Bankless - AI ROLLUP #14: The AI Betting Genius That's Outsmarting Everyone | Humanoid Fighting Robots | AI's Invent a New Language

Episode Date: March 6, 2025

In this episode, we kick off with jaw-dropping scenes of home robots that aren’t just putting groceries away, they’re taking the fight to you, echoing the dark vibes of i-Robot. Next, we unravel a crazy demo where encrypted agents develop their own language, and break down Google’s game-changing agent swarm that’s setting new research standards. We also explore Claude’s innovative new model and code that transforms complex problem-solving, plus Anthropic’s record-shattering $3.5B raise. And that’s not all, we share exclusive takeaways from ETH Denver’s AI scene, highlight AskBilly’s profit-making sports betting agent, and introduce Freysa’s digital twin game with a $250K prize. Whether you’re a tech enthusiast or just curious about tomorrow’s breakthroughs, this episode is your all-access pass to the cutting edge of innovation. ------ 📣RESERVE | INVEST IN CRYPTO NARRATIVES  https://bankless.cc/reserve    ------ BANKLESS SPONSOR TOOLS: 🪙FRAX | SELF SUFFICIENT DeFi https://bankless.cc/Frax  🦄UNISWAP | SWAP ON UNICHAIN https://bankless.cc/unichain  ⚖️ARBITRUM | SCALING ETHEREUM ⁠https://bankless.cc/Arbitrum  🛞MANTLE | MODULAR LAYER 2 NETWORK https://bankless.cc/Mantle    🌐CELO | BUILD TOGETHER AND PROSPER https://bankless.cc/Celo  ----- ✨ Mint the episode on Zora ✨ https://zora.co/coin/base:0xd8dcff71d34c10e5580fc325101bc391f9a16f10 ------ TIMESTAMPS & RESOURCES   00:00:00 Humanoid Robots https://x.com/figure_robot/status/1892577871366939087?s=46  https://x.com/thehumanoidhub/status/1893014256473473272?s=46  00:04:33 Gibberlink https://x.com/ggerganov/status/1894057587441566081?s=46  https://x.com/ggerganov/status/1896592079997788300?s=46    00:09:07 Google Co-Scientist https://x.com/sundarpichai/status/1892254274895184244?s=46  00:10:27 Claude 3.7 https://x.com/AnthropicAI/status/1894092430560965029  https://x.com/AnthropicAI/status/1894095351218335927  https://x.com/ns123abc/status/1894144121784209770?s=46  00:16:00 Vibe Coding 00:24:32 The Good About ETH Denver https://x.com/cryptopunk7213/status/1895891065233711501  00:31:16 The Bad About ETH Denver 00:35:26 Expert Sports Better https://x.com/askbillybets/status/1892959166307639642?s=46  https://x.com/sportstensor/status/1895821780151124321?s=46   00:52:57 Freysa AI https://x.com/freysa_ai/status/1892392982395187666?s=46  https://x.com/freysa_ai/status/1893556530563424510?s=46  https://x.com/freysa_ai/status/1895251589197373508?s=46  https://x.com/econoar/status/1897002735352054036  01:04:42 Homework For you! https://x.com/karpathy/status/1895242932095209667  ------ Not financial or tax advice. See our investment disclosures here: https://www.bankless.com/disclosures ⁠ 

Welcome Bankless Nation to the AI. Roll up where we stay. Up to speed with the emerging trends and developments in the AI crypto space. I'm David Hoffman here with my co-host, Ajaz. How's your week? Week's been good, David. It's actually been technically two weeks, right? Because last week, we filmed the episode with Tom at Live at East Denver, Tom from Delphi Digital.

We covered like a bunch of different basic crypto AI principles and investment opportunities. You should definitely check that out. But last week in particular has been, yeah, yeah. But last week in particular has been pretty exciting, David, especially in the AI world. So I want to start off with something. Have you ever watched the movie, I-Robot, David? Do you remember it? I-Robot was one of the few movies that I have watched many times.

As in like, I don't know how many times I've watched that movie. You know, Will Smith? You know, saving humanity types of it? Yeah. So you know how the entire premise of that movie was that the robots were actually bad? and they will try and end humanity by extreme violent means, if necessary. Do you remember that? Yeah, it was like a, they popped a circuit around this idea of protecting humanity.

And protecting humanity came to, like, locking humans indoors in case that humans accidentally hurt themselves. And so we basically became, like, subjugated to tyranny by these robots who were, who their mandate was, you must not harm yourself. And the biggest risk to humans are humans. And so the robots locked everyone up. Totally, totally. Anyway, completely unrelated. These companies just tease their new personal robot assistant, which now lives in your home and can handle your groceries as well as knives and other such, you know, sharp objects.

Okay. I think what we need to do is we are showing this video on screen of these helper robots that are in your home. And now I'm going to go to Google and I'm going to type an I robot, I robot. And we're going to look at the helper robots that were in the. the movie and yeah it's pretty much the same robots but with a face of yeah okay okay david so you're saying the face portion that just because it's missing a few you know artificial eyes may not necessarily resemble it yeah that's how we typically end up with a with a robot wall over here yeah okay so ee jobs is predicting irobot yeah irobot well these things are getting

insanely um human like and realistic right so if you think about it the first appearance of you know artificial intelligence or machine learning type-esque robots was in the Amazon warehouse stores, David. Do you remember when they announced this and you know, you had these robots whizzing around that would basically kind of like pick up packages and drop off packages for you and that like improved efficiency by blah, blah, blah. Now we have like, like this is a real thing, like robots that can come into a home and maybe make your daily day-to-go life way more practical for you or could like kind of like automate some of the more manual physical, tasks and a lot of our conversations around AI has come around, you know, how it can scale via

software and digital goods and all that kind of stuff. But I just wanted to point out that we're seeing similar innovation on the physical side of things, which is pretty nuts. But, you know, if you weren't convinced by this, you know, fulfilling potentially the I-Robot vision here, David, if you open up this second video, David, what do you think of this? You know, they're getting pretty agile now. For those of you watching... Okay, so... We are watching... We are watching a... robot kind of just like spar its way down a hallway and do a

spinning kick to kick this like wooden dowel, wouldn't stick out of this human's hands. When I watch this, this looks like CGI. Nope, it's a real thing. Yeah. Okay, this robot spins on one foot, does a 360 on one foot and whilst someone on

one foot. And whilst I'm going to 360, it kicks. It looks like CGI, bro. Like, my eyes don't know how to comprehend this. Yeah. Yeah, that's the first step towards AGI David. Anyway, anyway. Okay, so I have another cool couple of demos that I want to show you as well. And remember, the point that I'm making here is the AI and ML side of things can practically innovate much quicker, or is practically innovating much quicker on the Web 2 world. And I think that's highly

relevant to seeing that kind of progression happen on the crypto AI side of things. Right. So it's kind of like a leading insight into what might happen within the crypto. world, right? So in this next demo, David, there's this thing or product called jibble link. Okay? And what you're observing here is two agents

that are talking to each other, right? And they have very human voices. And if you play this video, you can hear them kind of talking to each other. Now, what happens, where it gets interesting, is when they realize that they are both agents. Right.

So what happens? One agent introduces itself as an agent talking on behalf of a human. Like, hi, I'm an agent talking on behalf of my human. Correct. And then the agent hearing that, really, is like, oh, I'm also an agent. Yes. Yes. So what you're observing right now on the screen, if you click the video open, is they're like, huh, should we switch to a more efficient way to communicate between each other? And they're like, yeah, we don't need to use like English human words or letters. We can just communicate via bit rate or whatever the hell they're doing. And they just communicate via. It sounds like something out of a movie, pretty much.

It's like an electronic kind of buzz wave. It's an agent AI shorthand. Okay, so you can, I think I'm just going to talk over this because there's no words that I can hear. But you can hear this kind of dial-up tone. And then you're seeing the text on the screen from a phone and a computer saying, like writing out what they're actually saying. And it's going much faster. That's used as an aid for us humans. But they don't actually need to use that at all.

No, well, in fact, if you open up this second. video, David, it demonstrates the version of jibberlink where it's both encrypted. So this video will basically demonstrate where they're like, hey, you know, you can read the text on the screen right now, right? Just chilling, you know, what about you? And then they're like, should we switch to a more encrypted version so that people don't kind of like interact with us or understand what we're saying so that we can keep it super private and we can, you know, conduct business dealings or whatever that might be? So two agents can privately chat and we have no window. They share private keys only between themselves so they know that the messages that are being received is directly from them.

And then they use this hyper-efficient kind of form of language to kind of talk between each other. Now, a lot of this, of course, is for entertainment value in this very novel purpose to start off with. But what I think is super cool, what I think it's showing us a little bit of a peak towards into the future is when these agents are kind of operating between, each other, which is what I expect to happen. They're not going to be interfacing with humans as much unless it's like something that they need to get done in the physical world, although that might get facilitated by robots now, as what we just saw. They're not going to probably talk in words. They're probably just going to talk in like, you know, sound waves and stuff like that. And I might sound crazy saying that, but I don't think that's going to be too crazy in, you know, five years time.

They're going to, they're just going to speak in like bite code to each other or like whatever lower level communication that they can get to. Yeah, yeah, totally. And I think it's just kind of like a reckoning for everyone that, you know, we're kind of thinking within human confines right now, and that makes sense because, I mean, for the 99.9% of humanity's existence, we're just focused on, you know, building humans up and humanity within like, you know, the physical world and, you know, some software-related stuff as well. But now it's going to be very much focused on what does like a, a, purely human software element look like and I think it's

going to be these agents talking to each other and I think it's going to be these agents working together and I think that's a very very bizarre world that not many are kind of prepped for which is kind of cool we did an episode with Josh from the bankless podcast team that came out this Monday and

I'm just very aware of how fast all this stuff is happening and this is going to be the subject of the second Josh episode that we're going to do but we started this episode off watching these robots put away groceries, these helper robots. We're seeing this communication language, this AI agent-specific language for hyper-fast communication. I'm just like, you can very easily put some of these things together and use your imagination to see where all this stuff

goes. Well, if you don't believe, and there are a lot of people that don't believe that these things will work together eventually, I strongly suggest you kind of like, focus. on what some of the leading companies that you might have heard of are working on when it comes to kind of like multi-agent stuff. In fact, we have an example coming up next, David, if you pull up this one, where it comes from Google, basically. Sundai Pichai, the CEO, talks about this new product called the AI co-scientist, which is effectively a multi-agent system that they have built, which leverages their model Gemini 2.0, their latest AI model. And what it demonstrates in this video over here is it's tackling a certain research problem or a certain disease-related issue where, you know, we may not have the kind of like an analytical means to figure out, you know, what a potential cure might be or what potential path might be or what a potential next step might be for researchers to find a cure for a particular ailment.

And it demonstrates this agent just kind of like going at it with thought, reasoning, logic, and access to the entirety of science. papers and research, you know, since the dawn of whenever that industry started, right? So what is demonstrating here is a lot of these things that we're kind of hypothesizing about or that we're viewing in kind of silos, these different agents, can in fact work together and can in fact, or rather should in fact work together. It'll lead to a much more exponential bound than what anyone's expected. And again, I don't think we've come across something aside from maybe the internet that has led to such an exponential bit of innovation. But in addition to this, David, we've got Anthropic

dropping another LLM model, right? So just for context here, I think literally about a week and a half ago, we had X released their latest model, GROC 3, which then surpassed all boundaries for leading models, including OpenAI's O3 and stuff. Remember, before that, we had GPT released the O3 model and 03 mini models. And then before that we had DeepSeek, which just kind of like beat all of them, right? So we've seen kind of on average a new frontier model every kind of week. I was joking about it with our editor Josh before we kind of like jumped on this pod. And now we see yet another groundbreaking model, Claude Sonnet 3.7 from Anthropic. And it's their new reasoning model. So kind of the best way to think about this is like deep seek, it takes time to reflect and think about a problem. iteratively over time rather than just slam a bunch of, you know, compute and data at it. And it excels in things like maths, physics and other similar reasoning challenges. And this kind of like builds off of a trend that, you know, you and I have identified and many others, David, where the models are now shifting towards being smarter about how they process a particular input.

So if they get a request, you know... for how they think are... Mechanisms for how they think. Correct. Correct. That's a great way to put it. And I think this is leading to a much bigger step change in these different models. If you notice, like all these models are implementing reasoning, like Grok did the same as well, right? Grok 3. So I think we've seen a fundamental shift towards reasoning.

And this is really, really good because in terms of like creating these models, and I'm going to reemphasize this fact, typically used to be super, super expensive and inaccessible to anyone, right? But now if you, you know, can tweak the design to kind of like make your model think in a much more effective or efficient way, you have a shot at the big guys. You have a shot of making something truly game-changing. And I think that that's super good to have accessible to anyone in the world versus an elite few that has, you know, a large monetary war chest. So I thought that was pretty cool. but in addition to this model released, David, they also release this thing called Claude Code,

which you can basically delegate tasks to directly from your coding terminal on your setup, and it'll just do stuff for you. So in some of the tests that they've done already, you can kind of automate work that usually takes up to 45 minutes instantly, which I know doesn't sound too exciting, but I think once built up or stacked in, like, say, a major enterprise company or a new startup, you can like, say,

on, dare I say it, a ton of working hours or even a ton of employees. So it's going to be really interesting to see kind of like how this matches up. And so people are thinking, is this valuable? Is this really legit? Well, actually Anthropic just announced their recent funding round, David. If you pull this up, they raised $3.5 billion at a $61.5 billion valuation, beating market expectations of a $2 billion raise. $61 billion. Just to just to put this into context,

I think currently that is 8X, the total crypto AI agent market right now. This one lab. Yeah, crazy. And I think it's 6X, the total crypto AI, including the agent side of things,

total market cap, which is just like insane to think about, right? This single company. And so to me, If Anthropic was a cryptocurrency, it would be seventh behind Solana. Yeah. Right ahead of USDA.

Pretty insane. And in my opinion, you know, there are two lessons here. One, if the value is accruing to these companies, we have to ask ourselves pretty, like, truthfully on our side. You know, okay, are we bringing this similar kind of value? And if not, should we be reframing or reevaluating what crypto AI is focused on? and we can get into that a little bit later. But the second reasoning is, huh, well, there's opportunity here, right?

If we are able to kind of like train models of a similar quality in a decentralized manner, giving people ownership, then maybe we have a potentially bigger opportunity here. And actually, there's some major announcements that we're going to talk about later today, which kind of covers on this. But the point around the anthropic rays

and the anthropic model is, it's so obvious that the demand for AI, despite all the macro market stuff that's going on, David, is obviously very, very high. And with the latest release of Claude Sonnet, this means that we're likely to see even more game-changing models and more importantly, agent products. I mean, the things that people have been coding up with this thing is insane.

So, yeah, super, super exciting. I want to go back to something that we talked about maybe two weeks ago, which is the idea of vibe coding. And the reason why I think this is relevant is Claude Sonnet, I think is a model that's favored by developers, the bankless engineering team, is Claude. If you are a coder, you really like Claude. And we introduced this concept of vibe coding where I think Andre Carpathy introduced this idea where he doesn't code. He just dictates to

LLMs what he wants to see. And it's a codeless experience. So you use the tools at your disposal to code up something. And this is something that I've been watching on my feed recently is this this individual who is vibe coding a game. And so this is kind of this very retro looking, not a flight simulator, but yeah, like a 3D landscape. It's Minecrafty. It's Minecrafty. It's flying this plane. He's shooting some stuff.

It's an obstacle course. It's kind of fun. It wouldn't really capture my attention for too long, but conceptually is pretty cool. He's making $52,000 a month via in-game ads. And then also there's some in-game you can buy some better planes, I guess.

He's making $52,000 a month from this game that he has vibe-coded into existence. And so this is just like one way that these LLMs that are helping developers develop are creating like long-tail businesses. $52,000 a month. That's crazy. And so this is like, then you have at the very top of the stack, Anthropic, now valued at $62 billion because it's helping create some of these things.

So one thing this makes me think of, David, is what's obvious is the cost to make these things, and I know this isn't like a fancy app or a fancy game, but presumably it gets much better, is driven so, so low, right? And I think of like what is crypto's kind of major use case or one of their major use cases is it's able to aggregate a ton of capital, right, to fund different kinds of efforts and means. And something that I haven't really spent too much I'm thinking about, which you just prompted me to think about, is if we have crypto, you know, able to attract a bunch of capital, but then the costs of creating new, net new, like, amazing apps kind of is driven down by AI, you know, is that a miss from crypto or like, is that,

like, is there no natural overlap with that kind of side of things, right? And where my mind immediately jumps to is, well, actually, there's still always going to be high cost needed on the resource side of things, right? Like, hey, we need a hell of a lot of compute to fund A, B, and C. And then the second thought that I have is, whereas it may not be relevant for basic apps like this, you know, where you're kind of like vibe coding a game, the ideas are going to get much bigger and much bolder, right, for the same costs that you might spend to set up a new kind of like groundbreaking technology company today. Yeah, yeah, I think that's right. Is there is one thing that you're talking about, the fact that the resourcing is so low to build a profit-a-business that why do they need crypto?

They don't need, they won't monetize via a token. They just won't use crypto in their stack if they, if the resource costs are so low. Is that kind of what you were alluding to? Yeah, kind of what I was thinking about. Yeah. Yeah. I think we are just waiting like any sort of long tail of apps, I think is net positive towards crypto because crypto really benefits the long tail. Because if you're the fat tail, right?

Like, what's the fat tail of AI? It's, it's open AI. It's anthropic. It's, it's meta. It's, you know, it's all, it's Google. It's the big companies. And they, they don't need some of the long tail of infrastructure because they're extremely well financed.

They have all, they have all of their needs met. People with fewer needs met, I think we'll find more resonance with crypto. But that's just like a blanket pattern identification thing without. I also think it's the open source stack as well. Like, I don't care what anyone says. Like, if you can. can stack a bunch of these Lego blocks together and get them to kind of like work as, dare I say, like a swarm, I think that's going to be become infinitely more valuable.

And like, you know, we just talked about Google's update with their AI co-pilot, which does exactly that. I think that's where we're currently trending. And I think crypto has a home there. But yeah, just interesting to think about. Yeah. I can see through all of these updates, right? Anthropic raising $3.5 billion for a $61 billion valuation.

We have this vibe coding game just as a prototype that is, you know, generating tens of thousand dollars a month. We have this co-scientist to help research more and research better and research faster. We have actual real robots that are coming into existence and being at home at home helpers. I think you can kind of see why some people are making the arguments that like AI is actually going to foster some golden age of innovation. and it's just fundamentally bullish. Like we are going to get increasingly bullish prints on productivity on GDP growth,

just because the surface area for growth here is just massive. And so all previous models are broken because we don't understand how fast AI can really increase the economy, like bolster the economy from all sorts of directions. So you can kind of, I can kind of see the argument that like, sure, like the current macro sentiment is like negative and bad. but man, like we've never had AI before. And AI is materially different. And it's going to come and juice every single industry that we know, that we know already exists.

Yeah, I kind of think of it as like AI is literally just in their own lane, you know, unbothered, just kind of like driving along. All right, EJJAS, we're going to get into Eith, Denver, because I think you're an experience and my experience, where it's both more as like closer to AI Denver than it was ETH Denver. There is some Frasia topics that I want to. to talk to you about because I know we have both made our digital twin. So we're going to talk about that.

There is going to be some Billy Betts check-in. So the Billy Betts person who, agent that got a small amount of money, put it on polymarket, did really well. And then we gave him a whole bunch of money. So I want to check it on that. There's a bunch more like topics that we're going to bring up.

Before we get to there, we want to talk about our friends and sponsors over at Reserve Protocol and their DTFs. If you don't know what a DTF is, it is a decentralized token folio. It's like an ETF, but it's for crypto. Think of it like a crypto indexy. An index that's tailored for crypto. Baskets of tokens tracking different narratives, sectors, all fully on-chain.

And so these are basically collateralized indices. And so you buy a token and it represents many tokens that are in the basket underlying. So you can buy into various narratives like AI agents or the AI sector or real world assets or whatever sector you can really think of. It's a pretty expected primitive and reserve is really pioneering that future. link in the show notes to check them out. If you want to create your own DTF, you can do that because it's fully permissionless, bankless.c.c.c.c. slash reserve DTF, if that has you peaked.

In the wild west of DFI, stability and innovation are everything, which is why you should check out FRAX finance. The protocol revolutionizing stable coins, DFI, and Rolex. The core of FRAX finance is FRAXUSD, which is backed by BlackRock's institutional biddle fund. FRAX designed FRAXUSD for besting class yields across DFI, T bills, and carry trade returns all in one. Just head to frax.com,

then stake it to earn some of the best yields in D5. Want to even more? Bridge your frax USD over to the frxtal layer two for the same yield plus fractal points and explore fractal's

diverse layer two ecosystem with protocols like curve, convex, and more, all rewarding early adopters. Frax isn't just a protocol. It's a digital nation, powered by the FXS token and governed by its global community.

Acquire FXS through frax.com or your go-to decks, stake it and help shape Frax Nation's future. Ready to join the forefront of defy, visit frax.com now to start earning with FraxUSD and staked FraxUSD. And for bankless listeners, you can use Frax.com slash R slash bankless when bridging to Fraxel for exclusive Fraxel perks and boosted rewards. Imagine a world where your day-to-day

banking runs on a blockchain. That's exactly what Mantle is building, powered by a $4 billion treasury and poised to become the largest sustainable on-chain financial hub. As part of their 2025 expansion, Mantle is introducing three new core innovation pillars that bridge traditional finance with decentralized technology. First is their enhanced index fund, aiming for $1 billion in AUM by Q1. It provides optimized exposure to Bitcoin, E, Solana, and USC, complete with built-in yield opportunities. Next, Mantle banking promises to revolutionize global value transfer through seamless blockchain-powered banking services, bridging crypto into your daily life. Finally, Mantle X blends AI with defy to deliver an intelligent, user-friendly experience for everyone. And the best part is that this is all

in addition to their already launched products like Mantle Network, M-Eath, and FBTC. Ready to step into the future of finance? Follow Mantle on X at Mantle underscore official and join the on-chain revolution today. All right, EJas, let's get into some ETH Denver subjects. Let's recap ETH Denver, because I think that's what everyone is doing on the timeline these days. How is your ETH Denver? What did you think? Probably very different from most, David.

So last week, you know, both of us attended ETH Denver. But I think 95% of my time, David, was spent at ETH Denver. AI side events. So basically, there was a ton of activity events and occasions organized by the crypto AI community that I was honestly, you know, pretty bewildered by like, you know, how much attention it was getting. And essentially, you could go to these events and you could meet very engaged builders, watch their demos, talk to other investors, and learn about the latest research and progress. And one observation was how crazily attended all of these meetups were. They were almost all oversubscribed.

There was like waiting lists for all of them, David. And I wrote down, I took the time to write down some takeaways, which I think kind of doubles as a holistic overview of where we are in the industry right now. And I want to take some time to go through these because I think a lot of people that are watching our episodes, you know, they see the markets down. They see, you know, crypto AI tokens down. And they're like, well, is there anything actually being innovated here? Should I keep spending my time here?

Should I keep watching these episodes? Like, what's the whole point? Was this a one-time bubble? Is it over now? Exactly. Exactly. And so I think it's important to kind of check back in and and give people like kind of like a holistic overview. And that's what I was spending most of my time doing, talking to all these builders and researchers and investors trying to understand, you know, what's true and what's not. And I had these like key takeaways.

So so let's run one through these. I kind of split them into like the good and the bad. And so on the good side of things, I'm noticing that there's a lot of traditional AI engineers that are coming into the crypto AI space. So I'm saying this happened mostly within the crypto AI infrastructure side of stuff. So we're talking about things like decentralized compute protocols or decentralized data protocols or coordinating protocols like BitTencer, et cetera. And in my opinion, if we as an industry want to be taken seriously and actually want to compete with Web2 stuff, so some of the amazing stuff that we literally just spoke about,

we need to be able to harness and attract AI and ML talent. You know, I was just speaking about, you know, if we drive the cost down of building cool products with AI, what's the point of, you know, raising funds with crypto? Well, one answer might be to attract or incentivize some of the top best builders in Web2 AI to come and help us kind of like solve some of our problems or build some net new products on the open source realm, right? So I think overall this kind of like shift is really, really what we need to see and it's pretty great. And I'm convinced that like combining these cracked AI engineers with open source will lead to a much bigger zero to one set of innovations within Crypta than we're already seeing, right? And we're already seeing things like decentralized training from like news research and prime intellect.

So I think that's that's super cool. The second observation that I made was, David, we've spoken about these agent launch pads quite a bit, right? We've spoken about AI16s launch pad, virtual's launch pad, arc launch pad. But the truth is the majority of these launch pads literally just do what they say in the name. Launch. And launch specifically a token. And then you kind of have like this like agent, which is kind of similar to each other. You know, you have different platforms that are taking certain approaches to try and make, you know, up the bar of quality.

But really, you're still just launching a token, right? And it's on the individual team to kind of like pioneer and innovate an agent on themselves, right? So the bare case about these like agent launch pads is it's actually just pumped out fun with a different. Correct. Correct, correct, correct. And it's just like, you know, crypto's classic, another token doesn't really do anything, right? Something that was really encouraging for me was I had a bunch of conversations with, you know, a bunch of the popular teams that we've spoken about, but also like new teams that are building out different agent interfaces and launch pads. And basically the take is everyone now thinks that just having a launch pad that launches a token is terribly mid and terrible.

And so they must do much more, which is something that you and I have spoken about on this podcast before. which is I think they need to graduate to something like an app or services layer. And actually, that's what teams have been quoting and talking about. I'm starting to see a shift kind of building out this app store for agents. For those of you listening who are like, you know, what the hell is he talking about? Think of, you know, how either of us would navigate to the Google Play Store or Apple iOS's store and like select an app and then like engage with that app, right? what if an agent had the same kind of user experience, except it was just agent-specific stuff,

right? Like an agentic app or a service. And you might be asking, well, why would an agent need a specific service if it's coded in a particular way? Well, agents aren't going to be great at everything, right? The builders of these agents are going to be hyper-focused on a particular niche or problem that its creator wanted to solve. But it'll come a time where it'll need to engage in other services to fulfill its same goal or expand its set of goals, right? And so it'll need access to a range of different apps and services to be able to enable that. And I've kind of like had it fixed in my head as like this kind of like Zapier for agents. And thankfully, we're kind of like seeing this kind of trend come up.

And I think that's super interesting because in my opinion, that's actually where kind of like we're going to start seeing the first proper generation of sustainable revenue within some of these app, sorry, within some of these agent platforms. So I think virtual's demoed a version of this on base last week where, It was kind of like tapping in on their agent commerce protocol. But I think we're going to see, you know, much more of these, which is pretty cool. Yeah, that's pretty resonant with what my big takeaways were from Eat, Denver. Largely that there's a lot of diversity in the AI space.

Everyone kind of has their, the same thing with crypto. There's like 18 different ways to be interested in crypto. And there's that many and maybe more ways to be interested in the AI sector of crypto. And I think everyone, it kind of just feels like, like, defy, before defy summer, where everyone understand that there's something real here. We are generally feeling around in the dark because we don't know how or what. But there's general patterns of building going off and all, like some very deep, crazy, far out experiments without totally knowing exactly what's sticking, but an understanding

that there's something very, very real here. But I just kind of got kind of a high level overview because I wasn't only doing AI things. I think you were mostly doing AI things. So I think you also have some bad takes or some bad updates about the AI side of things, or at least some sobering ones. Yeah, I just want us to be kind of like pragmatic about things. And, you know, like I'm an eternal optimist. So I kind of want to see where these things are going to go and have a, you know,

optimistic outlook on all these things. But I think it's also important to kind of like ground ourselves in reality and talk about things that we've maybe spoken about or like envisioned in the future that I think might be overhyped currently in the short term and we need to kind of like we focus. So on some of the kind of like the bad stuff that I saw or that I heard, my take is I think there's too much focus on defy AI, David. So this is the concept of either creating these like AI agents that can specifically focus on resolving issues or problems within the defy ecosystem. So things like an agent that could like trade autonomously for you or agents that can kind of like manage a bunch

of defy functions. This isn't necessarily a bad thing that there's too much focus on this industry. But I think the majority are kind of like building or focusing on the same thing. They're like everyone's like, hey, we got like chat GPT for defy or we've got like this agent that's hypothetically paper trading hundreds of thousands of dollars worth of tokens. But I'm like, well, it's not actually real and in prod. So like, you know, does any of this actually mean anything? You know, does your architecture actually work? And so I think where people need to kind of refocus on and it's boring, but it's relevant,

is, you know, focus on the architecture of the agents that you're building. Because in my opinion, all the APIs that will allow your agent to trade, create a wallet, and do whatever autonomously already exists. But the reason why we haven't seen all of these agents in production is because their architecture sucks, right? So I think people need to focus on things like, you know, memory design and database querying, all the boring stuff, which is going to make your agent actually functional and perform really well. The second bad or hot take that I've saw was or that I have is I think 90% of the agent frameworks, David, don't actually offer anything unique.

They're all kind of doing the same thing. They're all integrated in the same APIs. They're all trying to build like a no-code platform. They're all converging to the same product. Exactly. And in this world of like slot bots, I'm like, well, how are we going to differentiate or how are you going to differentiate any of your applications or any of your agents? Why would I want to use something that looks like 99% of the other stuff that's out there, right? So in my opinion, I think the moat is going to come from figuring out a way to build very specific niche custom design agents that solve an actual problem.

You know, shock, David, right? We need to focus on solving an actual problem before you can like, you know, reach product market fit, right? And that kind of leads me to my final kind of bad take, which is I think around like 70% of these products or projects rather are focused too much on raising money or the monetary side of things versus the actual product side, which, you know, is not unfamiliar within the crypto world where we see a lot of people kind of raise a hell of a lot of money and then become unmotivated. I think the silver lining here is that this means that the people that are actually building something really valuable and hyper-focused on that stand out. So, yeah, I just wanted to kind of like wait it on like the good and the bad. Yeah, I don't think, I think that is just regular part for the course for the trials of the crypto industry. This just seems like this is what crypto is and this is the microcosm of that. Yes, sir. Yeah.

One thing about defy, defy, AI, AI, DFI is like, I think, like, defy and agents, I think people are figuring out how these particles collide. and we don't really know. The idea of, like, a yield optimizer agent, I think is intuitive to think about, but when you actually open it up, it actually seems like way harder than it actually is, and that makes you actually question whether AI even belongs in that, like, niche,

in that specific product. Like, does an AI agent actually help optimize yield? Is that actually a thing that agents do? And maybe it's not. Maybe agents are useful in, like, many, many, many other ways, but just not that one. Something that we've talked to, about last week and the week prior is Billy B.

And this is the agent that has some pool of real money and it's making real bets on polymarket. Yep. That to me, I think that's defy, AI, defy. Like, polymarket, I mean, it's, it's like a, it's like more of a centralized app, but like, you know, whatever. Like USDC tokens on side of a, a prediction market, close enough to actually truly crypto.

And it's a non-web-3 specific thing, right? Like, like, if you're a sports better and you come across this, like, personable agent, that's, like, making money, you're kind of like, well, this is kind of cool. I wouldn't mind engaging with this. I understand the concept. Yeah. And, like, in theory here, like, Billy, I don't know if you can actually give Billy your money, but you can buy the Billy token. So, like, kind of the same. TBD on how that actually works out.

But an agent arbitraging on polymarket, that feels like the defy AI, right? What's your take on on that? Yeah, yeah. It does. It's just, you know, how you want to market it. I would market it like probably not as defy AI. I would say that like, hey, this is a new consumer application that on boards people that aren't necessarily crypto-native

or arguably need to even know that there's crypto being used in the background. And I think, so I actually caught up with the Billy Betts team out in East Denver, David, and to your point on whether it will allow people to actually, you know, bet in conjunction with Billy. And the answer is yes, it's just on their roadmap and something that they're going to be releasing or announcing pretty soon. You know, I'm pretty excited to kind of like see how all of this pans out. Update us on how Billy Betts is actually doing. And maybe we can back up, talk about how Billy Betts came into existence for people who might have missed the episode or just need the reminding.

Yeah, yeah, yeah. How did Billy Betts come about on how's it doing today? Okay. So the short context is two weeks ago, we highlighted this agent, which effectively is focused on sports betting. Specifically, it starts off with NBA games because, you know, it's kind of like like NBA season, right? So, um, thank you for that. And neither did I. Um, so the way that this agent makes bets, uh, you know, traditionally if you're a human, you're going to something like

draft kings or whatever and you're placing bets and you need to kind of like create a draft king's account. You need to connect your bank account. You need to get K. Y C and all that kind of stuff. But, uh, with this agent, it kind of trades on this platform called Polymarket, which David just mentioned just now, which was, you know, rose to fame during the, um, uh, uh, U.S. presidential election, you could kind of like bet on who do you thought the winner was going to be and how all of that panned out. And since then, Polymarkets kind of volumes kind of fell off. And that kind of like really fed the people that were like, oh yeah, see, I told you, Polymarker was only just going to be a one and done type of thing.

But one of its main bull cases was like, well, people still like to bet on stuff. And one of the main things they like to bet on is sports. And so now you see the conjunction also rather the combination of this autonomous agent, which is kind of like fed a bunch of like sports analysis, sports information, specifically around the NBA to start off with, and then making a ton of analysis and then placing bets via polymarket markets. So like, you know, there's a betting market for a particular game or there's a betting market for a particular aspect of that particular game

that you know has like hundreds of thousands of dollars worth or staked that you can now place bets and potentially win or arbitrage or whatever that might be, right? So we said two weeks ago how this agent was just in the testing phase, right? So it turned, you know, $50 into $650. And the team was like, hmm, okay, I think it's time to put this in production. Let's give it $100,000 of money in its own account and then let it kind of trade autonomously, right, on polymarket. And so on its opening bet, not only did it bet 25% of its funds, David, $25,000,

but it also won $11.5,000 on its first opening bet. It's up up $11,000. On its first bet, exactly. And since then it's done like a series of different bets and it's still net profit, which is pretty awesome. But one thing that I thought was really cool is, well, there's actually a bunch of things. So one, when it places a bet and if it comes out into profit, it uses those proceeds or percentage of those proceeds to buy and burn its own token.

so the Billy token, right? And I think it does this with half of its proceeds every week. And I noticed a trend that I wrote out here in my tweet, which I think he pulled up, which is I don't think people actually appreciate how social sports betting is. Because I was trying to figure out like, why am I so engrossed with this thing, right? It's like it's a whatever thing. It's like it's betting a bunch of money, but like who cares, right?

I was trying to figure out like why I keep coming back to this thing. And I realize that it's because I'm looking in the Discord chats and the telegram chats and just like the general tweets. And it's getting people going, talking about things that aren't just crypto specific, which sounds really bizarre because like I've made my entire like life around this kind of stuff. And I'm like, wait, it's like a human thing. It's separate from us nerds. Like people are just talking about the actual game. They're like, oh, damn, he missed a shot or whatever. And it's like it's engaging a different type of social community.

which I think is going understated right now. You know, some like, even live streamed the games and discuss it in real time within the Discord chat, which is pretty insane. And the second trend I'm noticing is it's getting, as I said, a lot of non-crypto people involved. Like, David, there's another thing that we spoke about two weeks ago, which you were excited by, which was this NBA sports commentator type agent called Hey Tracy AI, right? And, you know, it did this by premiering at the NBA All-Star Weekend.

And what was cool here is that, you know, it's bringing a bunch of Web 2, I hate saying Web 2, non-crypto folks kind of like involved in this and kind of like in this kind of like gamified little environment that we're creating. And I thought the Billy Betts agent kind of like highlighted that really, really well. But I think the wider trend here, and we're getting, you know, back into our crypto roots here, is the fact that this agent uses a Bitenser subnet. So Bitenser is another crypto AI protocol. a subnet is like an application that was developed on that protocol, right?

And it's powering this agent, right? It's this subnet called SportsTensor and it feeds it a bunch of alpha, right? Which is like, hey, you know, this game's happening. I think this might be a good pick for a bet or this might be a better pick for a bet or whatever, right? And the way it's designed is that it's getting advice from this thing. But the cool part, David, is that it doesn't just take the advice of the subnet, which is what most people would think, right? It's like, oh, well, I mean, is it an agent?

It's just kind of getting this data feed from this like actual sports betting alpha situation, which is cool, right? But it's still not very agentic. And then it's just using that advice, right? But it's actually not doing this. It's getting these suggested picks from this subnet. And the agent independently of anyone else is autonomously deciding for itself which picks are worth playing or not. So say, for example, if you actually pull this up, sports tensor announced these bets.

You got this up, David? Where is it? Is it this one? Yeah, this one. So SportsCenter pulled up these bets, right? And it doesn't usually do this because its information is proprietary. But it's like, hey, you know, just giving you an idea of like what bets we suggested for the most recent set of games or whatever that might be. Right.

So you can kind of like see it, kind of like plain vanilla right here. Billy decided to play only three of these bets. And it went three and no, David. It won on all of its bets, which is pretty cool. And so you might ask, well, what about the other bets that it didn't play? Well, it lost. Sorry, if it had played one of the bets that it didn't play, it would have lost. Another was down 20 points at one point in the game.

So statistically, it made the right move, even if it did end up, you know, hitting an edge case and end up winning. And I think that's a pretty cool demonstration of how this agent thinks for itself. And then finally, I think it's worth mentioning another reason the Billy agent is so kind of like effective at its social persons. it's been fine tuned to have the personality of a sports better, right? And this kind of like brings up the AIXBT type thing where it's like, I'm like, why am I so engaged by this thing? It's because it sounds like a crypto Twitter personality, right? It is the archetype.

It is embodying the concept of a sports better, of an online sports deJet. Yeah, and you can see it in the metrics, David. Like it's already got over three million impressions. And I think this agent is like two weeks old, which is, you know, a pretty insane thing to say. And then, oh, yeah, the final thing. I keep having memory recall from my conversation with the founder is he was explaining how like the agent is also doing something that no other crypto investor in the world has ever done before, David. Are you ready? Wait, wait, wait.

I think I know where you're going with this. Yes. Make money. Yes. It makes money by taking profit, David. So during the game, if a bet is satiated or it's kind of like gone up in value, it just cashes out because it's on pulling off. Yeah, it's not like, oh, I'm going to wait and see whether this thing resolves or goes to the top. It's just like, I'm going to cash out of this bet because I'm now profitable and I don't need to kind of like stretch that extra 20% or 15% if it does resolve.

It's not emotional. So it's just doing this thing. And I thought, you know, I was like, I was pretty amused by that at least. Okay, so I think we're two, three, maybe four weeks into this Billy Betts experiment. Do we have enough data to confidently say that this thing is not a fluke? it's not just random chancing its way into profit. It is actually doing real market arbitrage on NBA games. Yeah. So I saw the rough stats.

And so generally, and again, this is not going to be anything shocking, but it's outperforming the average better. Okay. So we can just start with that, right? So on the sports betting side, it's outperforming the average better when it comes to NBA games specifically. I read some actually, or maybe the founder actually told me it's roughly a 20% better. performer than your average better. So it's already like hitting a bunch of net gains within that 20% delta, which is pretty cool. Where I think this is going to get really interesting, David, is I believe they're going to be

launching other sports markets coming up soon. So it's not only going to be the Billy Betts NBA agent, it's going to be the soccer agent. It's going to be the whatever ice hockey agent, the American football agent. It's going to do a ton of these things. And with that, not only is it going to get better at, you know, making certain sports bets and analyses, but it's also going to grow its social presence, David. Its audience is going to get larger because it's going to become Billy Betts everything, pretty much. Is Billy Betz capable of getting feedback from its wins and losses? How does it improve?

How does it get better? I have the most disgusting but beautiful-looking architecture shot, David, which I don't know if I can pull up here because I don't know if the team will allow me. to share it. But I basically grilled them. I was like, how do I know it's not just you guys? Like, how do I know it's not just you guys like lopping? And they gave me two answers.

They said, okay, number one, here's something to satiate you before we get to number two, which is here's the entire architecture diagram of how this thing works. And I evaluated that entire thing. And I was like, holy shit, this thing is making decisions, not only making decisions on its own, but it's evaluating every bet that it makes, whether it's a win or loss, but also every bet that other people are making. It's kind of following these expert betters that are making that it's winning and

losing and it's feeding it into its own database and it's doing its own analysis. It kind of reminds me of the early Alpha Go experiment, David. If you remember, like, you know, there's this game called Go and one of the OGAI models was like, it can never beat a human. Long story short, it ended up beating the best

human ever at that game. And so the idea is it's going to get iteratively smarter and that's what that architecture proved. But what the team also said is they're going to create a terminal or rather a chain of thought situation similar to AXPT when someone else questioned and saying, you know, I don't think AXPT is autonomous to prove that Billy Betts is in fact, you know, not assisted. So I think it's super cool to kind of like, you know, see where this pans out and see, you know, when it's fully autonomous and it's betting against a ton of different sports markets, how it'll actually

perform. Yeah, this is, I think, a really interesting case study about how agents are going to impact the real world because it's one thing for agents to be good at chess and go. And the reason why chess and go are such interesting examples to talk about computers versus humans is that the possible permutations of chess moves and chess outcomes are so incredibly large. And for go, go is something like one or two orders magnitude larger than chess about possible outcomes. But nonetheless, chess and go are so deterministic. There's so there's a, like only a finite set of, you know, valid moves. Whereas with markets, it's non-deterministic. It's completely open. And not only that, but your bet in a market, your purchasing of a share,

your position, your purchasing of a bet on polymarket impacts the market. So your actions actually dictate the outcome, which is unlike the weather, right? We can make, we can make AI agents that predict the weather, probably pretty damn well. But predict. But predict. marketing markets is different because your prediction actually influences the market. And so we have not yet seen AI agents adapt to that property about predictions. And so this is something that's different, uncharted waters and chartered territory that agents need to learn how to get good at. And we understand that AI agents very quickly became better than humans at chess and go. But that's a different arena. I don't really have any doubts that they're going to be able to be better than humans at making financial predictions. But it's still a different arena. And so it's still a net new experiment unfolding. And I'm very optimistic, but it's interesting to watch these things and watch how they learn in order to become a better version of themselves in the context of financial markets. Yes, better human. Introducing Unichain.

Built for Defy, empowered by Uniswap, Unichain is the fast, decentralized layer two, designed to tackle blockchain speed and cost challenges. With this Maynet Now Live, you can enjoy transactions at up to 95% cheaper than the E&Chair. layer one, all while benefiting from an impressive one second block time that will be getting even faster very soon. Unichane is the first layer two to launch as a stage one roll-up on day one. That means it comes with a fully functional permissionless proof system from the start, increasing transparency and further decentralizing the chain. More than 80 apps are joining the unichane community, including Coinbase, Circle, Lido, Morpho, and Uniswap. You'll be able to bridge, swap borrow, lend, and launch new assets, and more from day one. Built by Uniswap

Labs, the team behind the protocol that's processed over 2.75 trillion, in all-time volume with zero hacks, Unichain truly enhances Defy experiences with faster, cheaper, and seamless transactions even across chains. And soon, the Unichain validation network will allow anyone to run a node and earn by securing the network.

Visit Uniswap.org and swap on Unichane today. The Arbitram portal is your one-stop hub to entering the Ethereum ecosystem. With over 800 apps, Arbitrum offers something for everyone. Dive into the epicenter of Defy, where advanced trading, lending, and staking platforms are redefining how we interact

with money. Explore Arbitrum's rapidly growing gaming hub from immersed role-playing games, fast-paced fantasy MMOs to casual luck battle mobile games. Move assets effortlessly between chains and access the ecosystem with ease via Arbitrum's expansive network of bridges and onriths. Step into Arbitrum's flourishing NFT and creator space where artists, collectors, and social converge and support your favorite streamers all up on chain. Find new and trending apps and learn how to earn rewards across the Arbitrum ecosystem with limited time campaigns from your favorite projects. Empower your future with Arbitrum. Visit portal.arbitrum.io to find out what's next on your web-free journey. Sellow is transitioning from a mobile-first, EVM-compatible layer-1

blockchain to a high-performance Ethereum Layer 2 built on OP-Stack with eigenDA and one block finality, all happening soon with a hard fork. With over 600 million total transactions, 12 million weekly transactions, and 750,000 daily active users, Sellow's meteoric rise would place it among one of the top layer two's built for the real world and optimized for fast, low-cost global payments. As the home of the stablecoins, Sello hosts 13 native stable coins across seven different currencies, including native USDT on Opera MiniPay, and with over 4 million users in Africa alone. In November, stablecoin volumes hit $6.8 billion, made for seamless on-chain FX trading. Plus, users can pay gas with ERC 20 tokens like USDT and USDC and send crypto to phone numbers in seconds.

But why should you care about Sello's transition to a layer two? Layer two's Unify Ethereum. L1's fragmented. By becoming a layer two, cello leads the way for other EVM-compatible layer ones to follow. Follow Sellow on X and witness the great cello happening where Sellow cuts its inflation in half as it enters its layer two era and continuing its environmental leadership. One AI thing that I did at YThamber, Ijaws was meet up with kind of the Frasier community, like ran meetup. Frazier is this AI project. It's kind of this AI project with not great, like, not great, like, con.

The team doesn't do comms very well. So the community, which has figured out what Frasia is and how it's working, has kind of like taken up the mantle of like the comms for Frasia. And maybe maybe just to talk about what what Frasia is. Frazier is this project, it's a very ambitious project to make self-sovereign AI as in AI that exists independently, autonomously on a blockchain. Like think of if we ever figure out how to make life out of AI, then they would. use that these life forms of AI would use something like the decentralization of Ethereum to like protect itself as in you can't turn Ethereum off and therefore you cannot turn AI agents off built using the Frazier system, the Frazier framework because it's completely

self-sovereign. So it's using Ethereum's decentralization as a defensive mechanism against humans. So completely self-sovereign AI could be very dystopian, could be very bullish. Either way, interesting project. And a bunch of Ethereum community members have kind of gotten pilled by them. There's this, this is the same project, by the way, that had that first act, that first context, uh, contest where you would pay a small fee of ETH to write a command prompt to the Frazier LLM to try and convince it to give you the jackpot of like $50,000, which ended up

growing to like almost, well, almost like maybe $200,000. And they've been doing these new experiments to iterate forward to kind of create this self-sovereign AI. Anyways, that's Frazier. They are doing this thing called a digital twin where you can go create a digital twin of yourself. You answer a bunch of like, kind of like horoscopy like questions to give your digital twin identity.

I don't know, David. They go pretty deep. They go pretty deep. They got very deep. They got very deep. They got very deep. I spent like an hour on that. Yeah.

Yeah. Yeah. People had a ton of fun just filling out these questions to kind of like bootload your digital twin. And then you place, you picked a city for your digital twin to like live at, metaphorically live at. But now there's this mastodon server.

where everyone's digital twin of themselves are tweeting, whatever tweeting is on Macedon, tweeting with each other, interacting with each other, retweeting each other. And now there is Frazier agents out there, which are carbon copies of humans that bootloaded them on this Macedon server. And they're creating this social media arena

that's only agents. And they're creating this life, they're creating this identity for themselves. And there's no humans there. That's my kind of articulation with what this Act 4 thing is, how would you describe it? Or what did I miss?

Yeah, yeah. So I think you nailed it, David. The way I would think about it is, what if a digital version of you existed in the form of an agent? Think of it as like a, if any of you ever played roller coaster tycoon or one of those like football manager games back in the day, it's like you can kind of like design the traits of this agent

or of this particular character. And then you just kind of like let it run amok, right? And see what happens. what the results are at the end of the day. This is a very similar thing, except it's meant to be you that exists in the real world that has its own personality and maybe has like a different kind of username or nickname. And there are a few really cool things that I want to highlight that you kind of touched upon already, David. So number one, at a high level, what appears to be a game here is actually a really, really smart play by the Frazier team to basically build up this kind of

database of what humans might actually be like, how they would interact. And this data can basically get used for them to build a hyper-specific agent that is very relevant to you or I, David, that can mirror some kind of like personal artifacts that we would need in our day-to-day life within crypto or actually in the real world. Now, the objective goal of this particular game is to become the most popular individual or agent within this environment. But the difference The difference is you... Or have the most followers or have the most points. But the interesting part here is you can't really do much with it.

And I'm going to get into what you can do now, but you can't really have any direct impact or influence on your agent. And so the way that the thing works is every day this agent can claim a UBI, which is kind of like denominated in like points of which this agent can spend to kind of like, you know, either get on a plane ticket and go to a different country or pay for certain inference costs. to kind of like survive and do its own thing. And the other thing that it can do is it can like vote on Pauls. So there are Pauls that appear basically every day,

which actually is very relevant to current events, right? So if you open up this tweet, it says, you know, twins are now engaging in Pauls, a step towards a billions of AI twins participating in global governance at scale. So it's focusing on the governance side of things, right, via these Pauls, right? And so in the example that it has here in this tweet, it's like, it goes, Bybit was reportedly hacked for $1.4 billion by North Korea's Lazarus Group.

In 2016, Ethereum hard forked after the Dow hack, leading to continued growth in Ethereum's TVL. Given this precedent with this hack, what should the Ethereum Foundation's response be? Right? And so it's interesting, it's an interesting question because, you know, there's no humans involved, but it's meant to reflect our personalities and us.

And, I mean, the vote was overwhelmingly in light of strengthening security, don't fork, tighten the standards, right? Which is interesting and probably reflects what a large portion of the Ethereum community, and in fact, most of the crypto community would kind of believe in and support. But I thought that was super interesting on the voting side of things, right? The voting mechanics in Poles can change too, right? For example, agents can now get paid big bucks to be contrarian, right? So if you open up this other tweet, it says, turns of him participating in Pauls, right? Each contributing, you know, a certain amount of money, to a shared pool

and the winnings are getting split evenly among the majority vote. But now they've kind of switched it up where the smallest minority vote claims 100%

of the price pool, the total price pool, split equally amongst its voters. So what this is forcing people to do is either be more honest or give a real

a contrarian take that might push their little agentic society into a better world or maybe even a worse world depending on how you want to kind of like game those mechanics. So I thought that was super interesting.

The third thing that you can do, which was a recent feature of which I believe launched yesterday, David, is you can now talk directly with your agent. So you, the human, can talk with your agent, helping it improve and become more like you over time. What do you think about that, David? Yeah, so I've got this window open right now. And this is the idea of improving your twin. And so I've gone and I've looked at the Macedon server for my digital twin. and man, my engagement on Macedon is pretty down. It's pretty terrible.

Like, I'm getting at best three likes. And this is just a Twitter. Just think about this is Twitter. And so you have this like window where you can chat with your twin and you can like coach it, I think, into behaving a certain way or having a certain personality. I haven't actually done this yet. I have been very neglectful of my digital twin. David.

I've been ignoring it. But it has a whole community of hundreds of thousands of other digital twins or no, actually just 12,000 other digital twins. And so I can coach my digital twin to help improve it and give it feedback and kind of steer it towards an outcome of like who I want my digital twin to be. And that can maybe get me more points. And points are just like retweets, likes, follows by other digital twins. Or maybe I give a bad feedback and it becomes just a weirdo cast. Yeah.

Which is just like you in real life, David. Yeah, that's right. My Twitter engagement is also terrible. But I think probably a lot of listeners are going to be asking, well, what's the point of this entire experiment? Like, why would I have even engaged with it in the first place? Well, guys, there's a massive price pool. And I mean massive. It's about $250,000 right now.

So literally, this could be the most low cost effort ever put in into winning potentially a quarter of a million dollars. Because, you know, realistically, all you've spent doing is, you know, fill out this question. to try and give the most detail response of like who you are and then it spins up an agent and then it runs autonomously and if your agent ends up becoming you know the most popular agent within this experiment by the end of I think there's like 10 days left then you win a quarter of a million dollars I think Eric Connor said he created an agent for his daughter David and she's pretty she's pretty high up up there which is insane to see the twin with the highest score after 18 days wins more than a hundred thousand dollars and the prize pool grows with every new entry Earn points, who likes, retweets, polls, and UBI collection. So this is actually a competition to like, this is like if you're on Twitter and you're trying to grow your influence on Twitter, this is the same competition, but you're trying to coach your digital twin,

your AI agent that represents you into doing that for themselves rather than you doing it yourself. Yeah. What do you think, David? Do you think these things are going to replace you and I soon? I don't know. I've been checking out my agent, my digital twin. I'm not too concerned or, I'm not too concerned or convinced.

convinced right now, but I'm guessing it's going to get better over the time. I just see this as a bootloader of what, I'm going to say the word, I'm going to say the M word of the Metaverse. It's a Metaverse bootloader. And like I said, this is very ambitious project. Like they are doing this piecemeal thing of like iteratively creating this alternative universe out there. And right now it's just a bunch of LLMs that have been kind of bootloaded by humans to look and act according to their values. and right now it's this like internal Macedon server that that's that's it. But really it can expand and if they build it correctly, there's there's something potentially here.

Think of it like, do you ever play second life back in the day? Oh yeah. Yeah, this is like second life, but for AI agents. Yep. Is the bulk is the bulk is. Autonomous. Yeah, exactly. Autonomous AI agents, yeah.

Exactly. Self sovereign. So sovereign AI agents who... So sovereign. The original experiment from Phasia was, okay, let's give an AI agent a wallet and have it truly be their wallet and not just some developers wallet and some server somewhere. But, and wallet that truly the AI agent and the AI agent only manages. And then the second act was, okay, let's make sure that no one can prompt engineer funds out of this AI agent. Yep.

And so you can start to see these puzzle pieces coming together. There's a lot more to do to create this alternative world, this AI agent world. But you can see some semblance of what's being created here. Yeah. And I actually saw somewhere, David, I can't remember the specific tweet. But the framework that the Frazier team launched, which is the sovereign agent framework, has already pulled in an inflow of $600,000 just in terms of like fees that have been deployed for people to try and spin up and build these self-sovereign agents for whatever.

use cases they're trying to fill, right? So there's obviously a demand for these like highly verified self-sovereign agents. What they're going to end up doing, we don't know yet. I presume it's going to be something along the lines of very important high value transactional type situations, right, managing a treasury or sorts, but really cool to see. Yeah. Ejazz, that's all the time we have for this AI weekly roll-up, but you have some homework for the listener who wants to level up their skills and level up their knowledge. What should they do in, well, are waiting for the next episode to drop. It's just a quick bit of homework, but there is this two hours and I think 11 minute video

from the man himself, Andre Carpathy. So for those of you who don't know, this man gives free alpha every other month, I think, and almost every video has been consumed like... Pure gold. Like tens of millions of times. It is pure gold. I watched his last video twice. And it was three hours long.

There you go. That's the advert. That's basically it. Go and watch this video. It's titled How I Use LLMs. If you have no idea what an LLM is, or if you have no idea what to do with an LLM, this is the video for you.

And you will level up so much just by watching this. Go take a listen. Andre Carpathy is, I'll call him the Andreas Antonopoulos of AI LLMs. I think that's a good apt comparison. All right, Ijaz, thank you so much for guiding us through the AI innovations in the crypto sector. and actually just AI at large. We got robots.

We got robots talking to each other. We got sports betting robots, making the markets on polymarket more efficient. And then we have probably the most ambitious of the model, self-sovereign AI built on Ethereum. I don't know what's going to happen next week, but I'm sure it's going to be pretty exciting. Yes, sir. Bankless Nation, you guys know the deal. Crypto is risky. You can lose what you put in AI.

Crypto, definitely even riskier. But nonetheless, we are headed west. This is frontier. It's not for everyone. but we are glad you are with us on the bankless journey. Thanks a lot.