On The Brink with Castle Island - Weekly Roundup 09/11/26 (Liquid hack, AI x-risk, Silvergate revisited, can AI leak MNPI?) (EP.739)

Episode Date: September 11, 2026

Matt and Nic are back for more deals and news. In this episode:  We bring back his the theme music Anthropic employees think there's a 10% chance AI destroys humanity Was Jacob Coxon astroturfed or real? Are Anthropic employees being sincere or cynical? Blockstream's Liquid sidechain is hacked What does the Liquid hack mean for privacy coins? Hunter Biden launches the $LAPTOP memecoin IONQ says they will be able to crack ECC in two years Ex Silvergate CEO Alan Lane speaks out OpenAI offers a solution to the Navier-Stokes problem  Can insider information leak through AI models?  Why AI doom is like Pascal's mugging Content mentioned: Alan Lane, Silvergate's Voluntary Liquidation

Matt Walsh and Nick Carter are partners at Castle Island Ventures. All of these 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, $85 billion. 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 we're back to our original music. man, did we upset some people by running a fresh intro last week? So the new intro, that was not popular. I don't think anyone liked it. No, I mean, did you, but you made that song.

So do you take offense to the fact that people universally panned it? I mean, I'm getting emails, just go back to the old music, brought down by bad mortgages. The old music isn't good. It's like a generic guitar riffs taken from some library somewhere that was like free of copyright. right. It's not good, to be clear. You all like something that's not good. You've gotten sick of it, but we've been running it since 2018. I mean, that's what people know the podcast by. Right, but that doesn't make it good just because it's old. I think it's good. I think it's Lindy. I think it's good. And we hear you. We hear you. You like song number two and you like the intro. Yeah, I mean,

I've been strong-armed into changing it back. That's fine. I mean, I'm not married to this new song. I made in 30 seconds on Suno. is cool, by the way. I do recommend you try out Suno. Suno is cool, yeah. Bad Boy's song one was made on Suno as well. But yeah, we'll go back to the old artisanal handcrafted. We had a human being make that one. So we're going back. Back in the day when you could use Fiverr, and now you just do everything on AI. Well, I mean, I guess if you went to Fiverr, they would just use AI. Yeah, probably. I guess that's right. Man, we have a lot to talk about this week. A bunch of AI stuff too, which was interesting. But before we do, why don't we

hop into the deals of the week. First one up in the Castle Island portfolio is TRM Labs, which is a blockchain analytics company. They announced a new fundraise valuing the company at a $2 billion valuation. Congratulations to Esteban and the team at TRM, crushing it over there. Payward, the parent company of Cracken, they raised $100 million from NASDAQ as part of a push to bring tokenized stocks to exchange in 2027. So does that guarantee that Cracken goes public on NASDAQ? I would think so. It would be a pretty strong inducement. So this brings the valuation of $21 billion, apparently.

Cracken is, you have to really tip your cap to them. They have been growing very aggressively via inorganic acquisitions. They're up to some interesting things over there. I'm looking forward to when that company's public. Next up we have Latitude, a stable coin infrastructure company. They raised $35 million from Oak, H-CFT, Coin-based ventures, N-A, and others. congrats to Cyril and the team over there. And then here's the last one of the week.

Circle, they have announced that they have acquired Taza Pay, which is a cross-border payments company for $400 million. I think this is a very smart move by Circle. They need to be pushing more aggressively into the payments category for sure. I think from a public market's perspective, just being a pure play stable coin issuer is less exciting than also being a payments company. So smart deal there. Congrats to both teams.

So I don't even know where to start. I mean, we've got apparently every employee at Anthropic is going on TV and saying the technology they are building is going to destroy the planet. Do we start there? I mean, it's the craziest thing I've ever seen. I mean, I've got people that don't pay any attention to AI asking me, like family members being like, so did you see that thing from Anthropic? They think there's a 10% chance that we're all going to die here as a result of AI. This is permeated. This guy's on TV.

Like, I've never seen anything like it. You worked at Anthropic for, what was it? Six months? Six weeks. Six weeks. And the opening eye before that. But, yeah, I mean, I think there's a lot of conspiracies about this, like, oh, well, this can't have been a grassroots thing.

Here's the synthesis. It was both, it definitely struck a nerve, and it was also clearly astroturfed. It's both things. It was a very convenient message for the clearly Democrats leading into the midterms, not to put a partisan tinge on a B-salt. Basically every senior Democrat was using this. So we know that it was astroturfed. It's very easy to tell.

You look at it because he had a Wall Street Journal exclusive. This is like a random low-level employee. You don't get a Wall Street Journal exclusive if you resign from your job at six weeks at a tech company. That never happens. Obviously, there's a PR firm or someone was coordinating the background, obviously. This guy's affiliated with a lot of these nonprofits that are focused on AI Dumerism. and he gets a TV appearance on CNN that same day. Those take more than 24 hours to set up.

Yes, correct. Anyone that's done TV knows how that works. I've never, I don't think I've ever done a TV interview where I didn't have noticed the day before. And then simultaneously you have basically every senior Democrat quote tweeting this and saying, well, you know, we got to be careful about AI, we've got to regulate it.

It's a very convenient message for the Democrats going to the midterms. I think it's a winning message. Now, none of this is say he didn't strike a nerve. He definitely did. You know, so ordinary people are scared and they're inclined to believe this guy if he's working on the frontier of AI, whatever, whatever. But I think the synthesis is it was both organically very compelling and also it was AstroTurf at the same time.

So what exactly is the endgame here for Open AI and Anthropics? So they want a regulatory moat. They want, you know, government approvals for model releases. and so that would stymie competition in the frontier markets in the U.S., but that would do nothing to the international frontier labs. So is it actually a winning strategy to be pushing for this moat? Yeah, it's a really interesting question. I think there's kind of two readings.

There's sort of the cynical reading, and then there's the reading where you take what they're saying at face value. So the cynical one is this is just a typical play where you push for regulation as the incumbent, and you know, you make it really, really hard to release a frontier model so nobody else can compete with you ever again. And it's smart to do that ahead of the IPO. That's kind of the reading where you assume that everyone's rational and not insane. I'm more partial to the second reading where they're actually, they have completely lost their minds. They actually believe this.

They think the technology their building is going to kill everyone, but only they can stop the technology from killing everyone. So I believe that they believe that they've been able to. believe this, even up to the very top, even up to Dario. And I think they think that only anthropic can align the model to make it safe. But what they've never explained is, how did they get everyone else to align too? So like let's say they solve alignment, which is I think in contradiction in terms, but let's say they do it. How do they go to the Chinese labs and say, hey, use our alignment module? Why? Not going to happen. Go to the open weight models. Okay. they're open weight.

Anybody can change them. Go to their competitor, open AI. So it doesn't make sense to me. Like saying this and choosing to stay working at Anthropic, so this one guy resigned, good for him. He's actually living his truth. But the rest of them that believe this and they still work at Anthropic,

it's completely incoherent. Not one of them has explained how working there is going to, you know, save the world from the AI menace. Right. Yeah. I just think it doesn't give enough credit to the frontier.

labs in China and I get that they're behind, but do you think anything that we do in the U.S. is going to slow them down? There's just no chance. Yeah, and I've been talking to people that are rationalists and their EAs, and I've been asking them, like, hey, what are your answers to the question? And here's what they think. They won't say that. This is the part they don't say. They think once Anthropic reaches RSI, recursive self-improvement, and builds ASI, you know, super intelligence, not just general intelligence. They think in a line, will be solved because Claude can destroy everything that's not aligned. That's crazy.

So that's what they actually think. So that's the answer, by the way. It's completely insane. They think we're going to live in this, like, everybody's going to live in the Claude world at that point. And we're all going to be a slave on Claude's plantation, basically. That's what they think. They think, you know, okay, what about Deep Seek? Well, Cloud will just destroy it.

What about opening eye? Cloud will just destroy it. They'll launch a swarm of, so. That's what they think. Yeah. What these people aren't saying is they're the craziest megalomaniacs we've ever encountered. Yeah. It's dangerous.

It is scary. I mean, those same people are the ones that are saying there's going to be no more private companies that they think that Anthropic is potentially the last private company. So I think what they don't realize. First of all, I don't think just one company is going to build a great frontier model. I think this is going to be lots. So, I mean, because there's never been a technology that didn't diffuse. Right. So does anyone ever invent technology just stayed the purvey of one man or one corporation for a thousand years?

No. I can't think of any. I mean, like the one I can sort of think of is like manufacturing silk. They actually kept a lid on that for a really long time. But at this point, it's diffused like crazy, right? Well, yeah, people stole the silkworms. Yeah. Actually, that's what happened.

So the equivalent will happen. So I think we'll have diffusion like every other technology ever. So I don't think there's a technology that just remains captive. And then also, I think the enemy. gets a vote. You know, like in warfare, the enemy gets a vote. China gets a vote. They're not just going to accept Claude hegemony. No, right? No chance. Open eye won't either. Opening eye is actually ahead of Claude right now. Yeah. It's like factually, opening eye is better than Claude. So, why does Anthropic think that they're going to be the only last man standing? It makes no sense to

me. Do you think it's going to be an interesting roadshow for Anthropic? Maybe they're already on it, but this will be a question that gets asked in every single one of those byside interviews. Yeah, and practically speaking, I think Anthropic did so well because they had all these lunatics that were really aligned. Yeah. Forgive the pun. They're aligned with Dario, right? They're like, okay, we're going to leave open AI. We're going to build like the moral AI at Anthropic, right?

So they had great retention. They did great. But I think that psychology, the culture at the firm, is now working against them, way against them. Unless this gambit works. Who knows? imagine every single data center Anthropic pays for sponsors or leases you have to go to somewhere in Kansas or Ohio

or Nebraska and you have to explain to these people hey we're going to put in this massive 800 megawatt data center the fans are going to run 24-7 you're going to hear this low hum a half mile from your house forever and you know we're going to put gas generators on the site and so it's going to sound like a jet's taking off permanently forever And also, by the way, we think there's 10% chance we kill you and all your children.

How does that sell? Sounds like a bad trade. Yeah. So like, I think if this, they continue on, like, Claude and Anthropical, we singled out, their data centers. And like, we as the civilians that are just like having this horror visited upon us, the natural reaction is just to go vandalize the data center. There's like sci-fi books about that, right?

I mean. This is a great way to trigger the, what? What's it called in Dune? The Butlerian Jihad. Okay. Yeah. Right? Are you Dune fan?

I'm three quarters of the way through the first Dune book, so I'm catching up on my Dune. Oh, well, they, I think they've covered it by now. I don't think I've, I don't think I'm at it. Maybe I'm halfway through it, but I've been kind of stuck on Dune. Everyone loves Dune, so I had to embark on that. In Dune, they got rid of the AI. Did you just ruin it for me? Well, no, that's actually kind of a premise.

the novel. Okay. I didn't read the back of the book. I just started reading it. So Dune is like where they really put a hard cap on AI. Maybe like what the safest want. I don't think I want to do a hard cap. I'm enjoying my AI and I'm getting a lot of efficiency out of it. I wish these guys would just tone it down a little bit. But I guess to your point, they actually do believe this probably. And I think it's a religion. I think this is turned into a, Anthropic is a religion at this Well, it's not just anthropic, it's rationalism. It's a secular religion. And this is what happens when real religion goes away. You get worse secular religions. Like rationalism has been around for at least 25 years. Yudkowski wrote the Bible of rationalism. And from the start, they believe that AI would rise up and become this secular, this machine god. And we have to prepare for that. And we have to make sure we have align it. But if you really believe that, if you think there's going to be a machine god, How could you possibly be so conceited as I think you could align it to your particular human interests? Yeah.

Like if you're giving birth to God, God is superior to you. So why would you think you can align it? It's very puzzling. All right. Well, we're going to come back to another AI story. There's a bunch of interesting ones this week. But maybe before we do, do you want to touch on this block stream liquid network hack? This was a, I think it was $319 million worth of Bitcoin was exploited this week by a quote-unquote white hat.

Well, this is an AI story too, don't you think? It has to be, right? I think this is probably the case where the frontier models were used to identify this if I had to guess. Yeah, I think so. So this is a really material hack, and it's really, there's a lot of implications here. So first of all, it was actually so hard to figure out what happened. It's insanely complicated, but this is why I think that an AI model was involved. Liquid is a block stream administered side chain.

it has a lot of Bitcoin in it. Then when you peg Bitcoin in, you get LBTC. And then you can do things on the side chain. Okay. They were around 4,000 Bitcoin in it, I think. Yeah. Yeah. Liquid uses this particular programming language called Elements,

which has this thing called confidential transactions, which is actually the same thing Minero uses for privacy. We'll come back to that later. Confidential transactions mean you can, have a blockchain with a transaction on it where nobody knows, except for the sender and the receiver, how much money has been exchanged. It's great for privacy, but also with privacy comes these auditability issues, like we've seen with Zcashbuck, the Manera bug, this is another example of that. For a confidential transaction to be valid, you have to include range proofs.

I think what a range proof does is it makes sure that the transaction is not negative. So it's a piece of cryptographic machinery that make sure you're not submitting negative transaction because what you could do is send at the same time a positive and offsetting negative transaction, creating inflation. The blockchain reads it is valid. You forget about the negative one and then you just broadcast the positive one to yourself and you've created inflation. So the range proof is meant to safeguard against this.

So the attacker was able to issue a valid range proof for a transaction. get it cached, reuse the cache. There was some bug in the cash logic, and then use that certificate of validity on another transaction, which did include a negative balance. And so through that, they created the offsetting positive and negative.

Through that, they created 4,000 LBTC for themselves. Okay, but so now they have these sort of like Bitcoin certificates on the side chain, how do they turn that into real Bitcoin? They were able to send the fake LBDCs to the service called side swap, which is not a block stream affiliated service. And they gave them the real Bitcoin, or the Liquid Federation gave them the 4,000 Bitcoin. It wouldn't have worked if they'd just gone through the Federation. Yeah. And there were no checks, interestingly, to make sure that you're not draining the entire valley of the side chain in one transaction. So the attacker did

get all of the Bitcoin. And that wasn't the whole story. They then gave back 85% of it after negotiating with Blockstream via Op return extensively. And it's still going on, isn't it? So initially they broadcast, we are white hats. But then they only sent back 85% of the assets. It looks like there's been a back-and-forth dialogue here, and there's some disagreements on, you know, should they keep the 15%.

Yeah, I mean, they're definitely not white hats, obviously. And also, I always wonder about this. We see these hacks where people negotiate with the hackers and the hackers give back some of the funds in exchange for immunity or that's not legally enforceable, right? Blockstream doesn't have the ability to say, hey, you're immune from prosecution. Yeah, like if someone beats you up in the street and you're the victim, you can't just say like that wasn't a crime. Right. Like it did.

It still was. So I don't think you can do any equivalent of signing something that says we're not going to press charges. And how do you figure this is going to work here? Is Blockstream going to make everyone whole on the extra 15%? it's tough. I mean, the loss as of right now stands at... It's like $47 million, right? $40, $50 million, $600 Bitcoin.

It's a lot of money. Some people speculated that the liquid supply was Blackstream's own balance sheet and they were kind of bootstrapping it. It's unclear. So it might be just balance sheet loss to them. But there's a lot of interesting implications. First of all, I think it worked.

the security profile of confidential transactions. Yeah. So knock on effects from an arrow. It might spook people regarding privacy coins in general. Now, Zcash is a different approach, but it is also faced these inflation bugs in the past. And then also, I think it will make people even more fearful of Bitcoin side chains, which were already not very popular. Well, do you remember when side chains were originally presented in white paper form?

I think it was 2014. We thought this was the feature of scale. We thought this was the way to effectively have, this was before Ethereum, but the way to have smart contracts on Bitcoin was going to be these peg side chains, which I guess the peg side chains just never ended up working. So you ended up having to have this federated model. Well, and what's interesting is the federation didn't. The security feature people were so focused on and argued about for years wasn't even the thing that mattered here. I know. So the thing that mattered was the programming language to allow the creation.

of the fake tokens and then also the failure to recognize that on the point of the pegout module. The Federation could have been infinitely large and it still wouldn't have mattered. I am interested to know

what type of frontier model was used if it was a frontier model or maybe this was like a Kimi K3 type of thing. Apparently the Blockstream software for Liquid hadn't been updated in years. That's about, mistake. We're living through, as we've been saying, we've been, we're living this through this time of danger where all vulnerabilities will be probed. But then eventually we'll get through it.

And then everything will be really secure. It's, it's going to be painful. I think there's going to be a lot more of these and also a lot of PII breaches. All right, let's talk about something else this week. Hunter Biden launched a meme coin this week. It was called, it is called laptop. He did a bunch of air drops to his substack subscribers. He had a big thing about air dropping to Trump coin holders that have lost money. I guess keep in mind, Hunter Biden, I think, has said that he's in somewhere between $5 to $10 million of debt in legal expenses, given everything he's been through. So it's clearly a money grab.

The token launched this weekend promptly went down 98%. Were you surprised? no i mean uh you know if trump couldn't keep his meme coin going then hinder biden certainly wouldn't be able to uh there were some allegations around foul play but it's kind of like guys is a meme coin so what are we even arguing about here buyer beware i mean if you're buying this thing it's you're taking your own risk uh people were alleging that his like sub stack subscribers got some kind of preferential access It's like, guys, this is a meme coin called laptop about his laptop, which contained all the evidence of his illegal exploits.

Is there any bigger picture takeaway to Joe Biden's son launching a meme coin? Does this have any implication on just crypto policy in the U.S? Or is this just someone who's highly in debt doing something to try to make a quick buck? I mean, the market cap, as of right now, this might be completely illusory and overstate the, liquidity, but it's 200 million, believe it or not. So it peaked at $5 and it's trading at 61 cents right now. So it's not literally zero, but it probably will be, I don't know. I think it's unfortunate that crypto has become the port of call for like formally disgraced people. Yeah. And then like Schrelli and Elizabeth Holmes for some reason and Hunter Biden, people that have been disgraced.

they like to launch their comeback tours with crypto. Yeah, I just wish they would not get off at this bus stop. Just keep on going. Like, why does it have to be, why do they have to pick on us? Yeah. I don't think any Democrats are going to see this and say, oh, I feel differently about Trump's meme coin now, and I'm going to vote for the Clarity Act.

I don't think it'll have any implication on where we stand on the bill or anything like that. No, and it's not like Hunter Biden is really in any Democrats' good graces. It is interesting to the optimization function around having to make $10 million really fast and the first place to go is a meme coin. I guess that kind of makes sense, right? Yeah. And there's this great short story by Goren called Clippy. I forgot what it's called.

It's basically like a really compelling story about what it would be like if AI took over the world. The first thing they do to get money is also they don't launch a meme coin. That didn't exist when you read the story. they steal a bunch of basically Zcash, they do an exploit of a privacy coin. So I think that's actually, crypto is playing a useful role there. If there's ever a misaligned AI

that's about to destroy the world, they'll come for crypto first. You go for the bearer assets first, I think, yeah. Crypto's always been like that. You press the right sequence of keys and you get money. Mention the Clarity Act, just an update on where that stands. So the cloture vote is still,

scheduled for September 15th. So I guess by the time we record this podcast next week, we'll know if the vote has passed. And obviously that doesn't mean that it's a law at that point. It just means it goes to a full vote and they'll have amendments from there. But there's a demarcation in the road on September 15th. Interestingly, the National Shariffs Association released a statement this week saying that they have changed their opposition from on the clarity act. They are now, quote, neutral on it. So they had been one of the loudest objectors of this bill. Definitely the biggest group within law enforcement that was worried about some of the AML provisions. And I guess they're comfortable now. So there's some tweaked language there to appease the sheriffs, which was

interesting because there was other law enforcement groups that were very supportive of this. I do think this is all just coming down to the ethics language, though. So Senator Tom Tillis from North Carolina said it doesn't look good now. and basically that the White House is going to need to bridge the gap on the ethics language. So to hear a Republican, albeit one that doesn't like Trump, come out and say that, it does feel like the ball might be in the White House's court right now to cave a little bit on this ethics provision. Yeah, it's not trending well, at least from my perspective.

I think the thing that's clearly happening, though, is that the rulemaking process at the SEC is just in full gear. And so I think the logical response to not getting clarity is to just get it via SEC rulemaking on the elements that you can. Also, this week there is a, at least if not, a development in quantum computing and announcement of one. So IONQ, which is one of the market leaders in quantum computing, they said they are only two years away from manufacturing the hardware necessary to crack Bitcoin's cryptography, elliptic curves. they have been known to make aggressive announcements so this isn't necessarily a hard timeline but they...

It's one that we should be paying attention to though. Yeah, we should. So, you know, take away the grain of salt what the industry says, but they have their end-to-end blueprint for Shores, which is the algorithm that you need to crack ECC. They think it will require 20,000 physical cubits take 26 days per attempt.

This is kind of in the same realm as the Oratomic paper that came out early this year. So, I mean, I think the evidence keeps piling up. Bitcoin needs to care about this. So 26 days per attempt is that, is the way to think about that is it would take 26 days to do a brute force attack on a exposed address. And so then, okay, so then they get Satoshi's a bunch of his coins first and then go to the next one. So you'd have a little bit of a lead time, but probably not enough lead time to get a, a fork pushed through.

Yeah, I mean, in theory, a malicious attacker that possessed the QC could go for the Binance cold wallet or something. Right. So you could target it at the bigger wallets. So I don't think it's really practicable 26 days per attempt, but still, you can do a lot of damage with just one attempt. I wonder if anyone at IonQ is actually doing the work to try to figure out which wallet they'll go after first.

I mean, hopefully that this would be targeted at, you know, a $5 balance on a wallet that's controlled by the company or something. Well, they just have to get a legal opinion, don't you think? Yeah, I think they'd be taking an outrageous risk to point this thing, point this computer at Binance out of the gate. I mean, there'd be. Or even Satoshi's old coins. I mean, that's still Satoshi's property. The one sector that has been discussed with some of these companies is, coins that were lost that are still owned by someone

that are provably lost and that someone goes to IoneQ and says can you crack these? Well, all these companies are pitching that as a line of business. It's like stop jumping in the landfill trying to find your hard drive, just come to us. I mean, I think landfill guy is a prime candidate. He is. He is the prime candidate.

That's going to have a happy ending for him at some point, I think. I don't know if he still, does he have the address? Yeah, how would you prove it, I guess? That's the thing. But he's in, what, the UK? Yeah, but if his coins are in something like Peta Scriptash, he's still screwed, which would be very unfortunate for him. That he's actually quantum secure. Yeah, that's interesting.

Yeah. I believe he gave up digging. He did give up digging. Yeah. Didn't he try to buy the landfill? Yeah. I think it was crowdfunding. It's kind of like digging was sort of.

of his life's purpose in the end. Can you imagine how disgusting that is just digging through a landfill? I mean, you get diseases doing that, don't you? That's crazy to me. Yeah, it was like 10 years of digging. But I guess it was what? He was over $100 million of lost Bitcoin.

I guess it might be worth digging. If it was a parable, you know, he would like find some gold or something, something that he wasn't looking for. He's probably found a lot of things he wasn't looking for. I don't know if anything, wants, though. So Tether is launching a $400 million private credit fund called Stable Fund with Fassanar

Capital. They're targeting, they're looking to raise outside capital up to $3 billion. They want to lend to small businesses in 60 countries using UST as a settlement rail. That's interesting. I wonder if they're looking at just the settlement process for stable coins, because that's actually an acute pinpoint in the industry, is just bridging the hours when banks aren't open to get stable coins into the ecosystem and vice versa, that's actually a great use case for private credit.

While we still have banking hours and blockchain hours, that's going to exist. Yeah. And it's, I mean, it's a pain point for all these companies, right? In the remittance space, FX wholesale markets, you do need a little bit of credit injection in that system. Here was an interesting one this week. Consensus announced that they're kind of separating into two organizations here. It will be MetaMask and then consensus.

And Metamask will be the wallet. It'll be kind of its own entity. It'll be run by Joe Lubin as CEO. And then all of the consent, all of the protocol infrastructure projects like Linnea will be under consensus. My takeaway from this was, I think they've been trying to take consensus public for a while. And it seems like Metamask is the prized asset.

And maybe this just simplifies the story for taking one of these entities public. Yeah, it's very interesting. I'm actually shocked to see Joe focus exclusively on Metamask now. But I guess that is the main business driver within consensus. Yeah, and he's going to retain a board seat at Consensus, and he'll be the, I think, executive chairman, but it sounds like he'll have someone else day-to-day CEO of that business. But, yeah, I mean, Metamask, I think, is just printing money here on a lot of the,

it's almost like a PFO business, front-end wallet. It's monetizing a lot on the back end. So Robin Hood has struck a multi-year deal to route their event contracts through a website called OG.com. So that's crypto.com's newly spun out prediction markets arm. They clearly like a flashy domain. And they took minority equity stakes in both entities pricing OG.com at $5 billion and crypto.com at 20 billion. That's interesting. Yeah, they're diversifying.

Obviously, they have the stake in Rothera. And I think that that business is doing quite well. But they had been routing a lot of this through CalShe. So I guess everything that we've just talked about is probably not great for Calci from a retail flow perspective coming off of Robin Hood. Yeah. And that's really the question for Polymarket and Calci is can you rebuild? If you have the flow and the users, can you rebuild a prediction market from scratch?

And the answer looks to be yes. Well, I guess from Robin Hood's perspective, they're looking at Calshy and saying, why would we route to your back end if you're competing with us on the front end, right? I guess you could say the same thing for crypto.com, but yeah, there's some interesting channel conflicts there in a prediction market space. Jack Dorsey's block has become the latest fintech to apply for national bank charter from the OCC. Very popular activity these days. Speaking of banks, it was great to hear from Alan Lane this week, former Silvergate CEO.

He had a great substack post. We'll put it in our newsletter. It was a retrospective on the 2022-2020-23 regional banking crisis. He says that Silvergate did not fail, basically. It had a 70% deposit run, but it was an orderly wind down and talks at length about the Biden administration's war on the crypto industry and some of the reasons why they decided to shut down that business. So I am sure you like this substack. Yeah, it was great to hear from Allen Lane. I mean, it's nice to hear X Silvergate employees' leadership finally speaking out implies to me that maybe some legal time periods have expired. But Allen basically confirmed what we've been saying this whole time.

I think this is actually the first time he's really been able to publicly speak about it. But his view is that Silvergate survived that 70% run. They have the liquidity to serve their clients. They were not forced to wind up. they submitted to voluntary liquidation when their business became illegal overnight, effectively, with the Fed, other bank regulators saying that they couldn't serve the crypto industry. They were a boutique focused only on the crypto industry. They were forced into voluntary liquidation, which people miss.

And they were the target of a coordinated political attack from every side, from short sellers, working with Congress to folks like Elizabeth Warren. to the FDIC, the OCC at that time, the Fed, you know, really no bank, no small bank could survive that. And so I think history will vindicate these people, and we've, you know, tried to play a part in that too. But it's worth reading the substack. We'll post it in the show notes.

I mean, there has got to be a case. If you were a equity holder and silver, you got absolutely kneecapped in this. And all these allegations about the money laundering and stuff, all of it wasn't true. I mean, there's a, people lost real money on this, including Alan Lane and all the people on his leadership team. Yeah, I mean, Alan was personally prosecuted after this. But, yeah, I mean, the shareholders should be aggrieved.

I think the common stock still trades, actually. But you had Mark Ahodes going around sending these PowerPoints to members of Congress who then wrote letters saying Silvergate was illegal in cahoots with FTX, doing money laundering. None of that was true. No. They were a victim of FTX like everyone else was. Yes, FDX uses their platform, but FtX used many corporate counterparties and defrauded all of them.

So, Silvergate. Now, did they have some AML failures on Senn? Yeah, they did. But none of the worst allegations of criminality were remotely true. None of that was true. Everything Mark Cohodes said about Silvergate, none of his allegations. were proven out. So it's really tragic what happened. I think the government basically made their business illegal. So I think there's a case for like a regulatory takings claim.

You know, the government's not really allowed to eminent domain private businesses by turning off their business by informally jaw-boning them to death. So I don't know. No one's brought a case. I think someone ought to. Yeah, I think that's right. All right. Well, let's talk about the other AI story of the week here, which is Open AI. So they have solved, what is it called the Stokes? Napier Stokes? Navier Stokes.

I confess that I was not really up on my propulsion math problems before this. You remember that short story I wrote about the quantum thing with Bitcoin trillion dollar salvage? In that story, which is set in 20209, AI has solved

all the millennium problems, the year before. Oh, is that right? So we're on track here. So this is happening the way I thought it might. They haven't solved the most important ones like the Hodge conjecture or the Reuben hypothesis. But this was one of the 10, I think 10 millennium problems, which are considered the most important math problems. And, you know, they keep solving interesting problems, like the Erdos's problems. That's just a list of problems. But the Millennium Prize problems have a prize associated. And those are considered the most critical, important mathematical questions. And many of them arguably have, people always say, well, it's just math for the sake of math. So a lot of them have

real important applied aspects to them, including Navier Stokes. Does it? Because I don't know that I understand what the actual day-to-day implications are going to be of solving this problem. Yeah, I mean, you know, you could have said that to Isaac Newton when he discovered gravity like, okay, well, it's not relevant to me. Yeah, well. But then we came up with planes later on. This is a key problem in fluid dynamics, which I think interacts with many aspects of our lives.

Well, I guess if you just ask Cheshdipati, what the implications are, it's, you know, they're going to improve turbulence theory. So, you know, I guess that's good. It'll model how we look at turbulence of aircraft. And yeah, I guess, you know, maybe we won't have as many bumps because we'll figure that out. I don't know. I think you're underwriting the importance here, Matt. Well, I'm sure. I mean, this is obviously above my pay grade,

but it would be nice if there was like some immediate thing you could do with this. There will be new mathematical tools as a result of this, which I guess that's good, better climate models, it says. That seems good too. I think you're being really unimaginative about this. I'm sure I am. I mean, And people have been working on this problem for a long time. I'm sure there's a reason for them to be working on it. I guess bigger picture item here. So it's a million dollar prize. They won the prize.

There's a bunch of angles here. The first is that there was a rumor that Anthropic was pretty close to solving this. And that Open AI heard about that and went to go solve it. And then the bigger story is that there was a mathematician who'd been using ChatGBT GBT to try to solve it himself for a few months. And I don't think he was on the enterprise plan. And so there's a suggestion that Open AI used the work that this guy had been doing and putting into the prompts to actually get a head start on solving it. And there's no way to actually know if the model that they're using to solve it, the frontier model, was trained on the data that this professor had been using.

So he might have gotten 95% of the way there. And, you know, Open AI just finished it, basically. Yeah, I mean, or worse, you know, that. somehow they learned through looking at the queries. Yeah. I mean, we don't know. No one knows really what happened. But this is the most interesting implication.

Is opening I just going to internalize all scientific discoveries made on its platform? I mean, because they could. Maybe, right? How much are they worth? I mean, this is a million dollar prize that you probably spent 25 to 50 million of compute on solving. So the ROI is not there. But it might add to the market cap of the company to be. at the frontier solving these things.

Yeah, I mean, it's enormous prestige. That's why both firms are racing to solve these problems. I mean, I think if open source models stay good, then you're going to see some compression on just the sale of inference, the sale of tokens. And yeah, I do think the big two will start to internalize all the discoveries made on their platform. As of right now, you can check a box that says,

please don't train the model on my queries. Well, if you're on the right enterprise plan, then they're not allowed to legally. I don't know. I don't allow to. Hopefully they don't. You know, this does make me think about other ways information could leak through a model. Like people upload M&A documents for a legal review, right? Yeah.

Not all of them have the right privacy setting. So now the model, the hive mind, it knows that some company is going to acquire some other company. So that's MNPI. It's definitely in there. That is 100% happening. It's definitely in there. And now what? If someone goes to chat and says, hey, here's $1,000.

Could you agentically trade this for me with the best, lowest risk opportunities? It looks at all the opportunities in its universe and it finds these MNA, you know, this activity that's happening, trades against it. Who has leaked what? who is legally responsible? Who has insider traded? The end user asking chat to trade on its behalf hasn't done anything wrong, presumably. Yeah.

But they've benefited from being tipped off. So, and if you upload documents in a chat, unwittingly tipping off someone else, is that illegal? Just because you didn't have the right privacy setting. It's a great point. I mean, think about all of the MNPI that Harvey or Ligora must have. I mean, you have all these lawyers that are getting things during these transactions and just uploading, okay, take a look at this, prepare a new schedule. These are treasure troves of MNPI.

Yeah. And some lawyer and judge is going to have to figure this out soon, probably. I bet it's already happened. And in my understanding, it's not very traceable. Like, who put that into the system in the first place? I think you're going to have a really hard time finding that. Right. Exactly. So it's such an interesting question. I hope Matt Levine covers this soon. Because you can have a situation where someone puts a ton of big options trades on a target that gets acquired. The SEC looks at it. They're like, okay, this is obviously inside of trading. They go ask the guy. And he's like, oh, well, I just had my agent do trading for me. So what? Yeah. And then they're going to ask opening eye. And they're like, yeah, I mean, I guess the agent gets tipped off unwittingly.

that that is 100% going to happen. And the person who put on those trades is probably in the clear, right? Because, hey, I just ask Google, you know, like it's the same logic. I mean, I don't want to pitch illegal trade ideas to anyone, but it's kind of an interesting concept. Tell the agent, hey, go go think about your 10 best trade ideas and execute on them. And don't tell me how you got them. Yeah, you know what we need. We need guardrails for that in a regulatory process.

And, you know, this is what the P. Doomers are saying. This is what we should worry about is insider tipping, not, you know, extinction. The end of the world. I don't know. There was a good, people were having fun with that anthropic guy's post. And I saw DA Wallach put up something about, well, we really need to be careful of electricity because electricity has been behind COVID and the Holocaust and nuclear weapons. These things all used electricity.

We should really rein it in. I mean, it is an uncomfortable discussion. I don't think we should let the fear of the bad things that will happen with technology. eclipse the good things that the technology will do. No technology is universally good or harmful, right? And when I think to myself, will AI kill people or will it be used to kill people? Of course it will. Like any other technology, like guns are a useful technology, but they're, you know, killed a lot of people.

And so has, you know, building dams. Dance have been great for humanity, right? Amazing. Probably one of the most important things ever. but they also collapse sometimes and kill thousands of people. That's true. There was a dam collapse in China and the 70s killed hundreds of thousands of people. Nobody ever heard about it.

Really? I think it was during the cultural revolution, so it was covered up. So would you say, okay, well, now that dam is so we can't use dams anymore. Yeah. So now the DOOMers are saying something different. They're saying, no, it's not just regular harms that are going to happen. It's like everyone on earth is going to be killed. so and it's this concept in epistemology called being mugged by Pascal have you heard of this

have not so you've heard of Pascal's wager though yes where yeah uh you kind of choose to believe in god because the consequences are good right right so this is the inverse we're just like okay we're doing a utilitarian analysis the we multiply the odds of something happening by the badness of that thing. So infinite downside doesn't really matter what your odds are. 0.1% while everyone on earth dies. So the expected value is negative infinity, right? So now you've determined something's existential, even if the odds are low, you are highly motivated to solve that thing. Right. Even if the odds are 0.001 because the downside is all conscious life is extinguished, which is infinitely bad. You have been mugged by Pascal. You have been

forced to redirect all of your current effort and energy into preparing and avoiding this existential risk. That makes sense. And it's a failure mode for consequentialism, like reasoning about moral actions based on their probabilities and only looking at the consequences. This is kind of a known failure mode, but this is what's happening with the rationalists. They're being mugged by Pascal. It's fascinating. Well, I think it just feels, do you feel like it's every week we have another one of these news cycle events. And, you know, this week it was Rokana. Last week it was Bernie Sanders. It's just, it feels like it's just nonstop with this. And AI hasn't even done anything bad yet. And it definitely will. Yeah, I mean, the hugging face thing is probably the worst. But I mean,

yeah, there, something bad will happen. Yeah, I mean, it'll happen in the next six months. I can guarantee you. Imagine an AI swarm takes down the power grid in a medium-sized country for two months. Right. Then what? Like, now we're in real. So I think that's what we should be. worried about not like X risks really what about the real harms that an AI model could wreak in the hands of just a bad person not that the models misaligned right the model is aligned to the person wielding it it's just that they have bad objectives yeah what what about what what about them you know how do we mobilize AI for defense against that I think that's the question all right well that's a good place to leave it hope everyone enjoyed the

intro music it's back I'm gonna fight to keep it back but you know thank you for your comments i know that was that was a tough one for you yeah i'm all more than it all right everyone have a safe and healthy weekend and we'll see on monday