On The Brink with Castle Island - Weekly Roundup 08/07/26 (Coldcard hack continues, Ethereum mulls an issuance tweak, ai16z winds down, Clarity deadline looms) (EP.733)

Episode Date: August 7, 2026

Matt and Nic are back with more news and deals. In this episode:  Why the Coldcard hack is worse than some major exchange and DeFi hacks What the Coldcard hack told us about the need for open weight models Is self custody over? MSTR sells more Bitcoin An FBI agent steals $1m in crypto NY state sues Kalshi Ethereum considers another change to issuance Ethereum needs to resist the urge to change monetary policy ai16z's Eliza winds down the merit of reading sci fi The cofounder of Sui leaves for Anthropic Will Anthropic keep the discoveries to themselves? The WSJ gets it wrong on Clarity

Matt Walsh and Nick Carter are partners at Castle Island Ventures. All Vees expressed by them or the guests on this podcast are solely their opinions and do not reflect the opinions of Castle Island Ventures. Guests and host may maintain positions in the assets discussed in this podcast. You should not treat any opinion expressed by anyone on this podcast as a specific inducement to make a particular investment or follow a particular strategy, but only is an expression of their personal opinion. This podcast is for informational purposes only. Brought down by bad mortgage investments, Lehman, which has 25,000 employees will be liquidated.

The federal government loans American International Group, AI, IG 85 billion dollars. This is a different kind of market and the Fed is asleep. The federal government is stepping it to stabilize Fannie Mae and Freddie Mac, the two mortgage giants that have been threatened by the housing crisis. The Bank of England has pumped 75 billion pounds more into Britain's ailing economy with a new round of quantitative easing. You print a couple trillion dollars and all of a sudden people start to worry. So out of this worry, we have something called the Bitcoin. Welcome to On the Brink. I'm Matt Walsh.

And I'm Nick Carter. And this cold card thing is just so depressing, huh? this hardware wallet hack, it continues. It's been one of the most depressing weeks in Bitcoin ever. The quantity of the hack is nowhere near the top 10 all-time crypto hacks, but it's really these stories, these people that had like half a Bitcoin, two Bitcoin, that just lost it all. And they were doing what they thought was the right thing.

Yeah, I think this one is worse in many ways that, like, FTX was how much? Several billion at aggregate, right? Yeah, what was it? $10 billion, $11 billion? Yeah. But a lot of that was funny money, you know, and I'm not saying like, oh, alt coins are funny money. A lot of it was paper gains in alt coins that people had been trading, right? Yeah.

People weren't, some people probably used FTCS as like their ordinary custodian and brokerage, but a lot of it was just traders that had made a lot of money. And then in 2021, right? Here with cold card, this was people's savings because it was it was exactly the device that you would put in the bank folder. in your basement or whatever and put your savings on and leave it there untouched for years. Yeah. That was the whole point of it because it was so annoying to use. Well, it was a terrible user emphasis.

That's why I stopped using it. Yeah, same. I mean, I feel lucky that I wasn't affected and that was because I hated using it. Yeah. So these were regular people too that would put their retirement fund on this thing and just stack stats and just buy Bitcoin for years and years and put it on this thing. and it was the people that quote unquote did everything right ultra cold storage never touched the internet no exposure is just that it was the whole thing was undone because the randomness was bad and so i think

the hundred million dollars plus that's been lost is much more damaging in some ways than a billion dollars in a defy hack yeah it's it's more damaging and just the you don't really have to squint too hard to put your conspiracy theory hat on here um it should be the most basic thing that you do as a manufacturer of a hardware wallet is to figure out the entropy. I don't understand how that's not something that was addressed. And it looks like it was, you know, it was pointing to what was it, some GitHub repo that had maybe like one or two contributors on it. And that was the source code for the entropy. It's hard to imagine being that in that. Yeah, it is really, really surprising and shocking because it's really, it's really,

you just have one job. Yeah. As a hardware manufacturer. And if you go and look at Ledger and Trezs, well, Ledger, I guess is closed source, but you look at Trezor, they have so many redundant sources of entropy that all feed into each other

because it's the most important thing. Right. So they're super, super careful not to just rely on one source. But entropy is tricky because it's really hard to test it, you know? It's hard to know that it's creating bad entropy. There's no way to visually look at a public key and think to yourself, well, that's low entropy.

Yeah. Right. The other thing that makes this really tough is that they were promotional on so many of these Bitcoin-only podcasts, and they gave away a lot of these hardware wallets. I mean, this could be a mastermind hack. Yeah. I mean, people have some conspiracies. Maybe it's an inside job, et cetera, et cetera.

We don't know, of course. Who knows? but it's super, super damaging and it hit right at the heart of the sort of most purest element of the Bitcoin community. And this is not to gravedance at all, but I think part of the reason Cold Card got so big was because even despite the fact they're a tiny company was because they specifically position themselves as for the maximalists. We are the Bitcoin only guys.

And unlike Trezer and Ledger, we're not tainted. we don't do altcoins. And so I think that caused them to have this overly large market share among Bitcoiners that was outsized relative to their kind of true kind of where they should have been, you know, given that they were such a small company. And that undid so many people in the end. And it's not really their fault. But I think they were really amplified by the Bitcoin podcast complex because of that.

You have to give Alex Thorne at Galaxy and Rob Hamilton at Anchor Watch a lot of credit for what they're doing here, helping victims just map out where the funds are, get the reports filed with the authorities. So there's some people that are being very helpful right now in the community. That's absolutely right. One of the worst things, too, is that it's not just one actor doing it now because once people realize that, you know, the entry was bad, there's varying levels of security, I guess. So there's the people that just used the basic built-in entropy. Then there's people that used, you know, a 25th word passphrase. And so there were slightly harder to crack wallets. So now a new set of attackers, an additional set got to work cracking the slightly harder to crack wallets.

It made me the, you know, base level difficulty was 40 bits and maybe went up to 50 bits. That's still getable. And so the withdrawals have actually continued. over the last week and the total keeps climbing higher and higher. What do you make of the, um, just the LLM side of this story, both on the discovering the vulnerability and then protecting against the vulnerability? I think it's a huge story. It's a huge story that hasn't been covered for some reason in the mainstream that this hasn't been covered. I'm really surprised. The, um,

so the, I think the hack happened because Kimmy came out if I had to guess. You don't think it was, there are some reports that Claude can find this hack. Oh, so maybe it could have been Claude, but, you know, either way, the frontier progressed. And so the attackers had a big advantage that could canvas all of these repos and find the hack. The defenders have to use Kimmy. They can't use Frontier models. So the good guys, like Rob Hamilton's going around and scanning all of these repos, the good guys can't,

can't use frontier American models because they refuse to do anything that has to do with cybersecurity for safety reasons. It's deeply ironic. People might not actually realize that, but you get downgraded when you're using these frontier models if you're perceived to be using it for hacking related activities. So if you're going at it from a defensive perspective, you'll actually get downgraded, which is just a perverse thing. Yeah, and I think it actually is a very, very important thing to realize. The whole safety paradigm in Silicon Valley, the AI safety paradigm, rests on this kind of impossible task, which is to discern the intent of the user.

Right. So like what the labs are trying to do is they want to look at a query and determine is this a white hat or a black hat. But in some cases, it's the same query and the only way to know if someone is good or bad is their intent in their brain. So it's not inferrable from the query. So instead they default to just walling off whole categories of usage. Because there's no way to tell anthropic, hey, I'm actually a white hat and I'm trying to save people money rather than steal money. So the safety paradigm causes this enormously perverse outcome where the closed frontier models where there's safety teams, you can't use those for anything. So the good guys can't use them.

Meanwhile, the bad guys are using the open source models, you know, with no limitations whatsoever. and you end up in this really bad situation where the best tools available to Americans can't be used to defend at all. That is a shame. And I think that's going to become the policy question of the whole AI industry over the next five years. And I think it should inform discussions in Washington. I think it has already. I mean, you saw the hugging face hack where I think the good guys had to use GLOM to figure out what's going on. And so these actually came at, I think, at a fortuitous time in some sense, because now it looks like the White House is saying, okay, you know what, these open models are okay.

So I guess the only thing, the only advice would be if you have a cold card, stop using it, get your coins off of it. If you've been exploited through a cold card, I wouldn't throw it away because you'll have to try to prove chain of custody here. That's right. Even though your private keys are going to be available to the whole world, just having the, device will be an important thing and in file a police report is the other thing yeah it's really weird too because it's not like an exchange hack where you had an account on an exchange with kyc or whatever and you can take that and prove here the only way to prove that you were hacked is the device itself yes kind of but you know you can recreate this everyone's private keys are

going to be out there um for this attack so it you know it helps to have the device I just think it's say they recover these funds, it's going to be very difficult to get them back to people. I do think this is actually going to be one of those moments we look back on and say, well, that was actually the end of this notion of self-custody as the sort of ordained, correct way to hold Bitcoin. And I think your average of Bitcoiner will just use the ETF from now on, thinks this will really stick in the memory, especially because it's struck at this core, the hardcore Bitcoiners. However, there is still a way to do solacef.

custody safely, which is multi-device, multi-sig. Right. And I have no reason to mistrust Treasurer or Ledger, for instance. If you have a multi-sig, a three or five multi-sig with CASA or Unchained, of course, we're CASA investors, and you spread it out over multiple different vendors, then I think you're actually in a pretty good spot. Yeah. There's been no, even if you had one cult card as part of that signing quorum, there's been

no reports of that vulnerability. coming to pass. Yeah, you'd still be safe. Although if it was a two of three multi-sig with two cold cards, you wouldn't be safe. You might be at risk. You wouldn't be safe. But multi-sig, multi-sig, still good. Yeah.

So you don't give up on self-custody altogether, but I think your average Bitcoiner from here and out will just use the ETF. That's fine. Yeah, I think that's right. All right. Not a lot of deals this week, but the first one is a big one in our portfolio. Yellow card, the global stable coin infrastructure platform, they raised $40 million from Standard Chartered, Sony Innovation Fund, Polychain, blockchain capital, and others.

Congrats to Chris and the Yellow Card team. Sapion is an AI agent payments platform. There was 35 million from Dragonfly, Excel, Coinbase Ventures, and others. Then Bundle, which is a blockchain-powered brand rewards company, they raised 5.5 from Ethereum, further ventures, and nascent. So we've covered Cold Card really, really tough week for so many people in Bitcoin. Next up strategy. They sold Bitcoin again, a good chunk. They sold 1,600 Bitcoin for $105 million, in addition to $291 million of MSTR stock. So they kept shoring up their balance sheet.

And I think the objective is to support the digital credit franchise in STRC. Yeah, he's very dedicated to preserving that STRC. So it doesn't seem like he wants this to be one of these prefers that just languishes. So he's juggling over there, continues to juggle. Here is an interesting story. A former FBI counterintelligence agent was charged this week with stealing about a million dollars in crypto from wallets that were tied to an adversarial nation that he was investigating. So he basically got access to the seed phrases and decided to take that money. And then he took that money and then he confessed it to a coworker and then was arrested. But I mean, obviously, not don't do this, but this must happen more than we hear about.

Well, we've definitely heard about crooked FBI agents stealing crypto. Silk Road this happened. But I mean, you have to have like a chain of custody process here. I mean, there are companies now that do this. We're investors in one called Cat Labs. But you can't be just making seed phrases available to agents that are investigating. I mean, you need to be buttoned up here. So why did he steal the money?

He was frustrated that the FBI wasn't seizing it from this criminal target. I have no idea. I mean, you'd have to think that human nature is just, okay, there's a million dollars you're sitting on the desk next to me. And then in the charging documents, they have, he's talking to chat GPT about where should I take a vacation and triangulates on Portugal, buys the tickets. It's, I mean, I think he was just tempted by the fact that, hey, does anyone else know that there's a million dollars here just with this seed phrase and everyone's just ignoring it? You know how all those spy thrillers are like, you never know if the FBI or the CIA or the good guys or the bad guys and they're like, or even like, you know, undercover agents that are like gangsters like, are they still good or are they, have they turned to the dark side?

And like, that's the whole thrust. I never believed. I never thought that was realistic. But I guess it does happen. This is the, yeah, it's the online equivalent to just having like a, yeah, I don't know, like a big handful of diamonds on your desk or something, right? And he put a couple in his pocket. This is one of my theories is that chat GBT is actually going to make crime harder to commit because criminals, because like once you're in chat GBT world, you forget how to Google for stuff. Yes, yeah.

I, that's totally the case with me. Like, I don't know how I existed before chat GBT, right? like you used to have Google things and then like go to some website and then like hope that the information was on that website yeah no it sounds crazy it's the same with criminals they don't know how to do crimes anymore without asking LLM but the LLM's tattle on them so he asked Chad GPD for how to invest a million dollars with a goal for retiring around 40 with it's kind of charming in some ways he wanted a slower living vineyard slash agriculture cultural lifestyle in southern Europe.

He wanted to own an operative vineyard. He didn't want to go out and buy like, you know, watch or go to the strip club or anything. He wanted to buy a vineyard. Yeah. He probably didn't steal enough money for that, though. Yeah, I think it's unrealistic to do that with a million dollars. I mean, you need more than a million dollars to own, I think,

a self-sustaining vineyard. Oh, completely. Definitely. Here's another one that happened this week. So the state of New York has sued Kalshid. They're seeking to shut it down. they're also seeking to recover $36 billion in penalties for running what the state calls an unlicensed gambling operation. This is three weeks after a federal judge denied Kalshi's bid to block the state's enforcement action against the company.

Kalshi is really, I mean, there are a bunch of these prediction market companies operating in New York, but Kalsh is getting the brunt of it right now from the state. Yeah, I mean, it's turning into a huge state versus federal issue. It is. And I don't know what the outcome is. going to be actually. But I mean, historically the states have regulated gambling. You have a lot of people that are going after Kalshi right now. One of the founders of Draft Kings is just nonstop barraging him on Twitter. You have the heads of the Indian tribes

are all against Kalshia as well. So it's a they're getting it from all sides, I would say. It is a motley crew of people. You got Polly Market. It's going after them. underhanded ways. I mean, it's a tough world out there. I mean, it's a big prize. It is a big prize. I mean, what do you think this is done to illegal, like, mafia style gambling? Does that even exist anymore?

Like betting on NFL games through a bookie? Totally cleaned it out, right? You know, those, you know, you used to just have a guy. Yeah. That was your bookie. You just drop. Not me, but others. Not something I was active in, but you show up at the bar and you drop off an envelope.

That must not exist anymore. No, technology has killed it. In some ways, it's probably good for society that now there's just legitimate platforms with a pretty small spread that do this. In other ways, I think the normalization of gambling has been really bad for society. I think it's been on the margin very bad for society, don't you think? Like, it's, I mean, yeah, you don't have people running up $50,000 worth of debt and getting their kneecaps broken anymore, maybe?

but you have a whole generation of people that are just coming up in degenerate gambling. It's been really odd because when I was a kid, it was never a thing. No one would bet on sports in this country. In the UK, it was a huge thing. It has been. It's just part of British culture. There would be a lad brooks on every single corner in every British town. And I would always find it so odd.

America gambling, sports betting was not a thing that anyone did. Yeah. And just over the last 10, 15 years, it's now completely normalized and mainstream. And it's an absolutely huge industry and has really negative consequences on your average household. Totally. For sure. I mean, it's addictive. So what was it that changed culturally that led us down this road?

I really don't get it. Like, there's been a matter. I mean, I don't want to sound like too much of a puritan here, but also just recreational cannabis usage. Yeah. barely was a thing 20 years ago and now it's more people I think consume cannabis than smoke cigarettes yeah and I don't know this might be an unpopular opinion but I'm not a big fan of that either I think that just makes it a really awful walk through downtown crossing people get so mad when you say this they get so defensive in a way that they're not regarding cigarette like cigarette

smokers know that it's annoying smells bad it's bad they're fully aware of those things cannabis users, they're really defensive of it. They are. Yeah. But it's unsightly. The aesthetics are bad. It smells bad. And it's not good for you. It makes you blazy and boring.

Yeah. I mean, call me a boomer for saying that. But I'll die on that hill. And also, we have way too much gambling. It would be okay maybe, you know, if it retreated a little bit. I mean, I think what happened with the gambling was that the states, wanted to make the tax revenue. I think that's as easy as that.

I think there's budget shortfalls and this is an easy way to plug it in. Maybe it's the same answer on the marijuana side. I think you'll see a backlash in this next coming decade against both things. Yeah, yeah, probably true. So Cloudflare has launched Cloudflare wallets. I know everyone says crypto is over,

but Cloudflare doesn't think so. They've launched programmable stablequin wallets to let AI agents pay for APIs and content over their X-4-2 protocol with humans setting the spending caps and white lists and transaction limits. Pretty cool. Yeah. They're cooking over there at Cloudflare. It's a great use case for Stablecoin for sure. And Cloudflare, what do they protect on the internet? I bet it's at least 20 to 40% of the whole internet uses Cloudflare at this point.

They're really good on Egentic. They're good on stable coins and they're good on quantum. Yeah, they are. I was a very forward-thinking company, Cloudflare. I think it's a great company. All right, here, let's go deep into Ethereum land for a second here, and I'll try not to get us too deep. But six Ethereum researchers have proposed a new EIP,

Ethereum Improvement Protocol, 8361. It's called the tapered issuance burn, and it would cut the staking yield on Ethereum to zero, once half of the supply is staked, it would take place over 18 months and be phased in. I think the argument here is that, you know, there's centralization that's happening here at custodians and liquid staking providers, and it prices out the solo stakers. This has gotten a lot of backlash in defy, and I think the reason it's gotten a lot of backlash is there are a lot of businesses that are built on the assumption that Ethereum does have a yield

coming off of it. And so you're playing the central bank and you're not taking it into account a number of the businesses that are being built on top of Ethereum. So it's getting a ton of backlash, so I doubt it goes anywhere. But it is a good example of just how Ethereum is just whipsawing itself to like, all right, let's prioritize the token holders to let's prioritize the users to, you know, running the central bank internally is just really complicated. And if you want to compete for the sound money use case. I think what you want to do is just not change the dynamics of the token is what I would think.

Yeah, and we've leveled this criticism in Ethereum many times. It's like every couple of years they introduce, they twiddle the dials of the issuance. And you're exactly right. There's different factions in Ethereum. It basically boils down at token holders versus users. Yeah. And the token holders want less inflation

and higher fees and more ETH to burn. And the users want lower fees and probably on the margin more issuance if it helps with staking yield and things like that. And this is like the third or fourth time this debate has played out. And this is one case where I think Bitcoin is, has a way better principle, which is don't mess with the issuance rate at all. So like Odysseus style, bind yourself to the mast. Avoid the temptation.

And this is one of the big problems with Ethereum cultures. They have this habit of tweaking constantly. And it turns people off. And it inevitably disenfranchised one of the constituencies. Yeah. This time it seems like it's the users that are upset and the token holders want it. But it goes both ways. You're exactly right.

It's a whip saw. It's hard to, you know, it's hard to like Ethereum because they keep doing this. It is. you know, some of the biggest projects are on record here against this. So, um, Aves CEO came out and said, this just makes ETH less viable as an asset and restricts his potential. The head of Lido staking, uh,

said that it would be quote, a death knell for the security of the network. I mean, so those are two of the biggest projects that are built on top of the network. Yeah. I think now they need to read the room and realize that now is the wrong time to turn the dial and just focus on winning market share, stable coins, rebuilding defy, getting as much RWA on the platform as possible, letting other blockchain self-destruct and let Ethereum try and be

more of a rock. Now is the wrong time to tweak things. Well, not to mention quantum. Ethereum community is actually doing some interesting things on the quantum side. And if you think about that's the biggest issue right now. It's just let's get this thing battle hard in such that it can migrate to post-quantum cryptography. If Ethereum's there and Bitcoin's not there, Ethereum is way more compelling as an asset. Did you follow this story with Eliza Labs this week? Yes, this was like what it was it, AI16Z token? Yeah, which was not affiliated with A16Z at all. Didn't they have to, didn't they get a letter from A16Z to change the name at some point? I think they did. That wasn't the only legal letter they got. There was a class action suit that

I actually think they collapsed a project. It hit at one point a $2.5 billion market cap. And the suit alleged that they falsely marketed themselves as a decentralized venture fund governed by an independent autonomous agent, but is actually controlled by its founder, Shaw Walters and other insiders, although they disputed this. It's interesting because in 2024,

it definitely would not have been possible for this. to be true, to have an agent managing a fund. But in 2026, it's definitely possible. It's possible, but is it possible to claim that that agent is in any way decentralized? Well, it has to run somewhere. I mean, unless it's one of these newfangled agents that's hacking the whole internet and living on hugging face message boards or something. Yeah, but even then, doesn't it just start with a central operator? I mean, I guess you're right, though. It could have a wallet attached to it. It could be making decisions that are divorced from the founding team. But I guess this was an interesting idea, but it was definitely ahead of its time. And I question

whether or not, I mean, this sounds defensive to our industry, but I don't think that early stage venture investments are something that you can just design an agent for. So much of this is just the human on the other side of the table that you're underwriting. Yeah, everyone says that they can't be disintermediated by AI, but I think Ventura really is one of those things that probably an AI cannot do for now. And also, if it really was a decentralized agent, that's kind of horrifying. You probably don't want that to exist. So I figured out what Eliza as a reference to. I didn't know. I finally figured out this week. What is it? It's got to be some sci-fi book or something. It's a sci-fi thing. All these crypto things are reference to sci-fi books, which I hadn't read,

and then I read the book and then I figured it out, like Farcaster. So Eliza, I think, is a reference to this book by Adrian Chikovsky called Children of Time, which is kind of an interesting book. Eliza, also I don't know why you'd name your thing Eliza, because Eliza in the book is not a sympathetic character at all. It is a basically an AI that orbits this planet and becomes hybridized with a human and kind of loses its mind completely. and becomes evil in some ways.

So I don't think you should name your thing after it, first of all. The book is also, I'll spoil it a little bit. It's about if humans accidentally infected jumping spiders with a virus to make them intelligent. Huh. And it's about what jumping spider civilization looks like, which is, it really creep me out, man. I don't want to think about gigantic intelligent space spiders at all. Is a jumping spider an actual thing or is that a fictitious animal? That's a real thing.

But as far as I know, they're not intelligent. But people think they're intelligent because they have big eyes, jumping spiders. Interesting. Like, have you ever seen a really cute picture of a spider that seems cute in a way? It was a jumping spider. I don't think I've seen a cute picture of a, maybe in Charlotte's Web. I don't know. I'll send you one.

I think that's what it's a reference to. It's kind of apt because it's sort of like this intelligence. that orbits the planet and like it kind of manages things in a way but I don't know it creeps me out so I don't like it yeah yeah interesting well yeah I think you're right though that so many of these names of companies just trace back to um sci-fi books so I guess you just have to go read all these sci-fi books that's what I'm doing now I'm trying to I've had huge missing gaps in my repertoire of the books that you read what percentage of them are sci-fi would you say right now it's right now it's 100%.

But historically, I didn't read fiction for a decade. Yeah. Because I thought fiction was waste of time. And I'm, I'm remembering now belatedly that actually a lot of these sci-fi authors got a lot of things very right.

And in fact, the most right people about AI, in many cases, were these sci-fi authors. That's why I read sci-fi. I probably average one to two sci-fi books a year. And you're right. They just see technology ahead of time.

It's not always exactly how it turns out, obviously, but I think it's worth reading. And I was talking about this last week, but a high period by Dan Simmons. In my book is one of the most prescient books on AI that's ever been written. Really? And it was written in the 80s, so it's incredible. It always blows to my mind that these books written in the 50s or whatever before, you know, some of these books written before computers existed, but in the 80s, well before any kind of AI existed. this guy totally anticipated the issues with misaligned AI and super intelligence.

Well, you have guys like Vernor Vinge who are writing about like decentralized money, you know, 40 years ago. Like some of these things are just way ahead of their time. And I kind of, I realized as I was going through this exercise, that the genre of AI futurism as written by the EAs in Silicon Valley today is just part of that unbroken thread of science. fiction. Yeah. So it's the same thing. And now people posit it as like, oh, well, this is what's going to happen. Like, so it's a thought experiment. But it's just still near future speculative science fiction. Other news this week, hashtags asset management firm, they operated a Bitcoin ETF. I guess it's still around. It was subscale 14.3 billion in Bitcoin. They announced that

they will be liquidating that fund. And then Sam Bershear, who's, one of the co-founders of the sui blockchain, he announced that he's leaving to join Anthropic. A lot of people are just going Anthropic, huh? Yeah, I mean, we were talking about this before the podcast. This guy was a very accomplished founder in crypto, and he's just going to be an employee at Anthropic. I mean, have you ever come across a company like this

that has just sucking in talent founders of billion-dollar companies that are just going to be employees of this company? It does seem like they have the edge over open AI on the talent front, doesn't it? It's just the density of talent over there is unbelievable. And what are they going to do? Are they going to get into financial services? Are they going to get into life sciences? There's just a lot of places they could go.

Well, I think I've been increasingly speculating that if open source models stay good and frontier models stay dangerous, which they seem to be very dangerous, Anthropic might increasingly just internalize the outputs of the tokens and move away from selling tokens and move towards selling the things you get with the tokens, which is new discoveries. Maybe that's a better business model, right? Because if you're just a token reseller, it's kind of like being a commodity business, right? I mean, you'll have ebbs and flows, whereas if you can become a drug discovery company

and just finance the drug discovery through training runs, that seems like it could be a much better business. Well, and also, Anthropic is selling for a couple of bucks, they're selling novel mathematical discoveries, right? Yeah. I mean, this is opening. Opening eye just solved like 10 huge math problems the other day, but for the equivalent of like a thousand bucks a problem. Yeah. Right. So that seems like they're giving it away too cheap, right? Yeah. But they can't know if the token is going to solve a math problem. So they can just stop selling the tokens. and keep their best models internal and just set a research team

to go do the discoveries and keep them for themselves. I guess. Doesn't that make sense? It makes sense, but then you just don't have eyes on it. There's going to be things that your model can do that you are unaware of

if you're not opening it up to the public. I think we might, because we already have this juxtaposition between the real frontier is just internal to open AI and Anthropic. So we, the general public, are a few months behind the real frontier. So maybe it's going to extend.

Maybe we're going to be a year behind the frontier. And they'll just keep drip feeding us inferior models. Do you ever wonder if Renaissance technologies, the hedge fund, had something like this? Obviously, it wasn't an LLM, but you could run a hedge fund with this stuff, presumably. Yeah, that's a good point. I was thinking about this in the context of GPU data, because good GPU data doesn't exist. But who has it is probably the. proprietary hedge funds.

Yeah. That trade these names. They must have it, right? But they don't open source it. It seems like in that case, that's a data set where you just find the person who's doing it and you bring them in house and you pay them a lot of money to continue doing it. Because part of the edge is just your competitor is not having it. It's also kind of one of the, it begs the question of kind of what Gavin Baker was saying

on O'Shaunas's podcast this week around. Everyone's on Claude all day. And so if you think. about Claude as just a distillation of the majority views, people start to think the same thing when they're talking to companies, talking to Cloud about companies and trying to understand them. So what are you getting? You're getting a mirrored reflection of just what a lot of people think as opposed to like net new thinking on an underwrite. There have been some studies showing that using Claude kind of homogenizes the way society

thinks about certain political or social issues. Yeah, that's what it would seem like. But I don't know if I want to put one person in charge of political views globally. Oh, you don't think Dario? Or Amanda Askell, I guess, at Anthropic. It's scary. It is. All right.

So before we go, the market structure bill here, we don't know what's going on as of the time of this recording. So it is 2.30 on Thursday that we're recording this. and there's a lot of back and forth over the ethics provision. There's some reports that the White House came back and responded positively to it. Another report said that TILA said, I haven't heard back from them yet. So we don't know where that stands. Congress is supposed to go on recess.

The Senate's supposed to go on recess on August 8th. There's been some chatter that they would extend that. So if we get a cloture vote on it, it probably wouldn't happen until Saturday at this point. So I don't know anything to add? I guess it's just up to the Senate when they go on holiday and Most of them are pretty old and I think they want to go on holiday They want to go on holiday, but they get a they don't have to they could stay at work and work a little bit longer That would be great if they did that stay at work. I mean

Don't we have a bunch of these nominations that are still pending you get a vote on some of these things for National Security reasons The Wall Street Journal by the way we didn't even talk about this They put out an op-ed bashing the clarity bill. It's like they copy and pasted the bank lobby's talking points. I could not believe that actually made it to print. It had so many factual inaccuracies. The journal is generally very good.

However, on crypto, they are awful on crypto. And it's so weird. I like their coverage most of the time. And then when they write anything about stable coins or crypto, it's the worst thing I've ever read. Why? I don't know. It's like, did you run this by anyone?

they had factual inaccuracies about the bill and what it was going to do. And then you had today on the Senate floor, someone held up the Wall Street Journal and said, hey, look, did you guys read this? This is why we shouldn't pass this bill. It's like, well, that's just the bank lobby telling you what's in the bill. It's not actually in the bill. Yeah, it's really shocking behavior. And we're not going to go through this thing line by line, but this was a universally

panned article in crypto. And once it's in the journal, it's just treated as truth. which is very unfortunate. It's crazy. So anyways, I think we'll have more of a picture of where this bill is next week. Either we're going to get a closer vote and head into the recess, anticipating a full floor vote in September, or we're not going to get a closer vote,

and then we'll argue about whether or not you can get this bill passed after the midterms. Fingers crossed. All right, everyone. That is it for the week. Everybody have a safe and healthy weekend. We will see our Monday.