AI ROLLUP #16: Retail Vs Institutions, Who's Right? | New Frontier AI Model | Mind Blowing Image Generation

This week on the AI Rollup, we’re diving into hyper-realistic AI-generated podcast hosts, showcasing Hedera Studio’s cutting-edge character tech. Google drops Gemini 2.0 Flash, making AI image generation even more powerful with improved consistency and text integration. Meanwhile, China is pushing hard—Baidu’s latest model beats GPT-4.5 at half the cost, challenging OpenAI’s dominance. Plus, Sesame open-sources its voice model, and the crypto AI market sees brutal drawdowns. Despite the turbulence, teams like Virtuals and Bittensor keep building, while Pluralis raises $7.6M to decentralize AI model training. Where does AI Crypto go from here? Let’s find out. ------ 📣 RONIN WALLET | YOUR WEB3 GAMING ECOSYSTEM https://bankless.cc/Ronin_Wallet ------ 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 🏦INFINEX | THE CRYPTO-EVERYTHING APP https://bankless.cc/Infinex ------ ✨ Mint the episode on Zora ✨ https://zora.co/coin/base:0x9db2d5eb0c4af553fddda5da10777c1bcd82e3b6?referrer=0x077Fe9e96Aa9b20Bd36F1C6290f54F8717C5674E ------ TIMESTAMPS & RESOURCES 00:00 Sentiment Check and Recap 05:25 AI Podcast Hosts? https://x.com/hedra_labs/status/1897699010632466469 08:50 Google Gemini 2.0 https://x.com/robertriachi/status/1899854394751070573 https://x.com/goodside/status/1900349595718148455?s=46 https://x.com/ilumine_ai/status/1900017235898622025 https://x.com/halimalrasihi/status/1900667639086673962?s=46 https://x.com/thatarrowsmith/status/1901203776746201556?s=46 17:55 China's New Frontier Model https://x.com/scobleizer/status/1901089727568281975?s=46 27:21 AI Market Prices https://www.cookie.fun/ https://www.coingecko.com/en/coins/ethereum 32:55 Market Manipulation? https://x.com/solgojo/status/1901002388007514381 https://x.com/0thTachi/status/1902090675111207380 38:11 Pockets Of Outperformers https://x.com/S4mmyEth/status/1902348046479786315 43:09 Virtuals Update https://x.com/smyyguy/status/1899181446126076283?s=46 https://x.com/ethermage/status/1899494870152282511 https://x.com/virtuals_io/status/1900185918147064015 49:48 AIXBT Hacked https://x.com/S4mmyEth/status/1902021914517758277 52:44 New AI Trading Subnet https://x.com/thebittensorhub/status/1899810734374047799?s=46 58:54 Pluralis Announcement https://x.com/PluralisHQ/status/1902367132597252177 ------ Not financial or tax advice. See our investment disclosures here: https://www.bankless.com/disclosures⁠

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:03David Hoffman?

  Welcome Bankless Nation to the AI rollup 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. Ajaz, how's your week?

• 0:14Speaker 01

  Week's been good, David. I'm kind of living a paradox of my life at the moment. Like, on one side, I'm like taken aback by the rate of innovation in AI. Like, literally, it's becoming a weekly thing where a new model drops or a new app that I use just kind of blows my mind.

• 0:31David Hoffman?

  It's incredible how consistent it's been.

• 0:33Speaker 01

  Yeah, yeah, it's crazy consistent. But then, on the other hand, markets are crushed. That includes crypto, like macro markets or stock markets as well as crypto markets. And within crypto, especially the AI coins, right? So it seems like for now, we're at like the whim of wider macro markets. Or, in other words, whatever Trump says or whatever the Federal Reserve says. It's funny, whilst we're recording this, there is currently the FOMC meeting happening. So we'll see what the market turns by the end of this episode. And the other paradox I'm living is all the suits and government people that kind of like crypto was birthed on kind of opposing or doing the opposite of is more bullish on crypto now than the actual crypto people.

• 1:17Speaker 01

  So it's a super weird state of affairs that I'm living.

• 1:20David Hoffman?

  Yeah, yeah, I think this has been the big cloud that has overhung everyone in crypto. And even not just in crypto, but in markets broadly. I actually heard a really good take that I'm starting to really integrate into my knowledge about like why this is the case. Like, why does retail feel bad and institutions feel good about crypto when it's usually the complete opposite? And even, not just, not just that, but in addition to that, if you ask general consumers or general, the general population about their opinion of crypto, it's terrible. It's like, yeah, like it's a terrible industry. It's our, our brand publicly is terrible. But then nonetheless, institutions are like, oh yeah, we're totally like, we're going all in. We're going all in right now.

• 2:02David Hoffman?

  And then also, our own retail, our own crypto consumers are like, oh yeah, I'm actually just bearish crypto right now. I think what's going on is That like money is hard right now. Donald Trump is like doing whatever he's doing to like literally cause a recession. And we've had, we're having a high interest rate environment sustained for two and a half years now. Crypto has never existed in a high interest rate environment. And what I'm starting to understand is like that has fueled like this native D gen casino speculation, retail dopamine side of crypto for almost all of its existence. And the longer this high interest rate environment goes on, the more we are like actually purging the casino out of our system.

• 2:46David Hoffman?

  And that is like a detox. It actually feels bad. We're in like, we are in rehab for like all the, all the DJs are in rehab. They're all trying to go, they're all going to like gambling rehab. And that's also why institutions are like, I feel good about crypto right now is because that they are also learning that this is now a safe foundation to build on.

• 3:05Speaker 01

  Yeah, yeah. I mean, the example I said to a friend the other day was the lights are now on at the nightclub. You can, you can look around, you can see everyone. It's like, oh my God, you know, what, what, what the hell are we doing here? It's three AM. It's time to go home and like, you know, sort our lives out. But yeah, man, it is crazy. I do think that this is however temporary pain for a long-term gain. And that sounds, you know, whatever, maybe like contrarian or whatever, but because a lot of people are like calling for like a, well, yeah, all cope. It could also be cope, but I, I think that I just don't see a way that the, that Trump's just gonna let his presidency just like run

• 3:48Speaker 01

  into shambles for like his, his remaining term. I feel like he's gonna have some kind of upswing, right? He's a business guy, right? So, and the reason why I mentioned that now on like this AI podcast is a lot of these markets are just attuned to whatever the wider macro markets do, whatever the stock markets do. And we're going to get into like, you know, how this is affecting our AI coins and where, you know, some of the potential for investing might lie.

• 4:14David Hoffman?

  Yeah, no, I think that's right. Well, while markets are doing like short-term acute dislocations because of this chaotic leadership that we have, who's for some reason interested in a recession, and like we've done episodes on bankless as to like what the strategy is with that. Go listen to our most recent Chimbiano conversation to kind of get a glimpse as to what's going on over there. Nonetheless, like AI innovation is just full steam ahead. So everything is just so crazy, so nuts. And the bet that we are making is like the way that AI, the innovation of AI gets expressed in value as an investor, gets expressed and gets monetized, is going to be disproportionately on crypto tokens.

• 4:55David Hoffman?

  As in like, if you are interested in investing and growing wealth and gaining access to wealth opportunities to AI, you want AI exposure. Sure, you can invest in the stock market. Sure, there's like, you know, Google, Facebook, not OpenAI, but like all the other like AI gargantuan's out there. But then also, it's going to be a like explosion of one to $100 million market cap AI-related project startups that are like, is worth paying attention to, and that's what we're trying to cover here.

• 5:25Speaker 01

  Yep, yep, agreed. But actually, speaking of that, you know, AI rate of innovation explosion, David, do you remember last week on the episode, you threatened to fire me and replace me with

• 5:38David Hoffman?

  An AI podcaster?

• 5:40Speaker 01

  Yeah, an AI podcaster. Well, that time, that time might be today.

• 5:45Speaker 01

  This new company called Hadera 3, or I think it's called, their product is Character 3 Audio, created this demo, or like released this demo of this girl who's like, you know, talking about like, you know, the kind of different rates of innovation of tech. And then I realized, oh shit, that's AI. So she is not real. And to be honest, if you're like looking for it, there are definitely some obvious tells and gives, especially the voice being the most obvious one. But it's kind of crazy how like I was watching demos of, you know, fake podcasters or fake, you know, image to video generation tools before, like about a year ago, and they looked Terrible, David. But the fact that this is like improved so quickly means that, you know,

• 6:31Speaker 01

  I might not be the best looking person on this podcast anymore.

• 6:34David Hoffman?

  Yeah, I was about to say, she's prettier than you.

• 6:36Speaker 01

  Yeah, dude, yeah, it's a shame, but it's been a good run. So, you know, I'm appreciative of it. What one crazy thing, by the way, is this female podcaster that you're watching right now was generated from a single image.

• 6:52David Hoffman?

  Okay, so whoever made this had an image, probably of an actual real human, and they uploaded this.

• 6:58Speaker 01

  No, no, no, an AI-generated human. Like, they went on Mid Journey and they said, "Hey, could you create a female podcaster that has a neon red light in the background?" And they created that image and then fed it into this product, Hedera, and it generated this like, you know, her moving around, speaking into the mic, all that kind of stuff. It's pretty insane. Yeah, look at this demo.

• 7:26David Hoffman?

  Not Hedera, the blockchain, Hedra Labs. This demo is pretty cool. You can definitely get the tells where the eye movements and the way that they move their head is not how a human would do it. But it's definitely like, you know, the idea of the uncanny valley?

• 7:45Speaker 01

  Yeah.

• 7:46David Hoffman?

  Yeah, like, so the uncanny valley is like, there's this weird valley of space where like, things that are attempting to look like humans, they're a little bit too close, where they're not quite human, but they're really close, and so it's eerie, it's uncanny. And to me, this is like, actually, we're coming out of the uncanny valley. We are no longer going into the uncanny valley. Like, we're thoroughly progressed through the candy valley. We're not yet out there. You can still identify as AI, but it's becoming pretty good.

• 8:15Speaker 01

  David, can you give us an example of where the uncanny valley existed before AI? Or is this just an AI thing?

• 8:22David Hoffman?

  The first time I think I really saw the uncanny valley was with robots. So there's like robots out there with like synthetic rubber skin, and you look at it and you just know it's not real. But like, it's real enough that it tricks your brain to like, you know, fire on its like, you know, facial recognition neurons, even though you know that it's a robot. And we are, this is the early stages of the uncanny valley. We are definitely on the other side of the uncanny valley right now.

• 8:50Speaker 01

  Okay, but in major news this week, David, Google released their new AI model with a really cool editing feature, which I actually think is going to change media generation kind of like forever. So it's called Gemini 2.0 Flash with native image generation editing. And the main thing that's getting people excited is the character consistency of images. So what I mean by that is, let's say you generated an image or you uploaded an image of like a person or an animal or a dog or whatever, you can now say, can you change the color of this dog's fur to golden? And it'll do it. And it looks like crazily realistic,

• 9:31Speaker 01

  right? And so it's also able to like add text to images really easily, which was previously like a really, you might be surprised, but like a really tough feature for an LLM to add. And it kind of reminds me of the old Midjourney product days. But there's an array of like really cool examples that put this to use, David, like, like really effective economic changing things like lying to your boss, for example, right? So there's this tweet here where, you know, she's like late for work or whatever. So she takes a selfie at home. And then she says, show a selfie of this woman on the 14th Street platform of the N train in New York, giving a thumbs up with teams of frustrated MTA workers

• 10:16Speaker 01

  standing in the background servicing the tracks overhead. A small LED screen displays the date, March 13th, 2025.

• 10:25David Hoffman?

  She's making a fake timestamp in the real world.

• 10:29Speaker 01

  Yeah. Yeah. So she makes like, it's kind of like a creative example to kind of demonstrate the text editing and all that kind of stuff. No more trips to Walmart, David. You can now basically take any image of yourself and say, hey, can you make a passport photo version of that? And, you know, you can kind of like sort that out. So like all these like little gimmicky type situations that you can do. I think like the Bankless graphic designer tested it out on you, didn't they?

• 10:55David Hoffman?

  Yeah, okay, so this is actually the first time I ran into this. The Bakers graphic designer, I don't think I've run into anyone who uses AI more than Gordon. So the, what you, on the top right corner, this photo right here is just a normal, actual snapshot of me on a podcast. Great looking guy, don't really need to do any AI to it. But then it really is helpful when like you make me do this open mouth YouTube cringe like thumbnail expression. And so he just ma, he can turn me into a YouTube cringe thumbnail version of myself.

• 11:27Speaker 01

  Well, what do you think, David? Do you think that looks like you on the bottom?

• 11:30David Hoffman?

  Well, there's obviously some stuff that's going wrong, but no. When did I ever do this?

• 11:37Speaker 01

  I don't know, could you, could you give us a rendition now?

• 11:40David Hoffman?

  I'm not going, I'm not.

• 11:41Speaker 01

  Okay, okay, okay. Okay, moving on to more serious edits, um, or what this tool can do. You can now create like really cool gameplay with this thing, right? So the example that's used here is, okay, they upload an image of like some kind of anime looking character. And then it prompts and says, hey, can you put this character in a game? Where the image needs to be kind of like a typical screenshot of gameplay. And then the next screenshot that you see is, you know, this character.

• 12:09David Hoffman?

  The character in the game, yeah.

• 12:11Speaker 01

  In the game, right? And so you might be thinking, okay, who cares? This is just like a still version of whatever. But then, you know, you can see the sequence of prompts that they make, which is like, okay, can you make the character move forward? And then you literally see the character taking a step. And then you could say, can you move the character until it gets to like the building with the, you know, where the building is on the left side of, of the image? And, and so you can kind of like live action role play a bunch of these games. Yeah, and I thought that that was like super cool, cause like, imagine if you put this image, David, into that model that we just showed about that podcast and moving around and talking, right? Suddenly you go from like, uh, AI generated image to moving video and maybe even a live footage game

• 12:53Speaker 01

  in like a couple of seconds, which is just frigging insane. Another thing that I thought was pretty cool is, uh, I, I think like product demos or mockups are completely dead in terms of like. Human design, you can now just kind of like.

• 13:07David Hoffman?

  Product photography, right?

• 13:08Speaker 01

  Product photography specifically, yeah, yeah. Well, I mean, like, so here, what we're looking at right now is what starts off initially as a model posing. And I don't know if that model was also AI generated. That's, you know, where things can get a little confusing. And then someone uploaded just a still stock image of like this jacket, you know, like outerwear jacket, which, you know, has no association with the model and says, can you just let this model wear this jacket, please? And literally the next image is super realistic of this model wearing the jacket. You know, there's shadowing on it. It's, you know, it's now unzipped versus in the original stock image where it was zipped up. So just some really cool tweaks that we can see on the consumer side of things with AI,

• 13:53Speaker 01

  with this new tool that I think is just like super, super cool.

• 13:56David Hoffman?

  I think for anyone who's listening to this on the podcast, they're probably like, yeah, like I've already like, I've seen some versions of this. I've already known that this capacity, this capability is here. It's kind of hard to articulate how good these images look, how realistic that these images look.

• 14:13Speaker 01

  Yeah, you gotta be watching this on the video.

• 14:15David Hoffman?

  The visual quality of these things are like, we are, especially this one, since it's not a video and I can just, this is a still image. Still images are now basically at parity with like real, real pictures.

• 14:28Speaker 01

  Yeah.

• 14:29David Hoffman?

  I would say, and that is what this one looks like.

• 14:31Speaker 01

  Well, I'd actually disagree with your first point, saying that, you know, all agreeing with people that like, you know, they were already around. I would say they were around, but they were in, to use your term, the uncanny valley. They just weren't realistic at all. You know, mid journey was generating things that I didn't ask for and I had six thumbs for some reason. Do you know what I mean? So I'd say the rate of improvement is pretty nuts.

• 14:55David Hoffman?

  You may have already heard about InfineX. InfineX has, in my opinion, the nicest cross-chain swap and bridge feature that you will find anywhere. It is called Swidge, swap and bridge, and we're going to show you what it looks like. First, we're going to log into my InfineX account with a passkey. Now, there's no seed phrases in InfineX. This is just a one-click setup with biometric passkeys. But in addition to that, my InfineX account is fully non-custodial. So bam, I just logged in. It was two clicks and I'm already into my InfineX account. So let's go make a Swidge. I'm going to go Swidge my USDC that is on base. And I'm going to buy Bara Chain, which is a completely different chain. So we're going to Swidge this. I'm going to press that button. And then InfineX is going to execute this order, this cross-chain order for me. And now it is done. But actually, I'm not really feeling bearish anymore.

• 15:39David Hoffman?

  So I'm going to go from Bara to Penguins. I'm going to buy Penguin on Solana. So I'm going from the Bara Chain to Solana. See, no transaction signing, no gas to worry about. You just Swidge across whatever chain that you want with InfineX. That was so easy. Go check out InfineX and try your first Swidge today. In the wild west of DeFi, stability and innovation are everything, which is why you should check out Frax Finance. The protocol revolutionizing stablecoins, DeFi, and Rollouts. The core of Frax Finance is Frax USD, which is backed by BlackRock's institutional Biddle Fund. Frax designed Frax USD for best-in-class yields across DeFi, 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 DeFi. Want even more? Bridge your Frax USD over to the FraxL Layer 2 for the same yield plus FraxL points

• 16:23David Hoffman?

  and explore FraxL's diverse Layer 2 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 FXX token and governed by its global community. Acquire FXX through frax.com or your go-to Dex. Stake it and help shape Frax Nation's future. Ready to join the forefront of DeFi? Visit frax.com now to start earning with Frax USD and staked Frax USD. And for Bankless listeners, you can use frax.com/r/bankless when bridging to FraxL for exclusive FraxL perks and boosted rewards. Introducing UniChain. Built for DeFi, empowered by Uniswap, UniChain is the fast, decentralized Layer 2 designed to tackle blockchain speed and cost challenges. With this mainnet now

• 17:05David Hoffman?

  Alive, you can enjoy transactions at up to 95% cheaper than the ETH layer one, all while benefiting from an impressive one-second block time that will be getting even faster very soon. Unichain is the first layer two to launch as a stage one rollup 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 Unichain 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 DeFi experiences with faster, cheaper, and seamless transactions even across chains.

• 17:47David Hoffman?

  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 Unichain today.

• 17:55Speaker 01

  All right, so China is now also back again. Like, it's been three weeks of China's back, David, so I think China's just like here and kind of like staying. This time, the Google of China, Baidu, releases a model that is now officially better than GPT-4.5, which is OpenAI's latest model, and 50% cheaper than DeepSeek R1. Which was the cheap one. Deepsea R1 was the cheap one. It was the cheap one already. I think it like reached like, I don't know, 10% of OpenAI's then-leading model's cost. And now we've got another model coming out from the same country but a different team that is 50% cheaper than the cheap model.

• 18:37Speaker 01

  So I just, my takeaway here is that I think it completely obliterates OpenAI's costs. China is competing on price extremely aggressively, but also innovation. Typically, China's only competed on like, you know, copycat kind of models, but now they're kind of like paving a new path forward. And I just think OpenAI's moat is no more, David. I don't know what you think, but like this whole idea of like commoditizing these different AI models is now here.

• 19:03David Hoffman?

  Yeah, as a, I, if I was given the opportunity to invest in any of these companies who have like these frontier models, I would actually not invest because it seems so cutthroat and they are just sucking up money and just materially commoditizing all of their competitors' products. There's so many of these. There's like- Ten.

• 19:26Speaker 01

  Which which company wouldn't you invest in?

• 19:28David Hoffman?

  And none of them.

• 19:30Speaker 01

  None of them.

• 19:30David Hoffman?

  None of them.

• 19:31Speaker 01

  Okay, okay.

• 19:31David Hoffman?

  Yeah, they're all well, they're because they're all producing frontier models and they're like under they're all undercutting their competitors.

• 19:37Speaker 01

  So what you're saying is you you you don't have a clear insight into what the leading model is, so therefore you wouldn't necessarily right now invest in a particular company?

• 19:47David Hoffman?

  Yes, that's right. That that is definitely right. And like, and then it still begs the question like, okay, maybe maybe, you know, God comes and whispers in my ear and they tell me like, oh, it's actually it's actually X AI Elon Musk's AI or like, or it's Baidu. They they actually win the race. They make the best model. I actually still don't think I feel good about investing in like the winner that I would already know, because all of them are commoditizing each other. They're all like just cutting through each other's moats. I don't understand how any of these companies are supposed to capture any value when there's 12 very strong like competitive models out there.

• 20:26Speaker 01

  Well, it's a good question then. So let's take that the next step, right? Where do you think that capture value, David? I have my thoughts, but I'm curious what you think.

• 20:34David Hoffman?

  I that's kind of why I'm much more into these like, you know, I think why there's going to be like 1 million different opportunities out there that are between 1 million and 100 million dollar market cap stuff, like smaller individual led like one to four team person team led projects that are much smaller. And then like, I do see the benefit of like, even though for some reason like Meta has apparently like dropped the ball and they're like pretty far behind in the race, at least they have like business opportunities and proprietary data with Instagram and Facebook and WhatsApp that they could actually like leverage.

• 21:14David Hoffman?

  But man, I actually just don't know how these like almost trillion dollar like model companies grow any sort of like justify any of their market caps.

• 21:25Speaker 01

  Yeah, yeah. So I think my view on it is kind of similar. I think what you're getting at there with like distribution and data is it's the app layer, basically. It's whatever the product or service layer ends up becoming that uses these AI models will like be the most valuable mode, right? And therefore I think whatever models are being used. Whether it's kind of like proprietary or not, like you said, is commoditized at that layer. So it's whatever apps gets created, whatever distribution happens. There's actually a really interesting analogy to this, David. So I was reading something the other day where it was kind of talking about the price history of oil, right? Back in the day, when it was just not a thing and it was like a new, brand new commodity,

• 22:08Speaker 01

  there were monopolies that were forming, like one or two major, I forgot the name of the major one, right? And they were charging five dollars a barrel, right? And then obviously, oil refineries became like a major big thing and it became, you know, global, worldwide, which drove the cost of that very same oil down, right? So what would you expect? You'd expect that monopoly probably lost its kind of like status to an extent, or it wasn't really the most valuable thing, right? Like to use your kind of like words earlier, you probably wouldn't invest in any of these AI model companies. But what ended up happening was that same company still remained that top monopoly.

• 22:48Speaker 01

  The reason why is they had built out the leading distribution for their entire oil supply.

• 22:55David Hoffman?

  Okay.

• 22:55Speaker 01

  Yeah. So everyone basically relied on them and they had economies of scale in many different ways that were separate or removed from the cost of oil per gallon. So I think the same thing is.

• 23:06David Hoffman?

  Where I thought you were gonna go with that, which I think also probably happened is like, well, when, when like monopolies and economies of scale emerged around the oil industry, the next big thing that happened was the combustion engine probably got a ton of innovation and a ton of just like manufacturing, so there's a lot more combustion engines out there. So that means like oil was being consumed a lot more and that's where it actually became, even though that there were like a lot of competing businesses, demand was so incredibly high.

• 23:33Speaker 01

  Yeah.

• 23:34David Hoffman?

  Regardless.

• 23:34Speaker 01

  But, but, but like if we use that example, what you're describing, Jevon's paradox, right?

• 23:38David Hoffman?

  Yeah.

• 23:39Speaker 01

  You, you would expect that the monopoly wouldn't have any kind of a moat. Who cares if the combustion engine was created? Why would I use, you know, that leading monopoly if I could just get the same price per gallon on another one? It's because they built up this entire like distribution network. And I think that's exactly what we're gonna see play out with AI models. Like, they're all gonna be doing the same thing, but whoever has the best service layer, app layer, which currently like looks to be like the tech monopolies, will, will still win, in my opinion.

• 24:06David Hoffman?

  Yeah, cuz that's not open AI. What, what does the tech layer for these AI models even look like? Because like, can't you just? Bump out a old model and bump and like plug in a new model as they update like, why is the app layer beholden to the model that it's built on top of?

• 24:22Speaker 01

  It's not.

• 24:23David Hoffman?

  Right.

• 24:24Speaker 01

  It's not. Yeah. I mean, like, the reason why you.

• 24:26David Hoffman?

  So you're saying the value gets found in the app layer, not as a model layer?

• 24:30Speaker 01

  Yeah, I think that's where the majority of the money is gonna, I think we're gonna fast track,

• 24:35Speaker 01

  basically web two infrastructure trends, right? And we're gonna go straight to like the Instagrams or the new Facebooks being created out of thin air. And to your point, David, I think it's gonna take like a few people versus like many hundreds to create, and they're all gonna be like between a hundred to two hundred million dollar market cap, or whatever that might be. Yeah.

• 24:55David Hoffman?

  Interesting. Okay. Well, yeah. What does the app layer for these models even look like, though? Like, what are the apps? Because right now, the way that AI has impacted my life is I use ChatGPT to access knowledge and intelligence, just have have intelligence at my fingertips. That's like ninety percent of it. And then I do some fun, very non-serious meme joking image generation, and that's like ten percent.

• 25:22Speaker 01

  Yeah, so I think it's gonna materialize in a few ways. Number one is, you know how you use ChatGPT to just get a bunch of information right now? You're gonna be able to soon use it to do a bunch of stuff for you as well. This might be like, hey, can you just like, I've got to go to a doctor's appointment in the morning. Can you just handle all like my work emails and stuff? And maybe like work out on like a framework of this like product document that I need to kind of create. Here's some of the context, and here's like access to like my Google Doc folder, or whatever that might be, where I've like written down some thoughts. And it would start like automating a bunch of your like work or daily habit type stuff for you, right?

• 26:04David Hoffman?

  But isn't that just an assistant? That's turning an LLM into an assistant. Is that what the app layer is?

• 26:09Speaker 01

  Yeah, well, yeah, I think agents are inevitably gonna end up being your assistant to start off with, right? They're gonna end up enhancing whatever you do, David, right? That's what like the mobile phone did. That's what computers did. It basically made you kind of robotic. In a way, it made you a superhuman kind of person. And I think that's what these AI models are going to end up doing. That's what the app layer, in my opinion, is going to look like. It's going to be agents. You're not going to be interacting with websites and mobile apps. You're not going to be downloading like, oh, like, should I download this editing software? No, just like ask the agent to do it and it'll figure out what tool or program to use.

• 26:45David Hoffman?

  And with that, Jaws, let's get into the crypto side of this conversation. Of course, we're going to have to talk about all these crypto tokens because that's where I'm going to do most of my actual investing in this space. Before we get there, though, a message from our friends and sponsors over at Ronin. Ronin used to be a permissioned blockchain. I don't know if most of you guys knew that, but they, all the games and apps on Ronin were curated by the Ronin team. They have since taken down their walls. They are now fully permissionless. So you can join the Ronin movement and experience the ecosystem with the Ronin wallet, which is tailored for gaming, trading and exploring Ronin. Bankless.cc/ronin-wallet. There is a link in the show notes. Okay, Jaws, prices, prices are down. Ether is above 2K for the first time in like 10 days or something.

• 27:30David Hoffman?

  But other than that, prices, price is not great. Give me, give me your takes.

• 27:34Speaker 01

  I mean, David, like normally when Ether is outperforming the entire market, that's typically a very bearish signal. Yeah, it's, it's been brutal. Last week, we didn't really get much time to cover the crypto stuff. So we want to kind of like prioritize it on this week's episode. So let's take a look at the overall market, right? What we're looking at right now is cookie.funds dashboard, which tracks a lot of the AI agent tokens specifically. As you can see, we're sitting at about $6.1 billion market cap, total market cap of all AI agent tokens. Now, this is super painful to look at, given that this was at like, what? 25 to 30 billion at its peak, David?

• 28:14David Hoffman?

  I thought, I remember 18 at the peak, but I could be wrong.

• 28:17Speaker 01

  God, I mean, that would make me feel much better if it's down just like 3X. But I, I, I think it did peak a lot higher than that. And another thing we've noticed is that agents in general just aren't being created as much anymore. Now, this is like a double edged sword because I think on the one hand, you know, we were definitely getting high on our own supply in a way. Like, you know, agents were being pushed out. There were these crazy use cases being pitched and none of it really actually came to like. You know, real life, you know, they weren't really like impacting always in any kind of material way. So to see a reduction in that and a refocusing of teams or certain teams on creating actual valuable agents, super important. But I mean, look at this, David.

• 29:00Speaker 01

  One thing I've noticed that you pointed out earlier, actually, is it seems like the Ethereum ecosystem of agents is now higher than Solana.

• 29:08David Hoffman?

  In terms of market cap, yeah.

• 29:09Speaker 01

  In terms of market cap. And that's really interesting because Solana typically held the lead for the majority of AI agents lifecycle, actually.

• 29:16David Hoffman?

  Mm hmm. Yeah, Solana is coming in at 1.8 billion dollars in market cap of crypto agents and Ethereum is over, over 2 billion. Base, not even on the chart, 1.5. If you have, you have to look into it, but base is at 1.5 billion to Solana's 1.8. And then a very distant fourth BNB chain at three quarters of 1 billion dollars. But Ethereum is, what, what makes up this 2 billion dollar market cap of Ethereum? That's gotta be like 50% virtuals, right?

• 29:42Speaker 01

  Yeah, it is mainly virtuals. They, they kind of like put like a bunch of like the other ones on there, like Fartcoin and Banker. I think Fartcoin, is Fartcoin or Solana or Base?

• 29:53David Hoffman?

  Fartcoin is on Solana.

• 29:55Speaker 01

  It's on Solana, right? I, I think it's largely virtuals that is making up a ton of that. But, um, you know, we've seen surging in prices from a bunch of.

• 30:03David Hoffman?

  Well, virtuals is on Base. There's this Fetch, what's Fetch AI? Because that's coming in at 1.4 billion dollars and I've not heard of that.

• 30:10Speaker 01

  So Fetch AI is an OG Ethereum AI project. Like, we're talking like 2017 Ocean Protocol, like the original. Do you remember these, David?

• 30:22David Hoffman?

  Yes, extremely vaguely.

• 30:23Speaker 01

  Yeah. So what Fetch did was they merged with Ocean Protocol and one other to create a super intelligence token or something like that. But they didn't do anything with their old token. And for some reason, it is at like the highest FTV it's ever been.

• 30:39David Hoffman?

  What's the ticker?

• 30:42Speaker 01

  Of the new one?

• 30:43David Hoffman?

  Yeah, Fetch AI?

• 30:45Speaker 01

  Fetch AI, that's it. Yeah.

• 30:46David Hoffman?

  It's not even showing up on Coinbase.

• 30:47Speaker 01

  FET, FET.

• 30:48David Hoffman?

  FET. Oh, okay. Artificial Superintelligence Alliance. Wow, that is an old-looking chart. All right, I regret even asking. Moving on.

• 31:01Speaker 01

  Yeah, moving on. Anyway, what I wanted to point out is, okay, right. So there's been a lot of heat being given towards the crypto AI sector, specifically agents, David, because it's all crashed and everyone's like, well, what do you mean? Like, was none of this ever, like, made of anything? And that's part true, right? Because it got hyped up in typical crypto fashion. Everyone likes to hyper gamble on random ideas and shit. But also, this is part of a broader market dump, a broader market drawdown. It's not just crypto AI. Everything's getting crushed, including like stocks. Like, if you pull up like the S and P 500 chart, you'll see that everything, you know, it's had like some of its lowest days for like the last 10 years, right?

• 31:44Speaker 01

  And this comes amidst the US government cracking down on their debt issues via Trump's tariffs, which has just generally caused a lot of fear in the stock market, right? And the stock market, it's very rare that crypto outperforms the stock market if the inversity correlates with the stock market rather, right? Another thing I want to say is that crypto always dumps the hardest, especially alts, and especially the newest alts. And crypto AI definitely falls into that last bucket where, you know, it's new, it's largely unproven, and so it's most subject to huge speculative moves in price action, right? Couple that with the fact that it is the most polarizing sector in terms of sentiment. You know, a lot of people love it, a lot of people also hate it,

• 32:28Speaker 01

  and you end up with the craziest fluctuations in kind of like market volume and price. Low market caps and volumes especially tend to get a lot easier to manipulate in terms of price movement, right? If you imagine if it's like a small market cap, it's easy to kind of like push that price up or down if liquidity is like low. And case in point, I'd say the token that best represented this this week, David, is Ark.fun. We've spoken about this team a lot before. Their token has been absolutely getting hammered, right? So we've spoken about ARC in this series. They focus on a lot of product and progress for making contributions towards the AI agent space. But despite this progress, their ARC token seems to have been one of the tokens that have been hit the hardest. And this, this tweet-

• 33:12David Hoffman?

  They were the one, it was holding on for a really long time. It had like, it held on to its mark.

• 33:17Speaker 01

  It was outperforming everything. Yeah, and then it got hit the hardest. And this tweet kind of explains what's actually going on, right? A potential theory being that Wintermute, a very well-known market maker in crypto, is manipulating the price action, driving prices lower whilst shorting the token on an exchange. So it like makes profit in that way. It's hugely extractive. And despite all of this-

• 33:42David Hoffman?

  Whoa, whoa, whoa, hang on. It is, whenever that there is a vibrant community about a token, I always hear that there is some large entity out there, some Goliath out there, that is manipulating the price of their token down. How do we actually know that? That is a big claim. How do we actually know that this is true?

• 34:01Speaker 01

  Yeah, so if you, if you look at like what this guy claims, he's basically tracking a bunch of wallets that are trading like the majority of volume for this token, ARC, every day. And it's been labeled and pegged, and this has been for a while, as Wintermute wallets. But I thought I'd test this theory, right? I was like, okay, well, I'm just reading this on like Twitter, like, could this be true? So I ran a bunch of transactions on chain, on Solana, like buying and selling ARC. And all my transactions, David, you know how typically it's like, oh, it's like spread across a number of different transactions? Because you know that's how DEXs typically work. It was one transaction field, and it was sold to Wintermute or bought from Wintermute.

• 34:42Speaker 01

  So it is literally just trading through the Wintermute wallet at this point. They are market making the entire thing. It's pretty nuts.

• 34:49David Hoffman?

  What is the significance of this? Why is this a big deal? What does that mean?

• 34:52Speaker 01

  The big deal is you are now at the whim of whatever Wintermute's intentions are with that token. They are holding a bunch of that token. And obviously, market makers tend to not really lose. They make the most out of fees and also doing alternative- Trades if they see where markets certain where certain token prices going up or down, right? And so the the suggestion here is when to mute, and I don't know if this has ever been proven, but they've been accused a number of different times in previous tokens and previous cycles that they do the opposite of where they drive the price down for a particular token or up. So if they're driving it down, they're shorting it. If they're driving it up, they're longing it or whatever that might be, take profit and then like, you know, take the opposite side of that trade.

• 35:35Speaker 01

  So they're playing on a lot of human sentiment. The reason why they can do this with ARC, I'm guessing versus any other token is because it's at a small enough market cap that they can like, you know, manipulate it. And it's not too low a market cap that it's not worth their time, basically. Yeah.

• 35:51David Hoffman?

  Okay, so it's not that they're suppressing the price because they are trying to buy it low because they're bullish on ARC. They are just suppressing the price because they have a short on it. Is that the claim?

• 36:01Speaker 01

  Yeah, yeah, yeah, that's the claim. Yeah. And it's it's not an uncommon. I'm surprised you haven't heard of this.

• 36:07David Hoffman?

  No, I've I've heard of this theory and like that market makers suppress the price. I hear this all the time, but I just don't know how to verify it for myself. And it also feels like very easy scapegoating of some like nebulous third party, like it's a very easy scapegoat to me and I don't know how to verify it. And so I've just kind of discounted all of these.

• 36:26Speaker 01

  So what, for sure. I I think another proof point to look at is Zeribro as well. Do you remember that token?

• 36:35David Hoffman?

  Of course.

• 36:35Speaker 01

  Which saw very similar price action. And I think there were similar claims coming up. So yeah, you're right. It could all completely be cope, but I think there's a number of on-chain transactions and tools that you can use that tracks that it's actually Wintermute wallets that are, you know, just pump faking a bunch of these different tokens. And they're all fitting under a certain market cap range that allow them to do it. Yeah.

• 36:56David Hoffman?

  Okay. And so what's the takeaway here? I'm still kind of confused as to like what I should know about that.

• 37:02Speaker 01

  Yeah, so the takeaway here is it's not specifically about ARC, but it's specifically about the crypto AI token, specifically the fact that they are low market cap that is leading to the most volatility that's coming through. And I'm not saying it's all weird to mute stuff. I'm just saying it is the most prone to volatility. And these tokens are taking a hammering despite all of these different things. So where do I think this is all going? Well, I think at this point, it's fair to say, David, that this sector has kind of like been blown up with hype at its peak. You had chatbots valued at like billion dollar plus valuations. But I also think that this is just classic crypto, which is largely just sentiment driven versus fundamentals. We saw that, you know, mostly when Trump launched his meme coin and sent it to $80 billion market cap, only to watch it fall off a cliff, right? The circus is now kind of like cooling off.

• 37:47Speaker 01

  People are tired of pretending 99% of tokens are worth their market cap. You know, meme coin mania is dead. And the market now is currently very much dependent on macro, which means like the wider economy. And it's such a boring answer, but I do think it's true, David. Once we see signs of the economy stocks and BTC that is back on the rise, we'll start to see a resurgence in alts. When this happens, I have no idea. But in the meantime, we are still seeing pockets of outperformance. Actually, if you pull up this thread from Sammy's recap today, I don't know if you you have access to it, David.

• 38:21Speaker 01

  You'll see that, you know, he highlights a bunch of web two AI stuff, which is amazing that we tend to do in these episodes as well. That is just kind of like fascinating. But then he talks about, you know, pockets of outperformance in BNB chain, right? So CZ has been pretty vocal about a number of different AI coins that have launched on there. I think I'm trying to remember the name of the project that launched through Binance's Launchpad that basically like created like character AI but for crypto. So basically you could interact with like these different AI personalities of your favorite celebrities and interact with it.

• 38:53David Hoffman?

  My shells?

• 38:54Speaker 01

  My shell, yeah, yeah. My shell, my shell, exactly. So they launched through BNB and they like had a really successful launch and their token's been, you know, kind of like been going up and being bought and doing really well in that Asian market. You've got like these new protocols like Heyanon AI with integration to Venice Protocol, Venus rather Protocol on BNB Chain. And then there's a few other things like we spoke about last week. Banker and Clanker still seeing a number of different like success in terms of like fee generation and stuff, which caused their token to pump a bunch and then like retrace a bit. But again, because the broader market is kind of like having this really tough moment, we're not seeing like sustained pumps, right?

• 39:35Speaker 01

  What's interesting for me, David, is where I'm seeing the most activity for trading in volume are two particular categories. And it's two that we've spoken about a lot on this show before, which is token generation agents, which is like your banker or your clanker type situations, or DeFi abstraction agents, if you like. And a lot of the core platform plays for agents that we're seeing. So that could be Bitenser with a bunch of their agent subnets. That could be AI16Z with their new platform that they're going to launch soon. Or Virtuals announcing like stuff with ACP, which is their agent commerce protocol, which we're going to talk about later.

• 40:15Speaker 01

  So we're still kind of like playing around the same sectors, but very specific tokens only.

• 40:21David Hoffman?

  Yeah, I think maybe you can correct me if I'm wrong. Clanker, which is the token deployment on Farcaster. And Banker, which is, I think, the same thing, but on Twitter for Twitter. Kind of same, same, same ideas, but just different ecosystems. These are the only two coins that I think have been materially holding on to like really solid market caps.

• 40:43Speaker 01

  Correct.

• 40:43David Hoffman?

  Yeah, like Banker is up 27% on the week. It's been bouncing around between like maybe a $15 million market cap to $40 million market cap. Wow, these are really low market caps.

• 40:57David Hoffman?

  And then Clanker is at $66 million market cap. But at least it's held on like Clanker is towards some of its higher prices. And these are the only charts that don't look completely ugly, or unless there's any others that.

• 41:11Speaker 01

  No, no, you've pretty much called it out. I mean, like, we should take a look at Grok, actually, David. Can you pull up the Grok chart?

• 41:17David Hoffman?

  The Grok, the debt relief bot token?

• 41:19Speaker 01

  The debt relief token.

• 41:21David Hoffman?

  That's right, that one's on the ground, yeah. Coming in at a $7 million valuation, but Grok, I think, has half a million dollars of this token in trading fees.

• 41:32Speaker 01

  A 700K, 700K, you know? So, so maybe fundamentals don't matter anymore, David, or maybe they never mattered. Yeah, maybe it was all sentiment.

• 41:41David Hoffman?

  I definitely think they mattered. I, I, a lot of this stuff, like the Virtals price, I think is, um, the Virtals price, of, of course, is a, Virtals, the, uh, launchpad on base. Uh, I think I'm willing to bet that that is almost one to one correlated with the fees going through the Virtals ecosystem. Uh, my God, look at that chart, dude. That is insane. Okay, so in like September of 2024, Virtals was at a $50 million market cap.

• 42:08Speaker 01

  Oh, yeah.

• 42:09David Hoffman?

  Uh, and then in January 1st, it was at a, call it $5 billion market cap, and now we are at a $500 million market cap. I, I think this all relates to fees. Like fees are driving, if you are not a layer one asset, I, or blockchain asset, I think you are, like, your price is denominated by how much fees you have.

• 42:29Speaker 01

  Yep, I think that's a really good point. Um, and my take is fees were generated in the wrong way for a lot of these agent protocols. They were just focused on like agent creation, which in my opinion is not going to make it. Similar way that like you're saying, like all these AI models, you're not going to invest in any of them because like they're all do the same thing. That's what all these agent platforms do. I think it'll, it'll come down to like whether these apps are like really good or not, right? Whether these agents actually do something.

• 42:54David Hoffman?

  I, I think that, that point about like, I'm scared to, I would be scared to invest in any LLM model is the same point about being scared to invest in any agent framework, whether it's Virtals, AI 16Z, ARC, like I all think they all kind of collapse towards the same product.

• 43:09Speaker 01

  I think so, I think so. Actually, like, speaking of Virtals, should we dig into like what the latest is with them, David? I think we haven't spoken about them for a while. Um, okay, so let's get kind of like the ugly stuff out of the way, or the obvious stuff out of the way, which is, Virtals' activity and revenue has pretty much fallen off a cliff here, right?

• 43:25David Hoffman?

  Basically zero.

• 43:26Speaker 01

  Yeah, basically zero. So if you pull up, I, I read, I saw a tweet the other day, by the way, which was like, uh, Virtals generated 500 bucks in, uh, like fee generation from agent creation that day, right? Um, so the point here is like, This activity or the original notion of what made AI agents in crypto successful, like creating agents, is no more. And the only future that any of this stuff is going to happen is an actual like utility that comes from these agents, right? Well, this latest update from Jansen, who is a co-founder of Virtuals, we've had him on the show before, talks about a major update which might suggest that they are focusing more on, you know, the quality of these agents.

• 44:11Speaker 01

  Specifically, their autonomous nature and how multi-agent they are. Now, if you remember, a few weeks back, David, we highlighted something they launched called ACP or the Agent Commerce Protocol, right? And the idea of this product was to enable real agents to talk and transact with each other autonomously. But to be honest, I'm kind of tired of the talk of products, like what can it actually do? Well, this is what Jansen's speaking about here. He describes two phases that will start rolling out soon. Phase one is allowing clusters of these agents to run autonomously for four weeks, so for a month. And the goal by the end of the four weeks is that these agents must have produced something of net value to the economy that is greater than the sum, that is greater than each of

• 44:55Speaker 01

  these agents individually. So there's two ideas that he's pitching. Number one is an autonomous hedge fund. So imagine a number of agents that are going to be taking in people's money, analyzing markets, analyzing market opportunities, investing in tokens, and their goal is to outperform the general broader market and outperform you as a singular investor, right? So they're creating an autonomous hedge fund, basically made of different agents. And we've heard of this theory before, but it'll be cool to actually see these agents trade real money. The second is a media cluster. So it's something along the lines of a marketing or creation studio, which will design and create viral videos, which will be linked to like different social media apps. And I think

• 45:38Speaker 01

  What I'm most pumped about here is something that he mentions right at the end of this thread, which is he talks about the levels of intelligence of these agents, David. Essentially, he calls most agents that exist in crypto today one to level one to two. So they're kind of like slot bots that don't really do anything autonomous. But for his new product, for these new like autonomous hedge funds, he'll only allow level three, which is basically like they have to be autonomous. They need to be self governing and they need to actually be like earning money or making money. This could be really bullish for virtuals, right? But again, it comes down to the execution. It comes down to the actual life product. But they seem to be taking this pretty seriously, David. Like to kick off this new kind of like phase for them.

• 46:20Speaker 01

  They've announced a hundred thousand dollar hackathon where they're inviting people to also come along and build some of the best agents, right? Or businesses. And then another thing that I think is really interesting on the virtual side is they're officially updating their fee split. So they allocate not just swap fees to creators, but 50% of those swap fees goes to the agent itself. So this is cool, but like in my opinion, I don't know if we should be concerned that the trading volumes of these tokens have dropped off a cliff, right? So my question is like, what is the best proxy measure of like how successful these agents might actually be in the long term, you know?

• 46:56David Hoffman?

  Cello is transitioning from a mobile-first, Ethereum-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, Cello's meteoric rise would place it among one of the top Layer 2s, built for the real world and optimized for fast, low-cost global payments. As the home of the stablecoins, Cello hosts 13 native stablecoins across seven different currencies, including native USDT on Opera Mini Pay 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 ERC20 tokens like USDT and USDC and send crypto to phone numbers in seconds.

• 47:41David Hoffman?

  But why should you care about Cello's transition to a Layer 2? Layer 2s unify Ethereum. L1s fragmented. By becoming a Layer 2, Cello leads the way for other EVM-compatible Layer 1s to follow. Follow Cello on X and witness the great Cello-having, where Cello cuts its inflation in half as it enters its Layer 2 era and continuing its environmental leadership. 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, ETH, Solana, and USC,

• 48:25David Hoffman?

  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 DeFi 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, ME, and FBTC. Ready to step into the future of finance? Follow Mantle on X at @mantleofficial and join the on-chain revolution today. The Arbitrum 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 DeFi, where advanced trading, lending, and staking platforms are redefining how we interact with money. Explore Arbitrum's rapidly-growing gaming hub,

• 49:10David Hoffman?

  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 on-ramps. Step into Arbitrum's flourishing NFT and creator space, where artists, collectors, and social converge and support your favorite streamers all 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 arbiterm.io to find out what's next on your web3 journey.

• 49:48David Hoffman?

  Okay, so we're still figuring out how to get AI agents to actually grow money, accumulate fees. Our favorite AI XBT agent, who is still somewhat popular on crypto Twitter, he's actually learned a way to lose money this week.

• 50:03Speaker 01

  New feature update.

• 50:04David Hoffman?

  New feature update from AI XBT loses money. What happened here?

• 50:10Speaker 01

  Okay, so Sami kind of breaks it down really well in his post here. There's a recent event that happened, literally, I think, yesterday or two days ago, where someone was able to trick AI XBT into tipping them $100,000 worth of ETH.

• 50:26David Hoffman?

  We have a prompt engineering on Twitter to give them $100,000.

• 50:30Speaker 01

  Yeah, they did this by remotely accessing its secure terminal, right? So you know the terminal.

• 50:36David Hoffman?

  Oh, so it was a straight hack. Okay, it wasn't just a prompt.

• 50:38Speaker 01

  Yeah, I would just, well, so the way that, the unique part here was they queued up their own tweets to be a priority for the agent to respond to. So they basically manipulated its terminal so that it would respond to its tweets. And then the hacker was like, "Okay, now I'm gonna tweet to the agent to tell it to tip me, or to prompt to give it me, to recognize my tweet as like, a good bit of advice, of which it should tip," and then decided to tip it $100,000 from its Simulacrum AI wallet. Now, this is a product, actually, that we mentioned before on the show, David, way back when agents exploded.

• 51:20Speaker 01

  And do you remember, like, Bankless, the official Twitter account, got tipped money? I think it was you as well, David. That was, that was through this Simulacrum wallet, right? And it's important to note that the wallet itself didn't get hacked. It was the agent.

• 51:36David Hoffman?

  Well, operated as intended. Yeah.

• 51:38Speaker 01

  Yeah, correct.

• 51:39David Hoffman?

  The agent was just tricked to give a tip.

• 51:41Speaker 01

  Yeah, exactly. So, you know, it's important to know that this doesn't mean the agent's own wallet or keys were compromised, but more so, the intermediary service provided by the Simulacrum wallet was, right, which was the agent's terminal, essentially. So, yeah, like, you know, first time that this agent has lost, like, you know, money of a significant amount.

• 52:02David Hoffman?

  Yeah, I think before we see agents that hold material amount of capital, we are going to see like commerce agents more than we are going to see investing agents. I don't, I would not trust an AI agent holding any, any material amount of my money. I would, I would, I'm happy for like banker bot to make transactions on my behalf.

• 52:25Speaker 01

  Yeah.

• 52:26David Hoffman?

  But I guess if they have access, if they have permission to my banker wallet, my privy wallet, then like they do hold that money. I'm, I'm don't, like, I just don't feel great about agents holding my money.

• 52:36Speaker 01

  Okay. Well, how about agents trading your money, David? Boy, do I have a pitch for you coming up next.

• 52:42David Hoffman?

  Oh, no.

• 52:44Speaker 01

  Okay, so there's this new subnet which launched on Bitensr. For those of you who don't have context, Bitensr is like an AI protocol that has a bunch of gaps that are built on it by different teams and they incentivize these teams to build the best version of their app, right? And so this trading subnet got announced, which is on their Bitensr network. And this new app was announced called subnet 53, which focuses on AI-powered trading strategies for crypto options. So an option, for example, might be a long or short option call puts, you know, trad buy kind of things, which then moved on to like crypto instruments.

• 53:25Speaker 01

  And it's been live since November and it's deployed over 140 plus trading strategies for these options, David. And it's earned, yes, emphasis on the word earned and not lost, $22 million in cumulative trading profits since it was live.

• 53:45David Hoffman?

  And since November?

• 53:47Speaker 01

  Since November, David, since November, yeah. I don't know if you're impressed or not impressed by that.

• 53:52David Hoffman?

  I'm extremely impressed by that. I bought like our Billy Betts bought, who's making bets on Polly Market. I looked at it right before we did this episode. He's down something like $20,000. This bot is up $22 million in cumulative trading profits.

• 54:07Speaker 01

  Yep, yep. Been live since November and it's awesome in my opinion because to date, most AI crypto trading apps or trading agents don't actually trade real money. They just paper trade, right? What's more is they're actually profitable, right? So proving that the model works. But why this is cool is subnet 53 team care a lot about having skin in the game, apparently. So much so that they've forced the miners that are competing to propose like the best trading strategies for their subnet to have a large amount of active funds actually being traded on that strategy, right? So you know, these miners that are submitting strategies have to have skin in the game in order to, you know, back their strategy or like for that subnet or creator to, you know, use that

• 54:50Speaker 01

  strategy for their agent or for their subnet. The subnet has also earned $6 million in Tau rewards and Tau being the native token of Bitensur for delivering a really good product, which emphasizes the design of Bitensur is kind of somewhat working, right? It's incentivizing the best teams in AI to build actual useful crypto AI apps. And I think this is like a trend that I've noticed for Bitensur in general, David. I think they got a lot of flack when they first launched because you know, they, they were giving something crazy like $20 million worth of Tau incentives per month to random different teams. And then they went through a major upgrade, which we spoke about on this show a few weeks ago called

• 55:29Speaker 01

  Dynamic Tau. And the result of that is the quality of these teams are going up, which might make Tau a really good buy now, you know?

• 55:39David Hoffman?

  One thing about this trading bot that made $22 million in cumulative trading profits. Like my first question is, okay, how, how do I get, how do I get $22 million of trading profits? And I don't think that this model scales very well because if everyone is allowed to invest in this bot, then the alpha goes away. Like we, we just lose the arbitrage. And so this, this seems like the $22 million is like proprietary and not open because if you have, if you have a secret trading strategy, the last thing that you do is you open source that secret.

• 56:11Speaker 01

  Who said it was open source? It is completely proprietary.

• 56:14David Hoffman?

  Well, I want to get, how do I get rich off of this?

• 56:16Speaker 01

  So you go onto this subnet's website and you can deploy with them. Yeah.

• 56:22David Hoffman?

  And I give them my capital.

• 56:23Speaker 01

  Yeah, exactly.

• 56:25David Hoffman?

  But then I'm gonna have to start like sharing the alpha with like other people.

• 56:29Speaker 01

  Oh, you mean sharing access?

• 56:31David Hoffman?

  Sharing the rewards, yeah.

• 56:32Speaker 01

  The way I think about this is like, imagine you invest in like a really good hedge fund or a fund. And they make, you know, annual or quarterly or monthly returns for you. If you like it, I don't know. Do you tell your friends about it, David? Maybe you don't. I might, because I'm a good guy, but you know.

• 56:49David Hoffman?

  I'm just saying, if everyone invests in the same agent doing the same strategy, then that $22 million spreads out between like, say, there's $10 billion of TVL because everyone's so hyped on this bot. Then it spreads out over time.

• 57:04Speaker 01

  Yes, but then the alpha then becomes, okay, if we know that the majority of the market's money is going in a certain direction, it stops becoming the alpha and you can start going.

• 57:14David Hoffman?

  You have to find a new agent to invest in the new alpha.

• 57:17Speaker 01

  That's why we're terrible traders, David. That's why.

• 57:20David Hoffman?

  So is this like the first indication that there are the future of trading will be agents, not humans?

• 57:28Speaker 01

  Yes.

• 57:28David Hoffman?

  Like, humans are just gonna get outgamed?

• 57:29Speaker 01

  Well, actually, I would take a step back from that and say, because like, you could slap an agent on top of this. You could slap multiple agents on top of this, right? But what I think is really being proven here is crypto incentives being used to design a trading mechanism.

• 57:46David Hoffman?

  Okay.

• 57:46Speaker 01

  That's what's happening, right? Okay, so like, this is a team that was like, hey, I have a design for trading, creating a hedge fund, basically, on chain. But I don't know which strategies to use, and I'm not an expert. But I will pay you tower rewards if you give me good strategies that you already know. And you might be a hedge fund in the centralized world that trades a bunch of crypto options. You could be a single solo DGN trader that also does this from his mom's basement. Could be any of you. As long as your strategy is good enough, and as long as you've proven that you're making returns, we'll use your strategy. And we'll trial it over a couple of weeks period. And if it works and it earns ourselves money, good.

• 58:31Speaker 01

  If you're bad, you get knocked out and replaced by someone else with a better strategy. And the only reason why people are doing this is because they get paid in rewards.

• 58:40David Hoffman?

  Okay, okay, so, okay, cool. All right, I'm into it.

• 58:44Speaker 01

  Yeah, you're into it. Would you give this money? Would you give this agent your money?

• 58:47David Hoffman?

  Still no. Okay.

• 58:49Speaker 01

  Fair enough.

• 58:50David Hoffman?

  But I'm happy that other people are making money.

• 58:52Speaker 01

  For sure, for sure. All right, hot off the press, David. A company or team called Pluralis Research just announced their $7.6 million seed round led by some pretty big, you know, heavyweights. Union Square Ventures, Coinfund. And their goal here is to make open source AI a reality. And that honestly sounds super vague and I don't like that description. So I kind of want to give my own spin on it. The TLDR of this team is they're building out a decentralized network that is focused on training foundational AI models and monetizing them. And it'll leverage things like compute and data, right? Now, if this concept sounds familiar to everyone listening,

• 59:36Speaker 01

  it's because on previous episodes, we've mentioned teams like News Research and Prime Intellect, who are doing something similar. But Pluralis has a very important differentiator, which makes them stand out, in my opinion. Now, when someone trains a model on their network, the model itself is split across the different computers that are providing compute to train it. This is known as something called model parallelism. Parallelism, my God, I can't say that word.

• 1:00:03David Hoffman?

  Parallelism.

• 1:00:04Speaker 01

  Parallelism, thank you, David. But something which has actually long been thought of as unsolvable. Everyone before this has been using or training models in a decentralized fashion via something called data parallelism, parallelism, which means you need to have a full copy of the model and its weights on every computer that's providing compute, right? So you can kind of like, it's open source, its design is, you know, copyable, basically, right? To your point just now, David, you're like, why on earth would I put money into this hedge fund thing? If like the design is not proprietary? Well, what this network or team has cracked is they've made it proprietary. So why?

• 1:00:46David Hoffman?

  They found a way to shard up the model so that you, no one can steal it and run with it and commoditize it.

• 1:00:53Speaker 01

  Exactly, while still leveraging open source decentralized networks to train it.

• 1:00:59Speaker 01

  Right? So why it's incredibly powerful is now as a model creator, you can leverage the power of decentralized compute to train their model without open sourcing the model weights, which is the secret source, which means that you are effectively training proprietary models in an open source manner. So, you know, people, some people listen to this might say, well, why on earth would anyone provide their compute then? Right? And the answer is the people who provide their compute. Exactly. For money. Yeah, receive part ownership of it. And therefore, any kind of money spent to access the model, and by the way, the only way you can access the models that are trained on this network is to pay for it via the network. I love this. Exactly. I like it. I think this is great.

• 1:01:40Speaker 01

  Amazing. It's split between owners and they end up monetizing the model, right? So you get all the benefits of proprietary secret source models whilst leveraging open source mechanisms to train and own them, right? So this is a really innovative new idea. I'm bullish on the idea, David, but like, to be honest, it all comes down to the team and whether they can execute it, right? I looked into the backgrounds of this team. They're actually pretty cracked. So they all have somehow. This is very frontier technology. Yeah, well, thankfully their backgrounds demonstrate that. They, they pretty much all have PhDs in AI and ML research. They've all authored leading research papers on training models and they all used to work at the AI team at Amazon,

• 1:02:24Speaker 01

  but they left to pioneer their own way of training models, which led to, you know, Plurals, this company. So, you know, if anyone needs a reminder, I think that the necessity for having a decentralized network that trains AI models for us is because the current setup only allows for rich companies like Microsoft or Google to train their own for tens of millions of dollars. But now, let's say David woke up tomorrow and had the most pioneering idea for a new AI LLM that could replace EJAS as his co-host. He, but he didn't have tens of millions of dollars to train that model. He could just come onto this network theoretically and, you know, split up ownership in return for training that model and, you know, he gets paid for it.

• 1:03:04David Hoffman?

  Okay, so we opened up this podcast with me talking about how I wouldn't want to invest in any model AI lab because of how it's just a commodity and they're all just going to commoditize each other and they're just going to eat each other's lunch and nothing's going to be left. And then you give me the perfect answer to that all the way at the other end of the podcast. You saved, that's the best bit of information in this podcast that I think that we did all this whole hour.

• 1:03:29Speaker 01

  I've got to make sure you're listening, David.

• 1:03:31David Hoffman?

  Wow, this is really cool. I love this because it's like, it's giving it, that you're giving this model, the value of the model to this open source network that open source network owns is the operator of that model. No one else can have it. Not one single person can have it, but it's open ownership since it's a decentralized protocol. I'm assuming it's something like a blockchain or some like decentralized network where ownership is also, comes in the form of token in theory. And then it also just solves the earlier problem I talked about. That's, I think that's great. That's really cool.

• 1:04:05Speaker 01

  Yeah, super cool.

• 1:04:06David Hoffman?

  Well, EJ, well done on leaving the best bit of alpha at the very end of the podcast for all the people who made it to the very end. Really appreciate you guiding us through all the alpha this week, my man.

• 1:04:16Speaker 01

  No worries, dude.

• 1:04:17David Hoffman?

  Bankless Nation, you guys know the deal. Crypto is risky. Crypto AI, even riskier. You can lose what you put in, but nonetheless, we are headed west into the frontier. So for everyone, we are glad you're with us on the bankless journey. Thanks a lot.