, Ellie, I feel like that's where you were going initially when you were talking about the crowdsourcing and crowdfunding. Yeah. But I feel like there's other things too. So I don't know, why don't you guys just talk a little bit about all the different ways that you think Dow's and AIs could work together? So I think a really interesting part is, and I also intellectually, it's interesting to think about, and I think we'll see this very soon, is let's imagine we have an organization that's run by AI, right? So instead of a CEO being a human, you have a CEO being an AI, right? And they have a mission. They have, you know, set of KPI and goals that they set out to do.

and they're actually allocating work to people to do within that company. Legally wise, we cannot do that right now, right? Like in the physical world. But on blockchain, this is actually pretty easy to do. We set up a DAO. The DAO coordinator is an AI model that produces tasks and pays for them to people. And you have, you know, shareholders, token holders, whatever the, you know, interest holders in this DAO who actually are setting up KPIs and making sure that AI is on point.

and kind of doing the things it should do. That structure is like very possible, you know, very soon. Like I'm expecting, you know, somebody. I mean, there's like a few technical things that needs to be built. And I know a few companies that are doing this right now. And we'll see some of this happening where you can have a project that's like kind of coordinated by AI based on the KPIs that the community has set out. And it's kind of, you know, splitting the work and, you know, analyzing the results

and kind of deciding what to do next. And, you know, it will make mistakes, but so as humans. And at the same time, it will be way more kind of cold-hearted in a way with sounds of decisions and not biased with some things. I think it's interesting to think, you know, expanding that, you know, a few years forward and beyond that, we talk a lot about, you know, AI is taking over the world. Like, people love this, like, Terminator story. But the reality, I think, will be economical, not kind of existential.

It will be because AI is actually better according to any humans that humans are, because they're not biased, because they're kind of can observe everything and can process all of this data at the same time. We economically will want to work for AIs pretty much as like better boss, like makes us more economically successful than kind of this existential AI becoming evil and wanting to take over the world. So I think that's an interesting concept for me. Oh, just help me for a second because, you know, as we mentioned earlier, there have been like prominent AI scientists who've come out and said, oh, these models are being. trained in a way that they're reinforcing a lot of societal prejudices. So you're saying that it's an

AI where they're only going to. Yeah, the community is becoming it's pretty much like governance body, right? Like, you know, they can decide if it's doing the wrong thing. It can decide if it's like, should be, you know, fine-tuned for a different thing. They can literally say like, hey, we need to source this types of data to find to, to adjust the AI model to be better at that or not do this or, you know, be less biased about that. They can, you know, commission pretty much crowdsourcing data, using again blockchain to actually get more training data for this model to become better at specific tasks.

And yeah, but also I felt like where you were going to is that since the AI will have a certain set of kind of goals or parameters that are more numbers based, then it's just a simple thing, almost like a check of, you know, did the Dow manage to do this or did it not? or like which Dow members did it versus which ones didn't. That kind of thing is that. It's like it's way less about, you know, any kind of observable parameters and a lot more about like work has been done and kind of, you know, just very transparent, very like kind of egalitarian or meritocratic approach.

Then, you know, as a humans, we strive to be that, but like we always fail, right? That's just inevitable. One other thing that I wanted to ask about is that I was also reading, you could do the reverse, where you could have DAOs that manage AIs. And I think we had kind of touched on some of these issues before, you know, when Jason was talking about like the transparency and different things like that. But do you see anybody working on that already? Or because, you know, we have like robust Dow infrastructure. I'm not as close to anything going on in the Dow arena. So I don't know if anyone is actually using this at the moment. So maybe Ilya in his hackathons has seen more

of these things than I have. I have you seen anything? I mean, people are definitely trying. I think the missing piece right now on infrastructure side is kind of right now, there's like two types of models. There's either like fully centralized models run by, you know, open AI and Google or like fully open source model that like just runs on the edge. And so there's no kind of this way to run some of this models in a, I call it under

consensus, right? So in a way that, like, you can prove on chain that it's correct, right? So some people are trying, you know, there's, for example, a way to convert some of the machine-dney models into ZK circuits, but it only works for kind of smaller models. And ZK circuits is what? Zer knowledge circuits. So, like, you can, you can run a model and then prove that the output is indeed the output of this model. Oh, okay. Yeah. So people are working in infrastructure for that, and this is kind of precondition of

any of the like Dow and machine learning interacting in a actually like trustless way. Outside of this, like what you can expect is more like a Dow commissioning pretty much somebody to do something with a model and then getting results back. Right. So this is more like, hey, you know, similar how Tao has commissioned somebody to go do something in physical world, right, and validate the results later. But I think the interesting parts for sure are more like around either augmenting humans kind of in their day-to-day work, right, and kind of interfaces and creating a better interface

or in this like making AI now become more of a coordinator role or kind of a pilot versus a co-pilot role. Yeah, I feel like what that combination could do, Dow managing an AI would address, like I'm sure you guys have heard, you know, people will say things like, you know, I mean, it's like with the FTX stuff, you know, they were always trying to mitigate against AI's doing something really potentially adverse to humanity. And I feel like that's what a lot of the fears are around the more closed-sourced AIs and even things like WorldCoin, like people just get nervous about certain things that aren't well known. And it's only like a small number of people that know them getting too much control. So I feel like if it's something where it's more public

and like the community feels that it has control of it, then it could mitigate some of those fears. So one other thing that I want to ask about was both of these are super emerging areas of technology and they're definitely not well understood by regulators and lawmakers. And Elia, I know that you have some thoughts about how you see parallels and how they're being regulated. And you mentioned to me an interesting idea about how they could be used to kind of regulate each other. So what are your thoughts on that?

Yeah. So, I mean, I think the similarity here indeed that like there's an emerging technology is that like literally changing every year you see like a complete shift from what it was before. You know, regulations don't move on that speed. And at the same time, like what we're really trying to do is to protect consumer, protect the markets, protect kind of the people. And like that is the core part. And then around that, A, we already have a lot of the regulations around products and around

internet and around just people doing wrong stuff, right? So like that all exists. And it's just kind of, you know, the point about blockchain always been like, hey, on the edges, you already have all of the regulations that are happening, right? Like people doing the wrong things, like they already fall under the existing, you know, everything from money laundering to it. It doesn't matter which tool they use to do that. And same on the eye side, when people doing malicious things like creating misinformation, like there's already loss against that. It's not like a new thing. Like same as like, you know, doctored images to present at application. and so or misusing copyright information or whatever. And so on the other side, like, what do we want to protect? Well, so I mentioned, right, using blockchain for reputation and identity. And same actually even for blockchain space itself. Like, we're not using blockchain to actually have the identity of people who are building blockchain.

Like, we're not using it. And like the tokens launched are not kind of cryptographically signed by people who launch them. There's no like security reports that, you know, produced by auditors are not cryptographically signed and put in blockchain. So all of this body of information organized properly can be the way to really regulate it. And for that, regulators should be just using this technology because it's actually designed for a lot of this use case of keeping track of records, making sure that everything is authenticated, allowing community to inspect it very easily, allowing other external tools.

Like imagine, you know, you're opening a wallet and if you want to transact with an address that is on some lists, right? It like automatically shows it because that list is on blockchain. And that to add to that list, you need a community vote of the interested part of the parties that are involved in these decisions. Same, you know, if you're trying to transact with a contract, if it doesn't have an audit, right, the wallet right away shows you that this contract does not have an audit. And like, are you sure you want to transact it? Going back to some of our earlier points already, the core basis for what blockchains would be good at would be a remarkable boon for society if we could use them that way. So think about even just like fishing attacks and emails that come in or text messages or whatnot. Like we haven't really fully used them.

There's, you know, this is again, going back to like one of the reasons why I like hanging with I was he's focused on making these things possible as opposed to just trading monkey pictures on the internet, which is yes, people are going to go. do and stuff, but ultimately, like, it's solving a real societal level problem. And if we use them that way, we can actually advance real needed things that are going to benefit everybody in the world to a degree. So, but we got to focus on them, build them, and we have to actually understand how they're going to be used. So going back to the regulator point on this, which is on both sides of the fence, no one who is going to make up any of the rules or regulations fully understands these technologies. And they don't understand the implications of the technologies, not the core of the line, but what's positive or negative can happen with these things. And so they rely on experts to come in

and talk about them and all that sort of stuff. But this goes to age-old problems of like incentive structures. And so somebody going in front of Congress and saying, hey, we need to put all these things in place for large language-run models at the moment. Like, well, let's just be honest. That's a very selfish view of the world for that one in particular person. We need people who understand these things. I'm going to go into a slightly different tangent to Ilya here and just say, I think we need showcases for the technology that will help folks understand these things. So right now, and like what Ilya described is there's a lot of practical that exists on the crypto side of the world, but then it's some theoretical for some of the things in place.

And I think we have to understand like that technology needs to get showcased. The capabilities need to get showcased so that people can fully understand that. Because Ilya is, is smart enough. And everyone who's listening to this is likely smart enough to understand and extrapolate out what Ilya just said that it goes, oh, yeah, I can see how that would be useful. But people who are not immersed aren't going to be able to see that. So they're only going to see the moment. You know, if we can move what is possible in the moment, we can showcase that sort of thing. So again, I hope people listening to this start to focus on those problems because those are real things that will advance the state of the art in a society forward.

I mean, for sure, speculation is just a huge part of the cryptic community, at least for now. So I don't think it's the majority of my listeners, but... No, but I think this is actually something that I get really bugged by because obviously, like, it was encouraged for quite some time because you can make so much money. But imagine this too, like half of these VC firms just to show that they were, quote, unquote, in the community hired people with anonymous identities on the internet. And like one of them was famous saying, we don't even know who this person is. Well, that was absurdly reckless, absurdly because the technology that they're supposed to be investing in is one of those things that should show them that they could actually prove who that person was and et cetera.

And they didn't even go through that process. They just hired a random avatar from Twitter. And it's silly, like in the moment, like now retrospect to talk about that, but they should have been showcasing what is possible with this technology too, in my opinion. Yeah, I mean, to mention like another example is, you know, using the technology, we allow to ensure that all the money are. are staying in the right place. They are said they're staying, not being reused in different places by editing some code and moving them around accounts, right? Like, this is, this is like, that's a part.

It's, like, it's been encouraged to build stuff on top that are not really using the technology and kind of allowing to speculate versus actually using technology. And, like, you know, there's actually projects like orderly, which are, you know, all of the custody is on chain and you can inspect everything while all of the trade, like, For example, execution and order book is off-chain for performance and for, so like stuff like this is possible. It's not like it's impossible to do performance things. It's just like it's been encouraged indeed to do a more, I mean, I guess, reckless way of doing things. But again, like I think to point to bring back to the regulations, what's important here is instead of saying like, hey, we're going to regulate with more, I call it language, right?

hey, we're going to come in and say, like, there will be, you know, you need to come to us and register for this and that and the other, right? And this is true about Web3 and AI, right? They're saying for AI right now, if you're using more than this much terraflops for your training of your model, you need to register. And it's like, well, that makes no sense, right? First of all, the capabilities that will emerge from lower than that, like, nobody knows what they're all going to be in a year, right?

This technology is evolving, like somebody will open source, the weights will leak. none of this makes sense, right? Versus saying like, hey, you know, we ensuring that, like, even if there's like super powerful model, it is like cannot actually affect us in any way because we have this cryptographic identity and reputation. And then on the other side, we're registering the right things automatically through a blockchain

and like augmenting it with more and more information from external sources because it's transparent, which data has been used, et cetera. That's a way to go. And again, like both 4Web3 and 4EI, And from the 3 side, you can use AI to explain to everyone else, like, what is actually happening, right? To actually interpret the source code of the smart contract in natural language to explain what transactions have been happening. Right now, when you open a metamask and you see a bunch of bikes that you're signing, that can not be, you know, the real world, right?

Like, we should actually explain exactly what you're signing in transaction. And we can use for that, again, like fine-tuned models to explain it. So, like, all of those things just like coming together really, like, reinforcing each other. But they need to be used in the right way and really focus on that versus focusing on some of that. And I think to even go further on this, too, with what Ilya saying, what we're talking about now, particularly, I'm obviously close to the AI regulations side of the fence, but I'm close enough to the crypto side too to talk about this. It's all this weird kind of performative regulatory stuff. To go into the tarifflops example here, it's like, sure, whatever. You can do that and have someone to audit you, but we don't know what the capability that's going to be in a year, let alone 18 months or two years, et cetera.

And so it's all this performative, classic nonsense from a regulatory perspective when, in fact, all someone has to go do is understanding Google build some of the systems that can showcase what would be possible to go do and have these conversations publicly in a way from a transparency perspective. So again, going to where the data comes from, problems and all that sort of stuff. Well, the system existed for us to register datasets and two and say this is what our training data sets look like and go ahead and audrum or more the weights or whatever, blah, blah. And we understand that. Yeah, I mean, I see what you guys were saying. I just think some of, because at different points, it seemed like you were kind of referencing, you know, like what happened with FTCX or whatever. And I do feel that some of the problems that you cited, I don't know really how you would use

blockchain technology to resolve some of those things because there would just be the sheer number of things you would have to build to make that possible. It just seems. So that's what I think. It's built. Like we have a project that's literally a as performance. and as easy to use as centralized exchange. But it does, it does.

And settlement and assets are all on chain. And it does not only proof of reserves, but also like liabilities going out. Everything is literally on chain. Like it's in smart. It's like the defy decks, but with the speed and execution of centralized exchange. Well, you know, you should do a partnership with Coinbase or something to show how this works. or, you know, Binance might be up for it since they're in regulatory hot water. Yeah.

I think you're assuming finance wants that in this case here. Well, I mean, yeah, assuming they want to keep their business, they probably need to, you know, give an apple to the teacher. Anyway, we've spent plenty of time on regulation and we're running long. Do you want to ask you one less question, though, which is, obviously we've now seen a flurry of activity around the application of AI. blockchain or crypto. So kind of in the next year, what are different things that you're kind of keeping your eye on, like developments that you are interested in following? I mean, from my side, I do think the AI coordinated DAOs are really interesting. And I'm expecting some first results, like, you know, to come out, everything from, you know, simple AI-based

treasure management to more maybe like coordinating people to do some activity in real world. I think that that would be really exciting to see. And they are will be government. by the Dow members, right? So to your point, Dow members will be kind of ensuring that this AI is operating within its kind of mandate and, you know, potentially be biased. So I think like that we'll see probably this in next year kind of first examples of that working.

And I'm pretty excited about that. You know, I'm focused on obviously getting my first model market for Poliside and making sure that it's already beating state of the art, which is what our expectations are based upon some more early model runs. But also, if I'm particularly paying attention to anything, it's going to be how the regulatory conversation is coming down and particularly who is in it and how to help push that to a good spot from the AI side of the fence.

All right. Well, this has been a very fascinating discussion. Thank you both so much. Where can people learn more about you and your work? Twitter, I know Black Dragon or check out near the door for blockchain operating system. And Twitter, Jason C. Warner at Twitter, and then a landing page at the moment. but Poolside.AI will be there soon. Perfect. It's been a pleasure having you both on Unchained.

Thank you. Thanks so much for joining us today to learn more about Jason and Ilya and the intersection of Crypto and AI. Check out the show notes for this episode. Unchained is produced by me, Laura Shin, with help from Kevin Fuchs, Matt Pilcherd, Zach Seward, Juan Aranovich, Sam Shryoram, Ginny Hogan, Leandro Comino, Shoshang, and Margaret Correa. Thanks for listening. Thank you.