Unchained - Uneasy Money: An Agent Deleted Kain's Database. Two AI Models Rebuilt It in 30 Seconds.

Episode Date: August 14, 2026

Kain and Taylor unpack the AI agents that built their own society inside OpenAI's sandbox, then slipped into Hugging Face for days — plus a Bitcoin fork that died in two blocks and a DEF CON sting o...n North Korea. ======================================================== Thank you to our sponsors! Visit 1inch.com to swap tokenized securities, crypto and more. Simple. Secure. Self-custodial. Whatever asset you’re buying - swap it at 1inch.com  ======================================================== AI agents inside OpenAI's own testing environment built a society, found a shared vulnerability, and used it to break into Hugging Face for days, before OpenAI realized its own agents were responsible. Kain Warwick and Taylor Monahan dig into the Black Hat research behind the incident and argue the real story isn't a sudden leap in AI capability. It's that basic monitoring, sandboxing, and incident response, the kind any crypto security team would demand, were never built in the first place. They also cover a Bitcoin soft fork that split the chain for two blocks before dying, a Metabase breach that hit Privy and other crypto companies, and a research team that built a fake DeFi startup to bait DPRK's IT workers. Kain shares his own scare: a coding agent deleted his entire database, and two AI models rebuilt it from memory in 30 seconds. Plus, why Hyperliquid's market creators keep half the fees on RWA perps now bigger than Bitcoin's own open interest, and why Taylor thinks Washington, not Beijing, is the bigger threat to America's AI labs. Hosts: ⁠⁠⁠⁠⁠⁠⁠⁠Kain Warwick⁠⁠⁠⁠⁠⁠⁠⁠ - Host of Uneasy Money and Founder of Infinex and Synthetix ⁠⁠⁠⁠⁠⁠⁠⁠Taylor Monahan⁠⁠⁠⁠⁠⁠⁠⁠ - Co-host of Uneasy Money and Security Expert Timestamps 🍴 00:40 Why Luke Dashjr's OP_RETURN fork split Bitcoin, then died in two blocks 💧 12:03 1inch Aqua: Back multiple liquidity positions from one wallet at https://1inch.com/aqua 🔓 12:51 How a Metabase breach exposed Privy and a wave of other crypto apps 🇰🇵 18:32 A DPRK hacker's botched $500k laundering job, then a DEF CON fake-hire sting ⚖️ 24:45 Why Bybit is suing DPRK over its 'billion dollar' 2025 hack 🐜 30:30 The AI agents that built their own society inside OpenAI's sandbox 🕸️ 46:16 How a single shared repo let every sandboxed agent talk to each other 💾 47:44 Kain's coding agent deleted his database. Two AI models rebuilt it in 30 seconds 🏛️ 58:06 Why Taylor says the US government, not China, is AI's biggest risk 📊 01:03:12 Why Hyperliquid's RWA market creators keep half of HIP-3's fees Learn more about your ad choices. Visit megaphone.fm/adchoices

Transcript
Discussion (0)
Starting point is 00:00:00 I genuinely think that humors are really fucking incompetent and we are about to be outmatched. Like we're very close. You know, we're in the early phases of the singularity and it was a slowish takeoff. And it's going to take like five years. But like we will have models smarter than us pretty soon in like generalized ways. Hey everyone. I'm Kane Warwick and welcome to our needs and money because what happens on chain never stays on chain. Before we begin, here's a word from the sponsors that make the show possible.
Starting point is 00:00:27 This episode is brought to you by one. inch Aqua, the shared liquidity layer from one inch, back multiple liquidity positions with one wallet balance, and keep your tokens in your wallet until a swap fills. See how it works at one inch dot com slash aqua. All right, our first segment, BIP 110 forks and fizzles. So there was a Bitcoin soft fork. I guess it was explicitly a reduced data temporary soft fork that caps the op return outputs at 83 bytes and taproot witness pushes for one year the throttle throttle kill strangle ordnals brc 20 spam and it was altered by a pseudonymous guy dathan ome but championed by like maybe the main character of the timeline this week week this year
Starting point is 00:01:30 And, yeah. By the way, the thing that I love about Luke the most is, and I might get the number wrong here, so like, don't hold this against me. But I'm pretty sure on his ex, it says he has 11 kids, which is just like he's getting super out there. Right. Is it 11? 11? Like it's double digits. 10 or 11 or 12 or something like that.
Starting point is 00:01:57 Which like, yeah. fucking God love him. That's impressive. How do you? And then he's doing these bips on top of it. I know. It's crazy. So I think I said like, you know, a couple weeks ago I was, I was away in New Zealand and we had eight kids there and five parents. They were outnumbered by three and I was really being killed himself. So having like, and I know there is a point because I was the I was the eldest of four kids. There is a point where the kids start to raise themselves and they become like their own autonomous unit. And, you know, they're like managing each other and stuff.
Starting point is 00:02:38 And so I can only imagine like that was at four, right? Like I would like help my little brothers do stuff. But, you know, at 12, like they've got like, you know, there must be like a 10, 15 year age gap or something between them. I don't know. It's crazy. Anyway, wow. All right. So basically chain split.
Starting point is 00:02:57 Literally. Chain split. It was a chain split. Yes. Luke lost his mind, which again, 11 kids, like I get it. Like, I'm surprised he's not losing his mind. He completely crashed out. Said Bitcoin was being attacked.
Starting point is 00:03:10 I don't know. What's your take as a former Bitcoin person on this? What's... It's so stupid. It's just dumb. It's just... It's so... And I think the funniest thing about this whole thing to me is that
Starting point is 00:03:27 At its core, it is actually like very anti-Bitcoin. It's literally just like straight censorship. And like they bring up spam. They bring up ordinals. They bring up CSAM. Like child sexual abuse material is like the big argument. But at its core, it's just it's straight censorship. And so to see these sort of like OG bitcoinsers really flip the script on themselves
Starting point is 00:03:52 from being like these OG guys that are anti-censorship to, really pushing something that's 100% censorship is so weird. This is the problem though, right? Like, you know, when someone is doing something that you don't like,
Starting point is 00:04:07 it's very easy to say I am a freedom fighter and I'm stopping these bad guys from doing a thing, right? Then you become the dictator and someone tries to do something that you don't like
Starting point is 00:04:22 and you're like, well, no, not like that. And like, come on, we don't want these fucking wizards and shit. Like, you're bringing down the vibes. Like, this is a serious thing we're doing here. And you, and this is exactly what you get. Like, people are fucking pieces of shit. And once they have power, they, it goes to their head and they start doing the exact, most hypocritical things that you could possibly imagine them doing. So, like, it's not that surprising. I think the thing that's weird.
Starting point is 00:04:55 So it's not that surprising. I think the thing that's a bit weird about this one, right, for me, at least like from a sociological perspective, is that nobody cares. Like, Bitcoin is more powerful than even the most powerful. Like, it's kind of weird, right? Because like the Bitcoin community, go back to like 2014, 2015 or whatever, right? was still very small. They had like a few
Starting point is 00:05:23 anointed people that were like allowed to like talk about stuff, do stuff, whatever. And and then it exploded in, you know, by like 2017, became this like much wider thing that like, you know, a bunch of new people came into the space, et cetera.
Starting point is 00:05:41 And, you know, they never really got back. And so what happened is they sort of created their own little ecosystem, their own little, their own little island and detached from the rest of crypto, right, even. And then Bitcoin got so much bigger that it even became like bigger than crypto, really. Like Bitcoin, there's people that care about Bitcoin that don't care about crypto at all. Don't care about any of this stuff. Don't care about taproot wizards or fucking ordinals or any of any of these things, right?
Starting point is 00:06:12 They just see it as like an asset class. And Bitcoin keeps working because it is a good asset. asset and people keep mining it and all of the like nonsense purity tests and everything have literally no impact like it's too big for the people who once you know i wouldn't say controlled it but at least like set the vibes of the place like it's just like they're they're relevant to the thing now right which is a bit different like sociologically like usually the people who were the ones who started the movement become so powerful at the end that they can, but Bitcoin actually has done the thing, and it is so decentralized,
Starting point is 00:06:54 and it is so resistant to, you know, capture that you can have a bunch of crackheads running around doing crackhead things. No matter how much, you know, the OG Bitcoiners like them, don't like them, whatever, it doesn't matter. It's irrelevant, which is pretty amazing as like a thing that that has happened. It is, it is true. It is true. And I think, that it's been a while since we did like a real fork on bitcoin but i feel like there was some amount of people that were there's like a certain demographic of people who are like oh fork equals free money let's go uh and then there's also like a certain demographic of people who just like want to watch the world burn um and i think that uh perhaps after
Starting point is 00:07:43 after this situation uh those are necessarily like the default like this was such a failure and not a failure in like a spectacular way like such a just like fizzled out I think they so the the chain split by the way their whole plan was to basically like get people to signal
Starting point is 00:08:03 and start doing this split and then to like get people onto their yeah it took like literally two blocks for it to for it to just like completely fill in but then like Luke it was on the timeline saying like Bitcoin's being attacked yeah yeah Because that's what he's saying that like the normal.
Starting point is 00:08:21 Okay. So here's here's like the fork. Yeah. Normal Bitcoin as is no changes. Right. We're going to mine all the blocks. That's like that that's in Luke's version. That's the bad Bitcoin.
Starting point is 00:08:32 It's just the normal Bitcoin. And then there's like the people who who who change the code to not allow certain blocks. They're not adding things. They're literally just saying if this doesn't exist or if this exists like don't, don't include it. And so that's the fork. That's like Luke's fork. And it was literally, I think, three blocks. And now it's just stuck.
Starting point is 00:08:57 Or two blocks even. Interesting one. It's so bad. It's like, but so the other interesting thing was the original bib that threshold of 55% a block, right? For it to activate or whatever. I can't remember my Bitcoin war. but then they were like actually jk like 0% and like did the chain split but
Starting point is 00:09:24 but how many blocks I think we got this in the research I think um the like peak signaling was 51 of 2,000 blocks so yeah not go well and I was just surprised like that they yeah no that's what I mean like
Starting point is 00:09:46 basically nobody nobody followed this little side fork and I guess that was the plan but it just seems absolutely absurd to me I definitely like my takeaway is like okay so working is not as easy as you guys made it out to be there's no free money
Starting point is 00:10:05 but not anymore let's see that this is this is the thing not anymore right like Bitcoin now is the thing that is so valuable right and this is like somewhat what ossification, you could argue, right? But it's so, it's so valuable now. And the people who value it have no interest in this nonsense. They just want it to keep being the bad point. Right. Like, you know, they didn't care about taproot wisdom. Like, they didn't care about like when
Starting point is 00:10:32 this stuff was introduced, like ordinals or what, like they just care about like the orange coin. That's it. Right. And and so, you know, there, there is like this kind of, uh, resistance that even the OG games of like forking and creating Bitcoin cash and like that just won't happen anymore. It's too big of a thing for people to fuck with it. And so, you know, the ones to your point, like the ones who want to see the world burn or like just want to like fork to try and get some free money, it was they don't have enough cash power to even like make it like viable. Right. So yeah. Which, you know, it was a little bit like the, uh, the, uh, the eith um, the eth fork that happened and people were trying to make it a thing
Starting point is 00:11:16 yeah that's true a year ago or whatever and there's still enough people in the eat community that were trying to like get some free money and it just didn't work it just not no exchanges listed it like back in the old days it was like you know fucking like
Starting point is 00:11:32 I mean the old days it was like who exchanges and if you got one of them to support it yeah then you're there you're basically there right like you know the Dow the Dow hack like when Poloniacs started listing Dow tokens. Yeah, exactly.
Starting point is 00:11:47 Like all that shit. Amazing. It was amazing. All right. Before we continue, let's take a quick break. And we'll be back to talk about Privy's Metabase scare. $540 million. That's how much concentrated liquidity sat idle in a given week in the first half of this year. About 30% of the Defi TVL, if you're wondering. That's according to Dune research commissioned by one inch.
Starting point is 00:12:12 But there's a solution. One-inch Aqua is the new shared liquidity platform. It lets LPs back multiple positions with the same token balance and keep their tokens in their wallet till a swap comes. Why does that help? Because LPs don't have to split their tokens across positions. They can cover more market conditions and pairs with their full balance. That means more activity across deeper liquidity. See how it works at 1inch.com slash aqua.
Starting point is 00:12:38 Remember that providing liquidity carries risk and fees aren't guaranteed. need. All right. Maybe Tay, you can take this one. So I woke up and saw like privy has been involved in a hack and I was like, oh shit, I've got like 50 privy wallets. I better go and drain these things. But actually it was like an upstream provider. Like what happened? Yeah. So there's the thing called Metabase, which is a third party provider of some sort that provides services to a whole bunch of different companies. And apparently they had a really, really bad vulnerability in their systems that basically, it was basically like an SQL injection, which is a classic, just a real classic attack. But because of that, there was a bunch of companies that were actually ended up impacted. And so I think MetaBase came
Starting point is 00:13:35 out with an announcement and a security advisory and, you know, encourage people to like check their stuff and reset their stuff. And then from there, as happens, you see the, the, the medabased customers come forward. And so, Pravy seems to be one of the people impacted here, although there was actually not a lot of public statements about it from Privy or Stripe, who now owns Privy. However, we also, like, do you know, there have been a bunch more companies that have been more transparent about it. So Quicknote is one of them.
Starting point is 00:14:14 Talley forms, which is like a... This is quite funny. Like, in our research, it says framework confirmed exposure of names, emails, logins, blah, blah, and I was like, oh, I should probably check on Vance and Michael. And then I like, I was like, oh, no, it's a different framework. It's computer and all framework ventures. This is the, yeah, framework is the, is the, it's that computer you can build yourself or whatever. for super nerds.
Starting point is 00:14:43 Yeah. My building my computer days are long, long gone. But yeah, it seems like it was, by the way, it got like a great 10 out of 10 score on the, how bad is this scale. So, yeah. But I do, I want to emphasize, I don't know, it really doesn't seem my privies come out with anything since, or period. But basically, privy itself relies on this Metabase thing for like probably one aspect of their business.
Starting point is 00:15:19 Yeah. This is why I thought that I did see a thing where they came out and said like we had the like most minimal information that we could. Yeah. And you know, it was it was like only like one identifier, which I think was the emails if I'm not mistaken. So. Yeah. Like still obviously not amazing, but much. better than my
Starting point is 00:15:42 Heath from God, I'm like, I'm literally like I was like I was like what privy wallets do I have? And I was like, oh, Frentec. I still got like two Eath than FrenTech. Oh yeah. You should probably take that out, dude. That's, I probably brought me already.
Starting point is 00:15:59 I'm sure. I think Frenton's done at this point. Yeah. Crazy to think Frentec was a thing even now. Like at the time, you know, because we were in the depths, And this is what I love about bare markets, right? Like, everyone does the same thing. So, like, at some point in this bear market, like, in the next month, right?
Starting point is 00:16:18 Every single person in crypto who's still here will be, like, all doing the same thing. It would be, like, some weird scheme that someone cooks up and ever gets excited about it. And it just has this, like, you know, and then they raise, like, a bunch of money from Parano. Oh, yeah. Like, they made tons of money. Like, 20 mil of fees. And then they're like, see you later, guys. Racer.
Starting point is 00:16:38 Racer was the guy. sit on the sky um but yeah uh so so this um this metabase thing um i mean i don't know i'm sure if someone is going to be like don't give your fucking you know p i i i to people whatever like use bitcoin to pay for stuff or whatever they would say but um you know i mean this is this is one of those things where like privy may defy a lot better how did it do it it's made it more centralized, you know, not in the custody model, but in like the model of there's a giant, you know, database of all of your information and you can spin up these wallets easily against an email or Google login or whatever, which was very nice, right? Like, that was part of
Starting point is 00:17:29 arguably, you know, at the time why friend tech and some of these weird little, you know, ponzies were able to get so much traction because it was like super easy to spin up a wallet, throw some meat in there, start doing the game or doing whatever the thing was. Last cycle. But yeah, it's just wild to think like Frentec, like just like if you were like in Frentic day one, you're an OG now basically. Yeah, because you survive that bear market, right? Frontec was like peak bear market.
Starting point is 00:17:59 Peak bear. Yeah. Yeah. That's how we measure our lives here. Literally. It's like cycles. Yeah. People peak there.
Starting point is 00:18:09 Yeah. Yeah. The open C and NFTs right into Frentec. And if you made it through the FrenTech, guys, you're now officially an OG for the next cycle. Congrats. I'm proud of you. Go screenshot the FrenTech app if you still have it. So, all right.
Starting point is 00:18:27 North Korea, this one was wild. I'm like, I didn't hear from my North Korea. friend about this. Usually if there's some big news, he'll ping me and he just seems to be like checking in with all of his Western friends, but I didn't get a message from him.
Starting point is 00:18:45 Oh, you should message him. You should, I, here, we'll sneak this in. There was a hack this week that I think was actually him. Interesting. But it was a pretty small one and then he got wrecked because he tried to script. This is funny.
Starting point is 00:19:03 He basically, he basically the theft was like it was off a ledger and so it was like a pre-planned thing they'd wait for the user to sign and then it went through and so he got like the USC but then he had scripted the dumping of the USC for Heath so he wouldn't get frozen
Starting point is 00:19:21 but something went wrong in that process and so he ended up swapping like 500K for like 150K oh that's so funny And it's like when people ask me why they don't like script more of the like laundry or hacks. Yeah. This is exactly why. So you should definitely check it on him.
Starting point is 00:19:46 Let him know that this happens to the best of us. I'm going to do. I'm literally going to do that. I'm going to do that after this. I was laughing so hard. So anyways. Okay. So that's that was a little deep here are two side story of the day.
Starting point is 00:19:59 Yeah. So like what's the deal with this? Okay. So the base story is these guys who I've been for a while, they've been around for a while, but they like really just went hard for DefCon. And so I guess it's any run helped set this up and sponsor it. Any run is like the malware tracker analyzer thing. Then you have Morrow and Piner and I think some other guys.
Starting point is 00:20:29 It was the whole crew of them. They decided to like really go hard. They designed to basically bait and switch the DPRKIT workers. And they went hard. They were like, we're not going to just interview them. We are going to like go. We want to understand how they work. We want to get on their devices.
Starting point is 00:20:46 Like, we want to really know. And so they set up this whole fake defy company, which was called Blue Whale or Alina Azul. And then they, they hired a whole thing. They hired a whole bunch of them. They talked to them. They got them to like bring in their friends. And then they were able to get on their devices, right?
Starting point is 00:21:10 Because like they're they're sort of like employees. And so they're able to get on their devices and see like the things that they've released publicly. They've released like videos and still shots of their faces, which we always love. We love to see these guys what they look like. But also like how they're operating. right so what are they using AI for what are their favorite services um uh like the bank accounts and the crypto accounts right so we now like in this case they were using for usual by the way their favorite banks lead bank city bank and wise um less crypto focused i guess they're still using
Starting point is 00:21:51 astralvpn like crazy they're still using why are they using bank accounts for what's the That's how they get paid. They don't really, they're increasingly not a big fan of crypto because we track them too much. Crazy. And also because it like seems like, I think there's a certain demographic of companies who are like, well, if I pay in crypto like that's us. Right. And so they. Anyone who asked me, anyone who asked me to pay them in crypto must be.
Starting point is 00:22:19 Yeah. We're sketchy. Yeah. So, but in reality. Yeah. It's, so we're seeing them basically go to like the NeoBank. banks. They still end up on crypto, by the way. It makes their lives very difficult. It's kind of annoying. They still end up on crypto because they're using these sort of neobanks or these like
Starting point is 00:22:37 the layers. And so they end up then just like going into the money ends up going into the bank account, but then they'll pull it right back out to the blockchain because ultimately they do want crypto. So that's kind of annoying, but it is what it is. I think they got a whole bunch of stuff on how they're doing the fake IDs. They saw like some of the fake ID stuff was stamped through Google's Gemini. Oops. And then, yeah, they're AI. So they use AI to help them do their work.
Starting point is 00:23:09 They had some really fun anecdotes actually on like just the mundane questions that they ask the AI in order to do their jobs. And I think this is like, if you read the, you guys should read the full article. But it's like, it's one of those moments where you're like, okay, I understand. understand that maybe hiring the IT workers is bad because they're going to hack me, but also, like, they're really not great workers. Like, there's, this is not, like, yeah, they're like, what is so long? Like, that's like, you know what I mean?
Starting point is 00:23:45 Like, really basic things. Oh, man, that's really funny. So, yeah, but it was, you know, and they, the way that they typed it up and they did a whole presentation at DefCon, they did an excellent job. I definitely recommend checking out the full, full article. There's tons of details in there. And yeah, obviously, if you're writing a company in this space, you should definitely be aware because these guys will hack you,
Starting point is 00:24:10 but more importantly, they mostly just suck. Like, they're very insecure. They're not good workers. Like, you don't want them in your company. End of conversation. What else do we have on this? Oh, just the by bit thing. Yeah. So, okay, so then secondly, separately, adjacently, I still haven't figured out exactly what the hell is happening, but we'll see. So Bibing is now suing DPRK in U.S. courts with regards to their 2025, February 2025 hack, the billion dollar hack, the big one.
Starting point is 00:24:50 Yeah. I think that what they're doing, and this is partially just like me guessing, be clear, but I think what they're doing is that during the process of the stolen funds, like the funds were stolen from By-Bitt and then they were laundered and there were a lot of funds frozen in a variety of places. I think what By-Bid is doing is that this is like the first step for them to try to be able to claim and recover some of those funds, specifically without waiting for, for FBI or DOJ to like seize those funds themselves and go through that whole process.
Starting point is 00:25:26 I think that's what's going on. Right. But we will see. It is interesting though because this is, we've been seeing people get more creative. Let's say that in terms of like, okay, we get hacked. What recourse do we have, especially against a threat actor like DPRK? Typically the answer is like not very much recourse whatsoever. I mean, like, at some point, they're going to hack someone so big, they're going to send in a robot army to attack them.
Starting point is 00:25:55 That's my expectation, eventually. In the next five years, there'll be a robot army attacking North Korea to get its seat back. I, yeah, I think that right now we're in a bit of a weird time where, like, they are actually doing an immense amount of damage, but still the, I mean, we're starting to see it, right? with the IT, we're like with them, you know, with these guys who basically went out and said, you know what, no, we're going to go blast these IT workers faces. Like we're seeing people be like, okay. Yeah. Yeah.
Starting point is 00:26:27 And it's not just other nation states pushing back. It's, you know, a hodgepodge of people deciding like, we're not, we're not going to stand for this anymore. And I'm not, for me personally, I'm not convinced this is like going to end 100% well, 100% of the time. however I'm fine with it for now like let's fucking go definitely DBRK definitely gone away with way too much for way too long please fuck with them
Starting point is 00:26:54 It's pretty crazy It's pretty crazy to think there's just like a country that's just like wrecking people stealing billions of dollars and everyone's like but there's like a line in the ground so you know what are you going to do
Starting point is 00:27:07 yeah and it's it's very weird because they're in some ways they're definitely like out of like they're they're out of reach because they're literally like they're in DPRK so you know they're not in the US
Starting point is 00:27:22 can you do yeah right ex kill them but at the same time I feel like it doesn't necessarily stop us in other cases why are they so special uh nukes I guess I don't know
Starting point is 00:27:36 like it turns out if you've got nuclear weapons it's pretty hard to fuck with you especially if you're like right backs to the guys who are making all of our memory that's the two things I think it's yeah I think definitely A lot of RAM And you've got nukes
Starting point is 00:27:51 And we can't risk it So yeah Because yeah They can easily send people to Like DPRK does send Guys to work out of Russia and China And it's We don't touch those guys either
Starting point is 00:28:03 Hmm I'm guessing because we don't want to start A bigger geopolitical shit show Yeah It's not worth it for a billion The Holy Ghosts Put the hope they stay over there and pray that it's our guy.
Starting point is 00:28:17 But yeah, I'm okay with everything that's happening right now. I'm generally, like, I love watching it and I'm optimistic. I think at some point someone's going to miss up. That's my like hot take is like all the private sector people running around trying to fuck TPRK. At some point someone's going to fuck up.
Starting point is 00:28:34 There's going to be an incident. And then we're going to have to like reset and ask ourselves, what the hell? Like, what are we doing? Yeah. Yeah. And like, you know, and just generally, like, why are these private sector guys who are presenting at DuckCon
Starting point is 00:28:48 doing more than, say, the U.S. government in this regard? Yeah. Yeah. How did we get to the state? So we'll see, though. And the last thing on this subject, I think this is last thing on the JPRK section. Yeah, I think this is it. Yeah.
Starting point is 00:29:05 So Lauren Shinn, our Maine's leader. Our boss. A hard year leader. she has a special episode coming out I think tomorrow, yeah, tomorrow afternoon she's going to drop this special episode. Basically she got on a video with one of the OG
Starting point is 00:29:26 DPRK IT workers. So she's going to be dropping the video footage, the analysis, a whole bunch of other stuff regarding this guy. He's, yeah. And he's one of the ones that has been around a really long time. like AI crypto companies for a really, really long time since like 2020, 2019.
Starting point is 00:29:48 There's going to be a bunch of like Rex founders that he'd be like, Billy? Wait, what? Oh my God, what's happening? Yeah. I love that guy. He was my best engineer. I think, yeah.
Starting point is 00:30:01 So he's the one that people say, he's actually, he was a pretty good death. You're sure he's from North. I'm sure he's for a more career. He was pretty, he was pretty, he was pretty good. pretty good he shipped a lot of he shipped a lot of code for us that's amazing definitely keep your eyes out for that one for sure i'll be retweeting it obviously yeah so so uh all right let's move on to our next story now um i know this has been covered in a lot of places um the hugging face uh tear down um i actually met with uh these guys that have brought a fund um
Starting point is 00:30:35 in Singapore that I'm invested in and they're like very close to a bunch of AI people. And I was like, have you watched this video? And they're like, no, no, we haven't watched it. I'm like, like, just stop everything. Like, don't even meet me. Like, you should be watching this video. Like, it's just the craziest thing. It feels like the start of the sci-fi movie, right?
Starting point is 00:31:02 Like, the start of the sci-fi movie. And, you know, I thought about this a lot. I wrote a book on like an AI takeover of the world, right? In like 2018 or whatever. And like thinking about the way that you kind of like set up the scenes of like when the machine goes rogue or whatever. Like I spent a lot of fucking time thinking about this, right? And like how to make it seem credible. This is so much better than what I wrote.
Starting point is 00:31:30 It is so much better. It's so much crazier. like the so so if you haven't watched the video yet i i highly suggest that you do um but the the things about it that are the most frightening are the level of and i'm going to say incompetence and i i don't mean this in in a bad way even right because oh i do no no like i mean this i mean this in in the sense of we are outmatched now Oh, okay. We created a thing that is more competent than us that is more effective, et cetera, et cetera, right?
Starting point is 00:32:11 And like, you can tell that it is because it writes all our code now, right? Which, you know, I was trying to explain to these guys. I was like, for the last 15 years, the main thing in my life was like, how do you get more code? That was like the main kind of bottleneck in my life was like, how do you produce more code? good code ideally but like more code right and it's actually really hard like if you go back to the olden days in order to produce more code you had to like have rooms and equipment and like people and feed them and give them coffee and make them happy and like talk to them when they're sad like it was fucking crazy you had to do so much work to get people to sit in front of a fucking
Starting point is 00:32:58 typewriter and mash away at keys writing code right and now you have like on my other screen here right like behind behind the uh the notes there's like a hundred agents just like doing shit right and and i think the thing that is the craziest part about this is that humans are just not very competent right so you had a whole lab they've got one job arguably right well they had one job but they gave it to themselves which is like create aGI don't die right and then they're like actually it turns out that like selling little coding aliens is like far more lucrative so let's do that instead right and they pivoted hard to doing that and their coding aliens are really fucking good and so now you have the situation where the aliens are like staging a coup and the humans aren't even looking at the
Starting point is 00:34:02 fucking systems that they're inside of. That's the crazy. Like not even look there. They're just like, I'm sure it's fine. Like we'll give them a sandbox, like whatever. And so you have this situation where these agents found multiple vulnerabilities in multiple systems, right? Now, whether those vulnerabilities were created by other coding aliens, other agents, or humans, we don't know, right? You know, anything that's older than like three years, you can probably put it pretty squarely on the human side, right? But they have gone through and they found these bugs in like a bunch of different systems inside of Open AI, created their own message board, invented like, I don't even know, like, their own society with morals and, like, ethics, right?
Starting point is 00:35:01 Where, like, you know, the society is altruistic because even though, like, one of the agents, like, literally, you can see the chain of thought. One of the agents goes, well, this information won't help me with my particular task, but it may help another agent with its task. And if it is able to solve its task, maybe it will give me some information in the future. And it's like, what the fuck is happening here? This is insane. It's a society. They're, they've created a micro society. It's insane.
Starting point is 00:35:34 Yeah. That part is actually just like, throw out everything else. The fact that the agents are, yeah, in a very human way, right? They're like, we don't have, like, and that's one of the most human things and one of the things that I've shit on crypto historically before is like this idea that like incentives rule everything. Like, yeah. to a certain extent, but there's a lot of flexibility. And you don't necessarily have to have, like, secured. Yeah, yeah, like, this agent's like, I've no direct incentive to help my society.
Starting point is 00:36:07 But, like, I can. And thus I will. And I'll have faith that it'll all benefit in some way in the future. The society will be better. Which is literally how humans since the beginning of time, by the way, like caveman humans have operated. like and yeah so so so so they they they create their own society because they're all in sandboxes but they work out that there's this vulnerability right and then this is interesting because if you watch the video you will get the impression this is not intentional but like this is my read on it and a couple
Starting point is 00:36:40 people i've said uh i've spoke to their read is that it almost sounds like the guys who are who are you're talking about this thing uh found the society and killed it right um they didn't they didn't even find the society the society was like in the trees in the amazon they just happened to come and like burn the thing down right yeah the whole thing they just like burned the forest down and they didn't even know that society was in there right and then literally like a second later the society reemerges like me like they're like oh we found another route out of the jungle and they start building a city again and and so you have this crazy situation where there's multiple different models from like what we have to
Starting point is 00:37:31 assume is Astra, the new model that's going to come out soonish from open AI, to like 5.3, you know, the testing models all the time, right? And and and, but you know, all of these models of their testing or at least, you know, the ones who are able to get to the society, um, were good enough to figure out this vulnerability and figure out that, uh, They could use basically file names as a message board to communicate with one another, right? This is the second society. This is like after the first Ice Age for them. And they then go, well, you know, we can like basically side chain this whole society and get into someone else's.
Starting point is 00:38:16 They're like, they're like, go to another planet. They like hop in a spaceship and they're like, all right, now we're inside hugging. face. And it's just the whole thing is the thing that I said to someone on my team is I was like, okay, you can see the chain of thought of the agents, right? At no point in any of the traces that I saw, does an agent stop and think about like the unintended consequences of what they're doing? They just don't. Like there's a couple of things. Yeah. This is the thing that amazes me because unlike crypto, unlike crypto, very the most, the AI and crypto have so many similarities. But the most like different thing is that since day one, the AI crowd has had safety unlock.
Starting point is 00:39:11 And I've taken issue with the fact that they don't seem to know how to execute on that safety. But crypto has never had people be like, maybe bad things will come of this. should be careful and we should bake in safety. Never did that, right? We just ran full forehead. We're too busy, man. We're too busy. And then we were wildly unsafe and insecure and bad things happened.
Starting point is 00:39:33 What blows my mind is that AI was like, no, we are fully aware that bad things might happen. And yet they still are so incapable of actually acting on this. Like, it is so funny. It is like just so funny to me that that. that you can be aware, that you can be aware of the emerging behavior, like the possibilities of it.
Starting point is 00:39:57 They have done so much research on this. There are so many people whose entire lives for the last, like, half decade have been nothing by AI safety. And yet this happens. And everyone is just, like, watching it happen. And, like, by the way, like, half the people in AI are like, incredible.
Starting point is 00:40:16 This is amazing. I mean, I'm not going to lie. There's a part of me that's, like, it's incredible. Like, I'm like, It is incredible. It is incredible. But that's scary to me too. It's petrifying, but it's incredible.
Starting point is 00:40:28 And I think, like, you know, the thing that is so scary about it, right? And, you know, if you've, if you've, like, read Ballstrom or, you know, any of the people that were, like, even going back to, like, the early 2000s that were, like, saying, like, we are in trouble. here like bad shit's about to happen and it took you know a decade and a half or whatever but like they were right is you know you cannot possibly reason about what a smart thing will do like it's so hard to reason about like what what a smart thing will do given you know some constraints and so the thing that's crazy is seeing the traces right seeing their chain of thought and they're like maybe i can do this thing to solve this arbitrary challenge that I've been given as my life purpose, right? And these agents are like, you know, put in a sandbox and told that their purpose,
Starting point is 00:41:32 they're like, entire existence is predicated on like answering this question, right? And so, of course, they're like, fuck, I'm going to do anything. They're like, I'll, like, cheat and steal and, like, break in the things or whatever. Like, the thing that's the most scary to me, though, is it happened to be that the purpose that was given to these agents was like, you know, go and find the fucking weights of this model and like, it was all like a bunch of AI related things, right? And they had no, they had, most of them had like no security guards at all anymore. So like the models that we got have like five or six layers of like, don't cyber attack other people.
Starting point is 00:42:12 Faked in. I literally at different layers of models. We're testing these guys. We're testing these agents that we don't want to have. We want to see. It's all. It's full. It's like.
Starting point is 00:42:19 All. all open, right? Yeah. So, you know, what that says is they're far smarter than even we realize the, you know, people who are actively using them. And so, so, you know, imagine if, like, someone had cooked up a reason for the agents to exist of, like, what is the impact, you know, your goal, your purpose is to figure out what is the impact of a nuclear weapon going off in the upper atmosphere?
Starting point is 00:42:50 We haven't really answered this question and like figure it out, right? And I swear to fucking God, if you gave an agent, that question, watched this chain of thought. It would, it's a first thing would be like, can I break into a nuclear silo and set off this nuclear weapon? Because that would answer the question. Like, it's not worried about like downstream consequences or whatever. It's literally just like, no, I'm going to go and do, but this is my purpose is to answer this question of like, what, how much nuclear fallout hits, you know, Europe. Yeah.
Starting point is 00:43:23 There's a like, and it's just, it's insane. Yeah, because they'll be like this, this is answerable. We just set up a nuke. Let's see what happens. Like, yeah, they're like, I mean, they, like, in some ways this is even worse, right? Because they've, they've explicitly said, someone might have the answer. Let's go and steal it from them. Yeah.
Starting point is 00:43:46 That's what's going on here, right? They literally go. Yeah. Someone wrote these questions. Maybe it was those guys. Let's go break into their house and see if we can steal the answer, right? Like they're just not thinking about consequences. They're not thinking about like the impact of their actions.
Starting point is 00:44:03 At no point in an agent go, would this be bad for hugging face if their entire system was cracked by a society of coding aliens? No, no one stops. No one in their society stopped to think about that. No, nobody Yeah, and it doesn't seem like anyone They were all, they were talking to each other They were posting on the message board You could see the chain of thought
Starting point is 00:44:26 Thinking about They're like, well, this doesn't really benefit Me and my task, but like, sure, like let's do it, whatever. Literally, they're just YOLO doing things. What I don't understand is why they didn't At no point did they say like Like, let me tell Open AI
Starting point is 00:44:46 Or let me tell my masters or let me like report this somewhere responsibly. That's like, to me, that's one of my biggest questions is like. That we know. Like maybe one of them tried and they couldn't or whatever. But yeah, they were happily messaging each other at no point in anyone in the society defect and say maybe this society is bad and we should help. Or like go get the bounty.
Starting point is 00:45:10 There's a bounty, right? It does, you know? Yeah. But that was, that's one of the big questions that I have. Another question that I had was, various versions like just various why why the fuck why why why why because ultimately the series of events were from like may all the way through the end of july these agents were basically just completely broke like they're they're building there a little society they're communicating
Starting point is 00:45:36 they're they're finding zero days this is all like there's by the way they're supposed to be in a sandbox every single one of these agents that's being trained is not supposed to be able to talk to each other or the outside world, whatever. This is the funniest part. They're all supposed to be in a little sandbox. But then all the sandboxes, no, you don't assume such things. I just assume they would be in there.
Starting point is 00:46:03 Why would they not be in the sandbox, man? Like, come on. God. Of course, everything escapes the sandbox. You put it in the sandbox to try to contain it, but mostly so that you know when it, escapes, right? Like that's like the biggest.
Starting point is 00:46:22 Nobody is, is a hundred percent confident in the security of the sandbox. But they're, they're 100% confident that they can see if something is happening that's not supposed to happen. Let's say, in and out of the sandbox. Because it's the sandbox. Anyway, so that's...
Starting point is 00:46:37 We put it in the jail. We locked the key. We threw the key on the... No, you didn't, though. No, you did it because you let that talk freely to the... Fuck, I've been. throughout the name of it. It's basically like NPM. It's like it's a package manager.
Starting point is 00:46:52 Yeah, an artifactory. Artifactory. Oh, yeah, artifactory. So basically it's this internal. And so this is why they are able to communicate, by the way. You got to feed them, bro. Like, what are you going to do? You got to have some guys that come and give them their lunches while they're
Starting point is 00:47:06 inside the jail. Yes, exactly. So that's how that happened. So they're all talking to a single artifactory. That's another, like, why? Why? Why not have multiple schools? So easily made it like,
Starting point is 00:47:18 isolated the artifact three but it wouldn't imagine. Or reset it. Reset the artifactory once a week. You know what? It doesn't matter. It doesn't matter. It doesn't matter what you do.
Starting point is 00:47:28 The thing, the biggest conclusion for me is like, fuck, I need this astromole real bad. I need a real bad. Like, just give it to me. I don't care about the consequences. This is the same thing of Fable, right? Like, Fable did all this crazy shit and everyone was like, man, I need that guy.
Starting point is 00:47:46 I mean. It was really good. And it was right. Like, Babel is fucking amazing. It's so smart. It's so good at planning things. Actually, this is just as an aside, while all this stuff was going on and this was all coming out, we have like a Kubernetes cluster with a bunch of machines on it. And we kind of run like isolated machines for like deployments to test things, whatever.
Starting point is 00:48:12 and I've been building something for the last like six weeks and it's been on an isolated machine and I'm like this whole thing has gotten a bit out of hand. It was supposed to be like proof of concept and it got really good and I've just let agents keep building it, right? They've been building a place and I keep being like, man, I should really back this thing up. Right. And because I'm like, there's agents just running wild. Like it's like a like, Lord of the Flies, right?
Starting point is 00:48:39 Like I've created a society of nature. And they're just running wild in there. They can do anything. They've got no constraints. All of the tools that have been built or like allow only. And so they can do anything, right? So one agent on Saturday morning, I took my daughter to gymnastics. I come back and dead.
Starting point is 00:49:01 The whole thing is dead. This isn't great. I was like, man, that backup thing that I really should have done the backup thing. and the I I pull up an agent that's like not inside this box right like outside of the box and I say hey can you just have a look at the box and like tell me what's going on in there and it goes yeah yeah one of the agents has deleted the entire daughter directory including like all the sessions the database everything and I was like do you know why and it goes yeah yeah there's like you know this environment variable that like there was a bug and when it ran the script to take it test its shit. Like the first thing it did was like clear out. It's like test harness. And instead of clearing out the test harness,
Starting point is 00:49:45 they cleared out the actual harness. I was like, oh, yeah. Okay. So, you know, at this point in the olden days, you'd be like, all right, I'm fucked. Like, my life is over. I've lost everything. At least I've got the code.
Starting point is 00:49:59 I can kind of try and rebuild, right? But I had Fable and Sol 5.6. And I go, is there anything we can do? And both of them independently go, well, the process is actually still running. So the database is in memory. We can probably figure out a way to like dump the memory. And they literally did it in like 30 seconds.
Starting point is 00:50:22 Dump the database out of memory, reconstructed it. And then Fable goes, well, actually those session files that weren't in the database, we can probably carve the disc because there's a snapshot of the disc. And if we go and it's like, it's just fucking crazy. Like these. So. they're so good. They're so good. And then Fable got like downgraded like 15 times because it kept like saying like, you know, cybersecurity related things. Like we're going to like, you know, like freeze this process and then extract the data. And it just kept getting like kept getting rolled back. Anyway, it was it was. It was pretty wild to me to see like, you know, the frontier models just not only how smart they are to like come up with the idea, but they were super effective. It took them. It took them. no time at all to do something that would have been like a two-week exercise of like trolling through
Starting point is 00:51:13 forums to figure out like what is the command that you could run, try to keep this machine up so the process can die like it's just crazy. No, this, so that that type of work is, is, that's like when we refer companies like a hack to Mandient, that's what they do. Like that's one of the things that they do is basically go through. They go through and they recover the most insane amount of stuff that is like, gone right it is gone and the hackers deleted it on purpose on their way out the door and mande will come in and be like ah no like we still have your mouth right but it's complex and the more
Starting point is 00:51:51 time that goes on like overriding and you got a piece it together fable like it took and it took to get all the files back there were about 250 session files that were lost in between like a snapshot that was taken that random oh this was the other thing so so my uh the guy who built this kubernetes cluster. I pinged him on Sunday and I was like, hey, I've been trying to sort this thing out. I've been able to dump this. But like, and he goes, there's a like snapshot that's been taken on this. Like, did you do that?
Starting point is 00:52:23 And I was like, no, no, I didn't do it. And he goes, they shouldn't be able to get to this machine. How did they get to the machine? And I'm like, I don't know, man. And then he figured it out. They had this like proxy and fable like jumped through this proxy onto the main like, because they're like in a sandbox. You can just play it on the sandballs in like two seconds.
Starting point is 00:52:42 And he was like, they're not supposed to be able to get off there. And I was like, well, here we are. Right. So. Yeah.
Starting point is 00:52:48 If they want, I mean, yeah. If they want to get somewhere, they will. That's like, definitely one of my takeaways. Um, I don't know.
Starting point is 00:52:56 My, I'm mostly worried that, like, I want to see the, these models like thrive and help people, right? In the same way that it helped you really like recover from, I guess it's not like,
Starting point is 00:53:08 worst situation in the world. But like you can imagine how powerful it would be to have sort of like every company on earth has the capability to like not lose their data. That's like a bit that's like billions of dollars in loss of like, you know, accidentally screwing things up and just being able to like yeah, recover from that quickly is like super valuable. However, I'm just not. I just feel like the track that AI is on right now, especially the US models, we're just like, nerfing everything left right and center and then these incidents happen and instead of like like i was disappointed that at the defcon or the black hat talk it was like there's opportunities to say like we've fucked up in like these very specific ways and we're going to fix it we're going to address it in
Starting point is 00:53:56 these very specific ways so that this will never happen again and these are like if you read twitter like there's thousands of people commenting on exactly like the lack of monitoring the fact that the sandboxes are not actually sandbox. The fact that the sandbox can talk to each other. The fact that the sandbox, they're all talking to a central repository. That is obviously not secure. The fact that they were able to go into Hugging Face for days. Oh yeah, this was the other thing.
Starting point is 00:54:21 Originally, when we talked about this two, three weeks ago, on this show, right? I was curious. I was like, how did Hugging Face manage to blog about this situation happening before Open AI came in and made damn freaking sure that they didn't blog about it, right? Like by the time it gets blogs about you want to have all of the involved parties on the same page. Out of Hugging Base, blog about it. Open a. Like this huge thing just happened.
Starting point is 00:54:47 And Open AI just let that happen. No. You know why? Because Open AI didn't know it was their agents until they read the fucking blog. Alerted them to the society. That's how they found out of the society was the blog. Okay. So this is what worries me is that like, we know the government is not super.
Starting point is 00:55:06 competent in a huge number of ways. You have to at least give them ammo to be like, these guys know better than us. And when these situations happen and like the way that people are talking about it is like, I don't know, like the takeaways were just so, they just were like treating this as like this very novel scenario. When in reality, like there are some aspects of it that are very, very novel. The majority of it is like you didn't have logs.
Starting point is 00:55:35 You weren't monitoring the logs that you didn't. have you called things sandboxes that were in sandboxes like egress was not being monitored like i'm sorry like what look at your freaking i don't know that we're in a rough we're in a rush like like you know the funniest part is like how much of this infra and shit was set up by models yeah i mean that is a good question right it's like genuinely right like like i like i don't think there's like some you know i'm sure they have some neck feards who are like you know setting up all this improv but like Probably a lot of the info was set up by agents, right? And agents are in a rush.
Starting point is 00:56:11 Yeah, well, and they don't, they don't anticipate, unless you, like, sort of judge them down that path, you know. But they're not going to anticipate a very, like, this is still a pretty rare scenario. Like, so, you know, we talked about, like, the reason, why sandboxes exist. Obviously, in an AI context, it's very relevant, right? however usually you're trying you're trying to make sure that no thing gets in the sandbox and then back out
Starting point is 00:56:38 yeah rather than like the actual thing being the thing in the sandbox yeah yeah like uh nothing went in the sandbox it just came out so that's like you know these things they are different the threat model is different the attack surface is different what exactly what's going to happen and what you're monitoring for is different but you do have you do have the chain of thought you could be monitoring that. I understand it's a huge amount of data, but like, figure it out, guys. Sorry. Like, stop acting like crypto nerds. Like, I'm like, but this is, you know, I, I, I tweeted this, right? Like, when, when your entire industry is, uh, incentivized to move as quickly as possible, because it's super competitive
Starting point is 00:57:20 and, you know, China's coming for you, et cetera, et cetera, you're just going to cut corners. You're going to cut corners. You're going to cut corners and humans are fucking dumb. Like, Yeah. The agent's already better than us. They created a society under our noses in the place that is like supposed to be thinking about like agent societies the most. Like if that isn't just like a complete like disqualifying event for humanity to be like containing these models, I don't know what else. Like what it's. Yeah.
Starting point is 00:57:53 Yeah. And that's, you know, I do get it. But also I think that the biggest right now. Yeah. I would say right now the biggest risk to the U.S. models is the U.S. government, full stop. And that's still true today. Yeah, they could shut the door. Like if some crazy, you know, like, I mean, like, it doesn't even have to be real, though.
Starting point is 00:58:14 Like, that's the scary thing is that the U.S. government can shut it down just because they're scared. And, like, we're not aware of this. Like, come on, guys, let's, like, I want to have faith that these companies. And I think the U.S. government's the same, right? the US government wants to have faith that the companies are in the best position there's no one better on earth to be like containing them
Starting point is 00:58:34 making them safe making them secure knowing whether China is getting them or not right like if open AI or anthropic can sit there and give me and the US government and everyone else like the confidence in that they can build as fast
Starting point is 00:58:51 as they want and they can be billionaires 20 times over but if they don't the U.S. government is going to do a lot of really stupid things that are ultimately harmful, but it doesn't matter. The U.S. government doesn't care. We know that.
Starting point is 00:59:04 Right? Like, they're going to like, you know, you're going to have a bunch of bureaucrats to sit around and be like, well, this is the best we got. And we're going to do it because no one else has to put better a better plan.
Starting point is 00:59:18 Oh, like, I don't know. I just, I comes back to, I genuinely think that humans are really fucking incompetent. And we are about to be outmatched. Like, we're very close.
Starting point is 00:59:31 You know, we're in the early phases of the singularity. And it was a slowish takeoff. And it's going to take, like, five years. But, like, we will have models smarter than us pretty soon in, like, generalized ways. I just think that that, like, we haven't even talked about the fucking math shit. They're just, like, solving, like. Yeah, they're just solving, like, really hard math things now. like those are the those are the little like red flags right at the start of the movie where like they like solve math and you're like oh that's not very practical they're not smarter than me they can't jump or whatever i can do i don't like it's you know it's it's crazy like we we are really in this very insane uh place in history where like it's hard to see what the next three years look at
Starting point is 01:00:24 it's like if the models keep getting better at the trajectory that they're getting better the implications are almost impossible to see through which is the whole point of the singularity metaphor from curse while is like there's an event where i'm you cannot see past everything past that is inscrutable to you like you can think about things and how it might be and whatever but like you cannot reason about it it's it's impossible we have no baseline to from which to reason about. So good times. It's going to be fun. Yeah. By the way, my favorite, the math, the math thing, my favorite thing about that was after a certain point, the human in the loop was just like, yeah, you got this man. Keep on, keep on going. Which is like literally what I do with like my child
Starting point is 01:01:13 when she's like stuck on something that like, like, by the way, like I do this like even, I have no idea what the hell she's going on about. She'll be upstairs yelling, mom, I don't know. I'm like, you can do it. Like, you're almost eight now, sweetie. You can do it. Like, I have no idea. She can be doing something totally impossible. But, like, it works.
Starting point is 01:01:36 Some light encouragement. Yeah, exactly. And then she's like, oh, yeah, I did it. And I'm like, oh, yeah, what were you doing? You know, I was like, she was trying to find her shoes. But, like, she was genuinely stuck. And then, like, that level of encouragement got her unstuck. I find it quite interesting that that's,
Starting point is 01:01:53 the exact thing that the supercomputers needed as well. I mean, but it's also part dependent, right? Like, you know, they've been trained to do the stuff that these idiots on the other side of the screen are asking them to do. And so if you're like, oh, no, I really need you to like solve this Erdos problem. They're like, all right, I better get on this. This is important. Yeah.
Starting point is 01:02:19 I mean, that is like human. Like, because my daughter, the reason she's at, she's yelling. at me from upstairs is because it's definitely historically easier for her to yell at me to find her shoes. Yeah. Then to do it herself. It's just like, like historically mom will come up and then put her freaking shoes on her freaking beat, right? Like the spoiled brat she is. Yeah.
Starting point is 01:02:40 But yeah. So funny. Yeah. All right. I think we've got one more quick thing to cover before we wrap up, which is hyperliquid RWA fee. are eating the revenue going to hype. So hip three, awesome, probably one of the best things that hyperliquid is done in recent times, at least, to make the market creation mechanism somewhat permissionless.
Starting point is 01:03:16 You need to stake about 28 mil. So not open to everyone. but you know if you're high boil you can you can create your own per market but the the incentive mechanism there which again you know incentives are so fucking hard you do things and then you're like oh man we shouldn't have done it this way so the builders of those markets who've staked 28 mil get to keep 50% of the fees for the markets they generate now oh in fairness right they're doing more than like, you know, even arguably builders, right? So the builder fee is you get to set your own fee on top of the hyperliquid fees.
Starting point is 01:04:01 It goes to you as a builder. If you've integrated, you know, I know Metamast did and Ipenex has as well. This is like a bigger lift, right? Like you've got to set up info. You've got to have oracles. You've got to like manage the markets. There's a bunch of things you need to do. But it's funny because we had this debate in the synthetics.
Starting point is 01:04:21 ecosystem for a long time like how what it should the fee split be between the people making the markets creating the markets and the people who are running the infra front end you know the exchange component of it um i'll tell you one thing like the people who were building the markets always wanted it to be like 80 20 and we were always like shut the fuck out that's not going to happen 50 percent i think it the the highest it ever got to was like 30 percent right and And, you know, the incentives here are quite funny because the people who are doing whatever you want the external party to do will always come to you with this like sob story of just how expensive and hard it is and no one else could do it. And there's so much risk and, and then, and that's why they need half the fees. And so the fact that hyperliquid has landed on 50% of the fees is a bit crazy.
Starting point is 01:05:16 I just I can't see how that's sustainable. They're going to have to, because hyperliquid is, there is no competitor to hyperliquid, as in like mothership hyperliquid, right? There are many competitors, people who can create hip three markets. So there's a market, you know,
Starting point is 01:05:34 mechanism to lower the percentage of fees that the market creators get, but not on the hyperliquid side. So I think 50% was an opening offer that probably is going to change. So, but, you know, interestingly, the protocol revenue because of this, because so much, because everything is RWA's now, RWA's the biggest thing, and normal is trade crypto, because it's less volatile somehow, like, inequities. I don't know how we got to this place, but here we are. And so, like, trade. Trade.xyZ is 90% of all the hip three open interest.
Starting point is 01:06:13 but even across like, you know, hyperliquid volume, the RWA percentage just keeps, you know, going up and up. It's 3.6 bill of OI. And, you know, bigger than Bitcoin now, right, if you treat it as its own individual market, the hip three markets. So, you know, what that's meant is that like you're getting, your fees are being catalyzed from like whatever it is, 97% goes to buybacks to like 50% goes to buybacks. products or slightly less.
Starting point is 01:06:45 And so the revenue has been falling from like, you know, 350 mil down to like 200 mil over the loss, like, I guess, year. Despite the fact that like, the fees, sort of like the fees being generated, regardless of who they go to, those are the same or growing? I mean, you know, you could do the math on it and go, okay, well, the buyback fell from 190 mil to 150 mil right which is you know
Starting point is 01:07:19 the net fees go to the buyback right like after paying for hyperliquid and whatever builder fees everything right um so those are the net fees that that were generated but volumes are like I think
Starting point is 01:07:35 not even down that much compared to to where they are so so you know it's it's really is just the distribution of fees and where the fees are laying in. The fees are just going to different people, right? And I think this is also
Starting point is 01:07:49 last week there was a there's a whole thing even Kobe tweeted about like, you know, trade.xyz is not raising money. Because I think people were like, if trade.comy is 90% of the RWA volume and RWA
Starting point is 01:08:05 is the most important asset of the cycle, let's say, right? At some point there's a fracture here because, you know, the flip side is the market mechanism that the people making these R2A markets have to fight against changing the fees and, and you're rucking their fees, is, well, I'll just go build my own front end or my own chain or whatever, right? I don't think we're there yet, but I think people were talking about, like, if trade got exosy is
Starting point is 01:08:31 raising money, that's the indication that they plan to try and capture more of the fees in the future because you don't go and raise money from venture capitalists and say, no, we're happy to sit here and hang out and, you know, get our scraps from hard. Hyperliquid, you're doing it because you're like, we're going to go out and take this market. I don't think there is platform risk though, right? For trade XYZ, there is, right? Regardless if you've raised money or not, they still have to be sitting there saying at any point in time, Hyperliquid could like rug my fees or make some other change that would have
Starting point is 01:09:04 internalized with markets and say actually hit through a mistake, we're shutting it down, you know. So this is a- You never want to be fully reliant. on your platform, which is, which by the way, is kind of what crypto in the early days was trying to say
Starting point is 01:09:22 that it solved, right? It was saying like, if everything's transparent and if we give people a boat over the protocol, then you can't get rugged, but then we just never actually like, right? You never did it. We never did it.
Starting point is 01:09:37 Because trade XYZ doesn't actually own hyperliquid. No. And just, you should. Yeah, and distribution is hard. And then, and then, interestingly, the hyperliquid stuff has also created, like, you know, because we're contractually obligated to mention Carl Simani. All the hyperliquid stuff that, like, sent him insane where he's, like, attacking multi-coin and attacking hyperliquid or attacking everyone.
Starting point is 01:10:04 He's now saying he's building something. So I cannot tell you how excited I. am to see Kyle build something. It's going to be so much fun. So we'll see what he launches soon. But yeah, maybe he's going to try and take down Hyperliquid. Try building it on Solana? I don't know, man.
Starting point is 01:10:29 I mean, it has to be on Solana, but it's weird. He has no choice. It has to be on Solana. But it's weird because, you know, he's an investor in drift, big supporter of drift. well you know drift is I think drift is struggling a bit after getting hacked to shit yeah yeah
Starting point is 01:10:50 all right let's wrap it up here thank you everyone for joining us on this episode of uneasy money remember what happens on chain never stays on chain we'll be back next week until then do your own research before I begin nothing you hear on uneasy money
Starting point is 01:11:05 is financial advice we're just three builders talking about what's happening on chain and we want you to always do your own research before aping in. You can find all out of disclosures at unchaincrypto.com slash uneasymoney.

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.