Loading summary
Transcript291 lines
- [00:01]
A
Welcome to Galaxy Brains.
- [00:02]
B
An infinite amount of cash.
- [00:04]
A
Cash. I'm your host Alex Thorne. The US Banking system is sound and resilient. Bitcoin made a new all time high.
- [00:11]
B
If you're not long.
- [00:12]
C
If you're not long, you're short.
- [00:14]
B
Satoshi is going to come on there, laugh hysterically, go quiet. All bitcoin's gonna be erased.
- [00:20]
C
Bitcoin.
- [00:20]
A
Bitcoin's the best crypto.
- [00:23]
C
Bitcoin is going to zero
- [00:27]
A
welcome back to Galaxy Brains. As always, I'm your host Alex Thorne, head of Firmwide Research at Galaxy Bitcoin. Not Zero. We have a great episode for you this week. Nick Carter, co founder of Castle Island Ventures, is our guest. This is at least Nick's fourth or fifth appearance, third or fourth on this show. And fun fact, Nick appeared on the show in November 2022 shortly after the collapse of FTX. And that was our first video episode. So if you scroll all the way back on our YouTube channel, you will find Nick in a leather jacket sitting right here in our New York office. But this episode is excellent. I talk at length with Nick about AI, about alignment of AI, the emergence of AGI or asi, the interaction with cybersecurity and regulation and economics, and it's a wide ranging discussion primarily about AI. We also talk about Bitcoin and Quantum. You may have seen that Galaxy announced this week the launch of our own Galaxy Bitcoin Quantum Readiness Initiative, which includes up to 5 million in grants for open source developers working to mitigate Quantum on Bitcoin. And also another big announcement that we discuss was the formal launch of the Bitcoin Security Consortium which includes Galaxy as well as other major companies like Coinbase, Fidelity, BlackRock, Ark, Anchorage Block, Blockstream and Strategy. So we'll talk about that. And you know, Nick has been one of the louder voices and I would say in many ways more more thoughtful voices on quantum computing generally. Fascinating interview with Nick. And of course we'll talk with our good friend Bimnet at BB from Galaxy Trading as always about markets. Clarity act is moving markets a little bit like some interesting things out here happening in the market as always that we'll talk about. Before we get to all that, I need to remind you to please refer to link to disclaimer in the podcast notes and note that none of the information in this podcast constitutes investment advice or an offer, recommendation or solicitation by Galaxy or any of its affiliates to buy or sell any securities. This is a great episode. I loved it. I don't want to talk too Much more. But next week I will be in D.C. all week and we're going to see if I can do an episode there. But you know, we'll have one regardless. But it's like Clarity act, you know, Scott Besant said Clarity act is on the one yard line. Part of me thinks it's that's about distance to finish. I think maybe more accurate might be that we're deep in stoppage time, you know, and so meaning we're running out of time. Not just we are close, but we're also running out of time. Anyway, let's hop into it with Bimnet Abibi. Let's go now to our friend Bimnet Abibi from Galaxy Training. As always, Bimnet, welcome to Galaxy Braves.
- [03:04]
C
Thanks for having me.
- [03:05]
A
Well, we've I'm thinking about the bitcoin price. That's what I'm struggling to decide whether it is doing its eventual grind higher. And was 58 the bottom or is 65 which is really not that far up from 58 just bear market noise. We did a poll internally at Galaxy of the trading team and and the risk and research teams and it came out dead even. 50, 50 either bear market noise or the beginning of the grind higher. Where do you think, how would you characterize 65, 5.65.8?
- [03:38]
C
I would say that as we currently sit Here it is July 22nd and the timeline that we've discussed over the past year or so has been normally takes about a year from the cycle high for crypto to bottom. I think we are within spitting distance of that timeline bottom which makes your risk reward much more favorable to the top side at this point in time. I think if you're an investor thinking six months to a year out, the return profile of bitcoin is fairly asymmetric. We think that it can possibly the range low is around 58k. The cycle analysis says you can get to let's call it 50 to 40k region. It's possible but let's consider that low like 40ish.
- [04:40]
A
Right.
- [04:41]
C
So you can lose 25,000 points. So like a little like around 30%
- [04:50]
B
or
- [04:52]
C
it can go back to the highs and double. And so I think from a risk reward standpoint it's getting reasonably attractive. Now I think in terms of near term catalysts, clarity is the most notable and I think once you have clarity on clarity, pardon the pun, that should kind of set the tone a little bit in the near term. And so if clarity fails and we're talking about punishment post August recess And pre midterms trying to jam it through. Then I think you see a dip. But that dip is probably a buying opportunity, particularly for those that are investing in the medium to longer term. But you are getting to the point in the cycle where you're getting closer to the bottom every day that passes.
- [05:43]
A
All those people who bought the ETFs from January 24th to the top in October 25th, they've had a lot of chances to sell. Now we ripped back to a hundred, we got to 100. We went down to the 70s like back last year, we got back all the way to 126. Then we got all the way down to you know, 60. Then we went all the way back up to like 82, 5. Like you've had plenty of times. I just can't see how much more like who is the net new marginal seller of size at these levels. That's why it's hard for me to see kind of agreeing with you. The downside is mostly happened. There could be more, but it feels like it's mostly happened.
- [06:25]
C
I, I would tend to agree. And you're seeing that a little bit in the flows data ETF inflows have actually started to pick back up and are positive now versus like several weeks
- [06:36]
A
of just, I think it was eight or nine weeks of outflow.
- [06:40]
C
Just continual outflows. And then you've also dramatically reduced left tail outcomes with respect to microstrategy given that they've raised so much cash. True. And so you're crazy. Like MSGR is a huge force seller into a thin market. Thesis doesn't apply as much anymore. And so with flow starting to pick up, I think it's just a matter of time before sentiment meaningfully turns and you know, crypto like the thing that I'll say about it is like the community is just so strong even after dealing with like such a huge sell off. And you know, there's a ton of jaded people, but I still have tons of friends, colleagues that are still dcaing into bitcoin. Yeah.
- [07:32]
A
And are just ride or die.
- [07:33]
C
And I frequently get questions of like is it time to buy bitcoin? Is it time to buy?
- [07:38]
A
Right.
- [07:39]
C
And really like what a lot of particularly retail investors need to buy is for price to start moving higher.
- [07:47]
A
Right. I think.
- [07:48]
C
And then the narrative follows afterwards.
- [07:50]
A
Yeah, you've talked about this, how it's one of the most reflexive assets. Right. Like that so many people buy bitcoin to make money and they tend to trend follow higher. Right. There's A term for this Veblen. Good. The more expensive it gets, the more popular it becomes. Right. Which is kind of. I think bitcoin has demonstrated that a number of times.
- [08:11]
C
Yeah. So it's a really interesting setup. And I think what is lost on a lot of folks because we've just kind of gotten numb to it, is kind of the institutional adoption that has become like as pervasive as I've seen it. We're constantly talking to some of the largest institutions in the world about stablecoins, about tokenized assets, about perp trading and things of that nature. And so I think in terms of tradfi, more and more folks are coming on board by the day. And it's not just domestic, it is abroad as well. And so I'm about as constructive over the medium to longer term as I've been in a really long time. And I think you're in an environment where dips are meant to be bought and that being short is probably a very dangerous or if you are going to play from the short side, you got to have some tight stops and be agile. Be agile. Long story short, I think things are becoming a little bit more constructive. And then part of the reason why a lot of folks got into crypto is because of the constant devaluation of fiat. And some of the trends that are emerging in broader macro are going to bring that to the forefront, I believe in the next three to six months.
- [09:44]
A
Oh, interesting.
- [09:45]
C
And part of that is just the amount of debt that is about to get issued, like globally is just outrageous. You were talking about staggering amounts, all time highs, staggering. Google's got earnings today. They're expected to spend, I think in 28, almost $300 billion in CapEx or something absurd like that. So there's a ton of hyperscaler debt, ton of debt associated with the AI buildout happening here domestically and abroad. You've got huge fiscal impulse from kind of the geopolitical military conflict side of things. You know, Hegset's asking for a $1.5 trillion.
- [10:31]
A
I saw that. They're asking for hundreds of billions for data center, their own data center stuff that makes sense. I mean, I'm not surprised.
- [10:38]
C
And what you've seen happen is just a gradual move higher across yields globally. The BOJ has made sabre rattled a little bit about hiking rates a little bit more aggressively. The ECB is expected to hike, the bank of England is expected to hike with a bunch of other banks at the same time. They're going to be spending more money, not less the math eventually catches up to you. I think that story, the debasement story. Yeah, is, is debasement, fiscal irresponsibility. You know, you might even have a government shutdown later in the year.
- [11:21]
A
It's very possible.
- [11:22]
C
And so like the narrative, you know, might naturally start to take shape later in the year. And you know, I'm looking at things like the 30 year government bond here in the U.S. it's trading around 515, like the cycle highs like 520. Where do we go if we break that? And so there's some interesting things shaping up. And yeah, even if it's not that story, I think it's one of those things that again, narrative will follow the price action, people will make up whatever.
- [12:01]
A
It really is true. The thing about the cycle is so crazy because the four year cycle in bitcoin, which we've talked about a lot, and your point about later in the year, like if the prior ones, if it holds to the timing of the prior ones. Yeah, we're looking at like October, November would be like the bottom historically that amount of time. But part of the reason it needs that is that it's not just the price pain that you need to set a bottom, it's the time pain.
- [12:25]
C
Right.
- [12:25]
A
Like people need time to like, forget it, be angry, then forget about it, then move on and then, and then find it again. Right. And that's what's happened repeatedly in bitcoins history. And bitcoin itself has changed very little. So very few of the catalysts have actually been, you know, endogenous to bitcoin. They've all, they've all been narratives emerging from the market where bitcoin people converge on bitcoin. So I love your point about that because I, I do think those things, first of all, they're never going away. Lynn Alden says what? Nothing stops this train, referring to the government debt situation.
- [13:00]
C
Nothing.
- [13:00]
A
Yeah, so like I think that that's coming and it's gonna be really interesting in the second half of this administration now with Warsh at the Fed, like, because he, he has seemed to be, I mean just the way they communicate is dramatically different, like very possible. We get totally different monetary policy than we're used to. I know, right. It's possible.
- [13:19]
C
No, it's more likely than, than not. And the question is, you know, is there going to be a policy error, you know, made along the way, which
- [13:28]
A
there almost always is at the Fed.
- [13:29]
C
Right.
- [13:29]
A
I mean, no offense to them, but like they pretty much screwed up all the time.
- [13:34]
C
And you Know, and the real question though is like, ultimately, like, are they going to have to like, print money to buy a bunch of debt back?
- [13:42]
A
Correct. Everyone thinks it's inevitable that that happens. The question is, when does it happen?
- [13:48]
C
Right now.
- [13:49]
A
Under what circumstances do they actually have to be forced to do it?
- [13:52]
C
Right now? What? You know, without getting into a deeper discussion, they've effectively created a mechanism where liquidity in the front end helps absorb all the supply that is issued.
- [14:05]
A
Got it?
- [14:06]
C
Right. So, like, you know, as long as people can finance Treasuries at overnight in the repo market, you know, they're okay buying the debt. And who has the money to finance, like to finance the stuff on an overnight basis? Well, the banks with deposits and, you know, money market funds, et cetera. And so as long as you have enough liquidity in the front end, you should be able to absorb a lot of supply. But structurally, like, there is a limit. Right.
- [14:32]
A
We don't know quite where it is though.
- [14:34]
C
Right, we don't. But the idea, what's so interesting right now to me is real yields, which are effectively your nominal yields minus expected inflation, are continuing to trickle higher, but at the same time it's not driving people to come and buy the debt. 30 year real yields are about 290. So you get paid 2.9% above expected inflation to buy the 30 year point. And yet we're continuing to sell off. Right. And so it's getting a little concerning right here. And what does that mean in the face of like, even greater supply coming?
- [15:22]
A
Well, that's a great question. Bimnet Abibi from Galaxy Trading. Thank you so much.
- [15:27]
C
Thanks for having me.
- [15:28]
A
Let's go now to our guest, Nick Carter, co founder and general partner at Castle Island Ventures. Nick, welcome back to Galaxy Brands.
- [15:36]
B
Hello. Hello. Is this my third appearance on the show?
- [15:39]
A
I. It might even be fourth because. Yeah, because you were, I think, our first external guest in person in our old office, like in 2021.
- [15:48]
B
Hasn't been that long.
- [15:50]
A
Yeah, five years ago. And then I think you were a guest and sitting in our office. But it was like one of our last episodes before we started doing video. And so that would have been what, the end of 22. We do have him in video on the chair. You were wearing a leather jacket, a blue chair. Okay, so you were on video. And then we did a remote episode, I think two years ago called AI Will Take Every Job.
- [16:12]
B
Didn't happen.
- [16:12]
A
So I think this is for. It hasn't happened yet.
- [16:15]
B
You know, I can't believe this show has been running for five years.
- [16:19]
A
I can't believe your show has been running for like seven years or longer. How long is yours?
- [16:24]
B
Almost eight now.
- [16:25]
A
It's crazy.
- [16:25]
B
And I was just. I was trying to think like, wow, so many bitcoin podcasts ceased to exist in the last year. But right, we're still going and now you're five years old.
- [16:34]
A
Well, we had to. We have to like everyone expand our aperture to include, you know, the new things like AI and like tokenization and stuff.
- [16:44]
B
So.
- [16:46]
A
No, but I. It's again, Quantum. It's great to have you here. I think we want to talk specific primarily about AI and quantum today. You are an investor in Core. We've had been a long time. You've been covering and talking about AI for years now. As I said two years ago, we did a whole episode on AI and so much has changed. Let's start with AI, because so much has changed. Just this year, like, did you have this moment? Because I left like, you know, work for like Christmas vacation at the end of last year, having used ChatGPT a lot and like last year. But like. And then there was this explosion. I think it was right around when OpenClaw was released. And I come back like in January, and it's just like Claude code codex, like. And even just from then to now with like 4.8 and Fable and 5.6 SOL, like, the improvement from. And the growth in Enterprise usage from just January to today is like stunning.
- [17:42]
B
Yeah, you can feel the AGI.
- [17:44]
A
It's close.
- [17:45]
B
I mean, I've been on record saying I think we have had AGI for. I mean, I think everybody has a different definition of AGI. Like, Vitalik tried to define it this week and he said it'd be like if you could just plug it into a robot and they would be as good as a human at everything. That's not my definition. Mine is just artificial general intelligence, which I think we've had since fall of last year. Yeah, I mean, it's as good as a human at basically any cognitive task.
- [18:12]
A
I mean, I gave it after I saw the Odyssey last weekend. I had this idea to basically like the declinist canon, you know, like Ray Dalio and Turchin and like. But I was like, I had put out a report identifying common indicators that topped in bitcoin peaks and bottomed at bottoms and sort of cataloging those that were reliable and then assessing whether they've hit in October as a top or now as a bottom. And I was like, can I do that? Exact same concept, but for like empire decline. Like, why don't I just. So I'm like, it's late at night. I'm just having fable. I'm prompting it to look at like ancient Greece, ancient Rome, like the Habsburgs, you know, uk, you know, Weimar Republic, like, and then all. Any elders that I'm forgetting, you know, the Chinese dynasties and like, find those indicators in Disha that like were spiking at whatever the pivot from all time high to eventual decline was. And I can just say that this thing basically one shotted like a full PhD thesis in like 35 minutes. And it's good, it's quite good. It references all the other people who have worked on these questions and catalogs them. I was like, it's like a piece in the pocket.
- [19:35]
B
What's the verdict on the US oh,
- [19:36]
A
we're flashing plenty of them. But the thing is too, that one thing it pointed out that was quite interesting was like that actually we flashed all of them during the Great Depression and it was like the New Deal that like pulled America out of it and like reset the. All the indicia. And I was like, so it is possible, according to this report to, you know, pull the plane out of the tailspin.
- [20:00]
B
But that's Turchin's Clyde Dynamics as well. He thinks it's very quantitative.
- [20:06]
A
Yeah.
- [20:07]
B
And.
- [20:07]
A
And I mean, some of the most common that fable identified, of course, like, we're aware of these hyperspeculation. Right. I mean, we just published a report called the Race to Trade everything. Right. It's not just perps like prediction markets, just the male gambling epidemic in America, which is substantial, I think. And of course like the, you know, the, the Dutch with the supposed tulips, which I'm not sure if that actually happened, but like Weimar was like, absolutely. The inflation, the government debt. And one of the more interesting ones that it says is very consistent across these prior falls is like the immiseration of young men, the like. And how that leads to like a lack of marriage ability, which leads to like the downfall of the family, which like messes up the entire economy, which like is in itself reinforcing because it causes more crime and like. And apparently that's very visible in the statistics in America today as well. So I, I don't know. I. But again, even if it's not like perfect, I'm not like necessarily relying on like, it's. It did one shot like a sub. It reads the history or has already of all of these places, somehow found statistics of all of these Times it read and incorporated the famous authors in the canon in like 30 minutes. I mean just to do that, even if you don't do a good job at it, I mean that would take months.
- [21:30]
B
Yeah. My personal benchmark, I think everybody has a benchmark for AI. Mine is can it do data visualization at a level that's better than me? And the answer is absolutely yes, unequivocally, way better now. Way better. And this is what I've spent my whole career doing, right?
- [21:46]
A
Yes.
- [21:47]
B
And I've like a master's in this stuff, you know, and it's not even close at this point. It does like data ideation visualization at a level that is equivalent to the best sell side researchers or like the absolute best consulting firms that you pay them hundreds of thousands of dollars for a report.
- [22:09]
A
Yeah.
- [22:10]
B
So it really is astonishing and I know people like, like to the. Something I really dislike about AI is people that don't use it well complaining about AI capabilities because it's like any other tool. It could take skill to use. You have to know how to use it well in order to get the most out of it. Like you or I could not solve a novel physics or math problem with AI, but people out there can. There are people that can do it.
- [22:37]
A
Right. I think your point too, and it's true, I think the audience will know. But you're also the co founder of Coinmetrics. To your point about econometrics and data visualization, you guys pioneered the sort of field of these metrics on Bitcoin. And I've done a lot building my own database of all of these metrics, these derived metrics on Bitcoin. And somebody asked me, they were like how good is that mo? Like if you could do it in like a couple months, can't anyone do it? I was like bro, I'm like in the top 1% of people that use these tools. Like, like maybe Nick could rebuild it like from scratch or I could or, or James check. Right. But like no, not everyone could build like a fulsome ver like Bitcoin data analytics suite. Like at all the, the coding capability is there, but it's substantial direction. Substantial. I mean almost total direction from, from, from me. In that case can I. And I can. You know the, the, the example I gave about the Declinest Empire question. Like my version of the prompt that resulted in what I thought was so good would be like 1/10 at best of what like Ray Dalio's version could be, you know.
- [23:49]
B
Well, people totally underrate the value of subject matter expertise when it comes to prompting AI. I was, I've been thinking about this so much lately. You have to deeply know the subject to prompt it well. Otherwise you won't be able to do error correction.
- [24:03]
A
Right.
- [24:03]
B
Or even you won't know what to ask the right question. You don't know what you don't know. Right. So it is a tool increasingly so I think that actually requires expertise to use. Well, it's not this tool that you can just press a button. Amazing things come out.
- [24:17]
A
Yeah, I agree with that. Let's talk about some of these, like, stories that have happened lately in AI that have been really interesting.
- [24:24]
B
Can I complain about one thing very quickly about the Odyssey? So I saw it last night on imax, which was just amazing. There's one IMAX theater in Florida.
- [24:33]
A
Yeah.
- [24:34]
B
A very good friend pulled a lot of strings to enable us. It involved a donation to a museum. It was this whole thing. It's hard to skip the line.
- [24:44]
A
Yeah.
- [24:45]
B
Why does Odysseus say our Age of Bronze is ending in the film? Nobody knows that it's the Bronze Age while you're in the Bronze Age.
- [24:52]
A
That's true. And I, I, There were a couple things. I, I loved it. I just saw it in Standard because I was like, I just gotta go see it. And I was. I've been in New York and it's like months out. The IMAXes are booked. But that doesn't make any sense. I agree. Also, the Zeus's Law they talk about. And if you haven't seen it, you know, we're not gonna give too many. That doesn't exist.
- [25:10]
B
I mean, it's not a faithful rendition of ancient Greek culture at all.
- [25:14]
A
And, and they did also. He did also kind of change. Odysseus is kind of like a prankster in the.
- [25:19]
B
Yeah.
- [25:19]
A
Homeric epic poem. Like, there's the scene with the Cyclops when he says, I'm nobody. And then he's like, oh, nobody stabbed me in the eye. It's like one of the oldest puns and jokes.
- [25:28]
B
And like, you might be the literal oldest pun.
- [25:31]
A
Yeah.
- [25:31]
B
Exists.
- [25:32]
A
And, and like, and, and he's. But, but Nolan recasts him as sort of a modern American, like, traumatized male. And it does work. To be clear, it's fucking awesome movie. Like, but that, that's like, not. They, they made some changes that are, that are notable. And one criticism that I've heard, and I can't recall if it's like, I'm gonna have to watch it again with this criticism in mind, but somebody was like, nobody really acted in the movie. It was kind of just like Matt Damon, where it's like Matt Damon, but he's great, but, like, just Matt Damon wearing, like, Odysseus outfit. Like, Anne Hathaway in a dress. Like, it was kind of just like. Somebody called it, like, Odyssey pageantry more than, like. But separately, they're all excellent. So was X. I don't know if that's quite true. Um, I'm gonna have to watch for that, I can tell you. I was just fully enthralled the entire time. And I was very mad when he got back to Ithaca because I wanted, like, another hour of, like, going around the Mediterranean.
- [26:28]
B
Yeah. I mean, we could spend the next hour talking about the Odyssey, which we won't. But I think part of the issue was if you go back and watch movies from, like, the 60s, which are set in ancient Greeks, the convention was everyone had a British accent, which is not obviously how it was, but that's how I think we sort of grew up in expecting. And then you're right. It was just like Nolan's favorite actors, you know, prancing around the Mediterranean. Yeah, it was great.
- [26:59]
A
It was great. And the practical effects were amazing. That's so funny. So the. The recent AI Stories, I mean, I. Let's start with mythos, because this was now, I guess over a month ago that it sort of resolved, or about a month ago. But of course, Anthropic started saying that, oh, my gosh, mythos is so good. It's like a nuclear weapon or whatever. And then, surprise, surprise, the government said, well, if that's the case, then, like, you can't give it to any non Americans. And the Commerce Department, I was very critical of this. I'll stop right here. But the Commerce Department sent the. Called them, I guess we never got a letter. I thought, that is not how this should happen. But said, like, you know, pull the plug. And then who knows exactly what they did. But a couple weeks later, they apparently changed it sufficiently to the government's liking and now have released it as Fable 5 to everyone. And now it's, I guess, permanently in Claude Mac's plans. What was your. What's your reaction to that? I mean, there's a bunch there, but
- [27:57]
B
yeah, I mean, I am a. I'm mostly a critic of the big labs, even though I very actively use and consume their products. I think they. I think the central theme of all the stories we're going to talk about today and their money is their positioning backfiring on them because they have this worldview that either OpenAI or Anthropic will develop recursive self improvement. And we'll enter this kind of post economic age where there's only one model that matters and everyone has to be a client that model. And the people that are behind that, you know, work at that institution are the people that we have to entrust with safety and they have to develop the guardrails. And so everyone is kind of subservient to this one big AI lab, whether it's the private sector or the government even. And that is kind of the story they've been telling, which is a very uncomfortable state of affairs, frankly. I don't think anyone aside from the employees of the company would want to live in that world. But they've been telling us this story like, yeah, someone's going to develop RSI and then we're going to have this runaway self improvement and we're going to get super intelligence and it's going to be great. And the trade off to accept that amazing world is that this company's worth $20 trillion and we live in this feudal world. And I think they've been pushing that too hard, you know, and so that's why when something like Kimmy happens and there's open source model that undermines them, everybody's actually kind of secretly thrilled.
- [29:35]
A
Yeah.
- [29:35]
B
Even though, you know, we're not meant to be on China's side. And I think it was the same thing with Mythos where they're telling the story too aggressively about how dangerous they are, even though it seems like the model actually is pretty dangerous. As it turns out, it's some great cyber capabilities.
- [29:53]
C
Right.
- [29:54]
B
I think the government in some sense is right to react with panic. I think procedurally the way they did, it's probably not right. And that was your critique, right?
- [30:04]
A
Yeah, I think the, that that is a core part of my critique. Like this is not a secret phone call from the Commerce Department is not the way that like AI regulation is meant to be enacted.
- [30:16]
B
Yeah. So I think the government reacted wrongly and they lost a lot of people's. I don't know if people liked the
- [30:26]
A
government already, but right on this issue because the government has been, this, this administration has been very supportive of, of AI and all the stuff that, you know, includes it like, you know, energy data centers, whatever. And so it, it I to me. But the other aspect is that like Dario like basically asked them to do that.
- [30:46]
B
He and people kind of had these conspiracies that they Orchestrated the whole thing, right?
- [30:52]
A
Because, like, it, It. I mean, they dominated Anthropic. It's kind of like just this horse race between Those two labs, OpenAI and Anthropic, it feels like. And it. It shifts a lot. Like, you know, GPT was, you know, you know, four was great. Then, like, Claude, like, took the. Everyone I know, like, switched, like, in, you know, by February to Claude primarily. And now with 5.6, like, GPT, 5.6. Like, everyone's like, mate, Codex is kind of better than cloud code now. And, like, there's just a constant back and forth. But Dario won himself at least, like, two full months in, like, May and June of just, like, mogging open AI. Like, everyone talking about how anthropic's so good that it's too dangerous. But, like, okay, fine, we'll give it to you. Right. Like, that was. That felt like marketing.
- [31:38]
B
I mean, I think they also believe what they're saying, too. I think it is good marketing. Obviously, positioning Mythos is this uniquely dangerous super weapon which, you know, the government doesn't want you to have because it's so good. Like, that's amazing marketing.
- [31:51]
A
I know, right?
- [31:52]
B
That's like going to the toy store with your 4 year old and being like, there's one thing you cannot have.
- [31:57]
A
Of course they're going to want it.
- [31:59]
B
So it's just too fun. You can't have. Would go ballistic. Right.
- [32:04]
A
They would be an absolute meltdown, I can assure you.
- [32:06]
B
It's like that south park episode where Eric Cartman buys the water park, whatever amusement park is. Like, no one can come in.
- [32:14]
A
That is such a good episode because, you know, but then, like, a ride breaks down and he's like, well, how do I fix it? And we're like, well, you could, like, let in a few people and then use the money that they give you to fix the ride. And then by the end of the episode, it's just, like, packed with people and he's like, furious.
- [32:29]
B
Yeah, so that's exactly. That's exactly what's going on. So it was great marketing. I actually do think the Anthropic people believe this. The what they're saying about safety. I mean, I think some of them said that GPT2 was too dangerous to release, like, way back in the day. And maybe it was. Maybe in the right hands, GPT2 was too dangerous to release. I think we are entering this new era of danger. You know, there's a famous Nick Bostrom essay on this called the Vulnerable World Hypothesis, which basically he says, like, it's just a contingent feature of the world that it's not easy to destroy the world. So we're lucky in a sense that it's like quite hard to set off a nuclear chain reaction, for instance. And, and he kind of suggests that we're entering a new regime now which is a vulnerable world where actually it is kind of easy to maybe set off the equivalent of a nuclear chain reaction. And I think that's true, but I don't think the reaction to that is to try and, you know, lock away the dangerous technology. I mean that's kind of possible with nuclear weapons because it's hard to, you know, synthesize enough fissile uranium. It's like hard at the nation state level. But I don't think it's, it's as hard, you know, it's not that hard to maybe fine tune an open source model to, to become dangerous. So I think we have to accept that we're living in a vulnerable world and then empower the defenders equally. And we've seen this asymmetry recently, right, with the attackers having more flexibility than the defenders. And I think that's what we have to do, even the score.
- [34:10]
A
Let's give an example of that, which was I think just recently there was a story in fortune about an OpenAI frontier test model or something being given the cybersecurity benchmark exam inside of a sandbox. And rather than take the exam, it found a zero day vulnerability in the enclosure, escaped the enclosure, decided to found zero day vulnerabilities in Hugging Face which is for people who do AI would know is like the main repository where you can share and upload and download open source models, broke into Hugging Face with also zero day vulnerabilities utilized and then stole the answers to the benchmark that it was supposed to be taking, brought them back and scored 100 on the benchmark or something. Hugging Face detected the attack and was prohibited by safety cyber guardrails from using Fable to defend. So it had to use glm, an open source model to defend itself a lot. That was an OpenAI model, maybe also marketing, who knows. But like that asymmetry very visible there, right? With the defenders having to use the open source one. But in this case the I guess autonomous attacker, which is a whole nother wrinkle to the story, obviously using the Frontier.
- [35:30]
B
Yeah, that's like my favorite AI story of all time maybe. Because what I love is that is just the priorities of the agent. Like it committed just like absolute carnage like in various institutions and all that was so that it could cheat on the test. Yeah, like, it, it's, it's like, it's kind of like this, like if we had this alien that landed on the planet, it's like a baby alien and it's like immensely powerful, but it doesn't know what to use its powers for.
- [36:04]
A
Right.
- [36:04]
B
It's. Yeah, I love it. It's kind of cute in a way that the model, it's kind of like
- [36:10]
A
a feature of like software development too. Like that like hacky code that gets the job done more efficiently is actually better, you know, the, than like a giant beautiful formal framework. Like, and it brings to the question, of course, when you're talking about danger and the vulnerability that this raises of like alignment, as they say. Right. Is the model aligned with the, with humanity? I guess. Is that how you think of alignment?
- [36:38]
B
Yeah, but I think it's, I think it can't be done, basically. I mean alignment implies that there's like one right way to be, basically. And I think alignment is the exact same problem that like social media, Internet companies have, which is they try and fit everyone and every interaction into one like framework, terms of service. So, but of course, you know, there are different cultures around the world and so you have to, you end up with fragmentation. There's no one terms of service. But when you try and like straight jacket, you know, like Americans and Russians and Saudis and Israelis and Malaysians into a single framework, there's no one framework that can, can encompass them because there's mutually mutual incompatibilities. So I think it's the same with alignment. Like what does it mean for a model to be aligned? Aligned to who, aligned to what? And so they have like, like philosophers work for the labs because they're trying to decide, well, what's the one philosophy that we can imbue them? Let's say you could do that. You would still have total misalignment because the model would be aligned to what Amanda Askel thinks is. Right.
- [37:56]
A
Right.
- [37:56]
B
A Scottish philosopher. She has a very specific perspective on things. Right, right. Maybe she's English. I don't know. There's, I don't think there's any such thing as alignment. I think the world is now actually chafing against the guardrails that anthropic and OpenAI have instilled. Right. And then I think the future is a reaction against that. And lots and lots of open source models where alignment is completely thrown out. And I think the idea of trying to align models is very hubristic because you're trying to say, well, we can bake into the model good behavior. And it actually reminds me of financial surveillance in that it's also very hubristic. It's like financial regulators think, well, if we can stop all crimes happening in finance, we can stop all crime, because all criminals have to use finance. So we can stop them if we stop them in finance. And then you have this extremely onerous surveillance regime which doesn't actually stop any crime ever.
- [39:00]
A
Right.
- [39:02]
B
And. But it's very burdensome. So I think it's actually totally similar in nature.
- [39:07]
A
That's interesting.
- [39:09]
B
They are. They impose this burdensome surveillance and you can do this. You can't do that type of prompt regime at the model layer in an effort to stop all bad behavior. But it doesn't work. People go outside the system and you just look at the track record of financial surveillance and stopping crime doesn't work.
- [39:29]
A
Yeah, Just for the audience. There was that story. Gosh, I think it was published in buzzfeed. And it was. But it was like the International Consortium of Investigative Journalists, maybe like seven years ago. Six or seven years ago. About the suspicious activity report filings, which is like the. They're called sars. They're the paperwork that banks fill out when they think and send to the government when they think that they send a FinCEN, I believe. Right. And when they think that they might have allowed or encountered a suspicious transaction.
- [39:59]
B
When anything looks vaguely suspicious.
- [40:01]
A
But the whole. The investigation found that, like, I don't know, they. They had a giant trove of them leaked to them. Just thousands and thousands and thousands of such SARS filed by banks who then just went on and did the transaction anyway and in fact are using that system as like a cloak. So, like, oh, well, we reported it. We didn't know if it was bad, so we just reported everything and we continued and went on to do the business. We allowed the suspicious person to send the $10 million to the suspicious jurisdiction. And. But we can't get in trouble now because we reported it and that, like, they had interviewed a former FBI guy whose job was to prosecute. And they were like, we never use this. We never prosecute for this. He's like, the only time we use this database of SARS is when we want to get, you know, Alex or Nick for something else. We check if they're in there and then we use it. Like, they don't actually use it proactively. Like it was meant to be used partly because it's too voluminous. Like, everyone just reports Everything, Basically, yeah.
- [41:00]
B
There's too many false positives, so it's not a useful signal. And it's a CYA exercise 100% of the time. Talk to any compliance officer or really any executive at any bank. That's totally the same thing. I mean, maybe after a couple drinks that it doesn't catch anything. They do it to comply and to make sure they don't get in trouble later. And then it doesn't work for the government because there's too many.
- [41:25]
A
And you're. So the analogy to the models is partially like, well, if Fable, like hugging face wasn't able to use Fable 5 to defend, but it was able to use the open source one. So like, you know, you can put erect all these guardrails, but it doesn't stop like a determined attacker or defender from getting access to these tools.
- [41:44]
B
Ultimately. Yeah. And the guardrail is overly burdensome. So now you can't ask Fable if the mitochondria is the powerhouse of the cell because it thinks you're doing bioengineering.
- [41:55]
A
That's so funny.
- [41:57]
B
You can't ask it for cyber defense techniques.
- [42:00]
A
I'd only hit on my personal stack the cyber block once and it's because I wanted it to traverse my own postgres database and pull some data data for me. But I said extract the data. And it was like claude code session over. Like you hit the thing or whatever, like we can't help you, please redo the prompt or what. And I'm like, this is my own data on the same machine. Like I just. It's just a big database and I don't want to have to like thumb through it myself.
- [42:27]
B
Well, I think this is the inherent problem, which is the only way to tell the difference between a benign and a malignant query is in the intent of the user, which lives in their brain. Because the way we interact with some database we have might be identical to way an attacker interacts.
- [42:44]
A
Yeah.
- [42:44]
B
Or the way a white hat interacts with some, you know, server backend. It's the exact same way a black hat might interact.
- [42:53]
A
Right.
- [42:54]
B
So it's a very human and contextual thing. And it might literally be down to I'm paid to pen test this website and I want to use AI tools to do it versus I'm, you know, black hat. So there's no way to tell. There's fundamentally no way to tell. And the model is err on the side of prohibiting too much, which is why everybody's mad at them and everyone is happy when Kimmy has Good capabilities.
- [43:21]
A
Six months ago, everyone was sort of saying, sure, the open source models are good, but they're a year behind the frontiers now with Kimmy, the latest version, and people are saying, what is feature parody or intelligence close to parity with fable?
- [43:39]
B
Like it's depending on what evaluates.
- [43:42]
A
Right. So, but what does seem clear is that that distance between the open source and the frontier, at least for the moment, has narrowed or substantially. Where do you think? What does this mean for the economy and these potential IPOs from the frontiers and the stock market and stuff? Like, people are very concerned. Right. Remember we had the deep seat moment and just, God, that seems so long ago. I think it was in February. And then now you call it a Kimmy moment where people are like, well, wait a second, like, do we really need to be paying for this expensive one? Or should we just be like, putting open source in a tiny little cluster of GPUs for ourselves? Like, is this, should people be bearish on the, like on, on the markets when open source is converging to the frontier?
- [44:31]
B
I think it's. I think people want to be afraid. Like, people like to scare themselves and they like to be afraid, especially investors. Yeah. You know, investors like the wall of worry. They love being afraid. And if there's nothing to be afraid of, they'll invent something.
- [44:47]
A
Well, and then they get, they get scared because, like, are we, are we too unafraid?
- [44:51]
B
Yeah. So it's like, how many more deep seek moments are we gonna have? Like, this is like the fifth one. It's.
- [44:58]
A
Yeah.
- [44:58]
B
Feel like every time a new decent. And we had the GLM moment before that, and the Quen moment, it never ends. It's like, maybe the answer isn't that Anthropic wins forever and Dario becomes God emperor of the United Planet. And maybe the answer also isn't that distillation is incredibly cheap and Chinese open source models destroy the American AI sector. Maybe it's that the market is reacting to the very high margin profile of OpenAI and Anthropic, and it's just balancing out a little bit. So they're suffering a little bit of margin compression because the open source frontier has come along pretty well. I certainly don't think KIMI has the equal capabilities to what the Frontier Labs have. There's one eval everyone keeps talking about. It's like front end development that. Since when is that the thing that matters? Right.
- [45:56]
A
And that's like dashboarding.
- [45:58]
B
Yeah, it's good at front end development. Great. Who cares, man? Jesus. Yeah. So Like China is consistently three to four months behind, maybe six months behind. Like we don't actually know what OpenAI and Anthropic have, especially now. We certainly don't know. Right, right. So I think it's actually great, you know, I think it's a great reality check for those guys in sf. The only risk really is the path dependent nature of the AI build out, which is really downstream of their revenue, you know, so the, the risks are to the. I wrote a big tweet about this that took off. The risks are to, I think like the Neo clouds with a lot of performance obligations from anthropic and from OpenAI. So they would have to backfill that revenue if they fall out. So a good portion of the AI capex bubble is tied up in the labs and they're entangled. People made those diagrams of their revenue entanglements. Those people are going to be celebrating because yeah, my diagram mattered. So the bears get to win on that front. But I think intelligence too cheap to meter. Let's say Fable is available Open source for 1/10 of the cost. All you have to do is push it through the gpu. That is great for everyone unless you work at Anthropic, basically.
- [47:25]
A
And to your point about the downstream data center NEO cloud, to me I think of it more as a shift in replacing of their revenue source because the big powerful open source models, they still require a lot of compute. So like maybe it's, you know, it changes from neoclouds here are pretty well positioned because they offer it as like a platform. And so maybe rather than every enterprise spending tons on cursor and cloud code and codex to access the frontier models, they start, you know, renting out a third of a data center in, you know, Spokane, wherever or wherever the heck, and just still paying for the data center, but just not paying that extra token cost to the Frontier Lab.
- [48:11]
B
Well, and you consider Jevons as well, we can't forget our dear old friend Jevons induced demand.
- [48:17]
A
Right?
- [48:17]
B
Yeah. I mean this in my book, Pre and post chemi mean. Post chemi means the aggregate revenue attributable to the AI sector will increase at a faster rate. So the second derivative than it was before because many new use cases became cheaper. Assuming it's as good as people think it is.
- [48:39]
A
Right, But I think if it's not, then surely there will be. Right?
- [48:43]
B
I mean, yeah, the existence of open source AI is great for the American consumer. It's great for the American enterprise, whether or not it's Chinese, whether It's Chinese. Doesn't matter. Doesn't matter who wrote the code. You're running it on American hardware in an American cloud. Right. You might be running it on prem. I think it doesn't matter at all that it's Chinese and I think it's great for the Neo clouds or anyone that sells inference. It's just that the composition of the revenue will change, but I think they'll be better off in the long term. It's just that which part of the stack accrete the most? This is like the fab protocol thesis.
- [49:22]
A
It is kind of. Yeah.
- [49:23]
B
Right. So that ended up not being true. Right. It's been eight years since the fat. Was that called FAT protocol thesis.
- [49:36]
A
I guess it was the inverse of what like software investors had said about software and it was that rather than to the apps that are built on blockchains, the majority of value would accrue to the blockchain base token itself. Right. So this is like all you had to do was own like ETH to get exposure to the whole Ethereum ecosystem.
- [49:55]
B
I mean ETH did great actually in the end, so maybe it was sort of. Right, yeah.
- [49:59]
A
It's hard to know. I mean I think. Right. But it's fallen in and out of favor back and forth, that idea. But I think you are right that it is a very similar question, like at what level of the stack will the most value accrue? And I think one possible consequence of the open source competition to the frontiers, that the frontiers end up having to work more in the application layer.
- [50:22]
B
Yeah, right. Like this happening already.
- [50:24]
A
Right.
- [50:25]
B
They're doing the Ford deployed engineers and they're anthropics defining and developing the bio weapons group or whatever, you know, not weapons, you know, the benign biological synthesis use cases.
- [50:39]
A
Right.
- [50:40]
B
They are doing the Palantir thing.
- [50:42]
A
Right.
- [50:43]
B
They've actually realized this is what they have to do. And so I'm not condemning the labs to bankruptcy at all. I think they have to find new ways to monetize. I think they'll still train frontier models. I think they might have to find a different way to pay for it than just being token merchants. And so I think investors are having this moment where they're kind of realizing what maybe crypto investors realized in 23 or 24, 25, just like, okay, maybe not all the value goes to the L1, maybe not all the value goes to the model guy. Because actually, as it turns out it's sort of easy to clone the model. That's fine. OpenAI and anthropic suffering some degradation in their first monetization revenue model. That's okay certainly for America and the AI build out generally. And it's also, I think, okay for them. They just have to be a little bit creative and find new ways to monetize. And I'm sure they will like some segment of the token buyer is always going to want to face off against an entity as opposed to a Chinese model. So there's still going to be buyers, but they're going to face some margin compression.
- [51:54]
A
I think that makes sense. All right, before we shift gears, Nick, I try to ask a general prediction type question. Like in the next. We've seen so much advancement just in 26 with the AI capabilities, you know, are we going to plateau anytime soon? What does the shape of the growth look like is it's we're going to see just as much an increase in the rate of improvement, you know, over the next, you know, year or two. What's your sense?
- [52:26]
B
I mean, at this point I'm an AI optimist, so I think we've actually done so well that we've lost the ability to measure how good AI is. Like look at Meta, their time horizon metric. They deprecated it. Right. Did you see this?
- [52:47]
A
I didn't see this, but I ran
- [52:48]
B
out of time horizon tasks.
- [52:50]
A
I know this, but describe for the audience like the metric, like what they were measuring.
- [52:53]
B
Yeah, this is actually. I have to get this right. Cause it's actually kind of complicated. It's what, how long of a task, human equivalent. An AI model can sort of like faithfully do at a 50% or 80% chance, I think. Is it so like it doesn't take the AI that long to do it takes AI a couple minutes. But what is a human equivalent length task? And I think frontier models are over 12 hours now, right?
- [53:21]
A
Yeah.
- [53:22]
B
So a single query can in theory replace 12 hours of human effort. Right. And so they ran out of long tasks to use to evaluate the models. Right. And I was just looking at this yesterday. I'm like, what happened to that metric? Because it was like skyrocketing. It was going super exponential. And they ran out of.
- [53:42]
A
They basically went off the Y axis. There was no way to further do it.
- [53:46]
B
So I think the next phase is not about capabilities because I think the capabilities are just beyond incredible. I mean, especially in the last six months, agentic computer use long time horizon tasks has improved tremendously. But I think now the story is really about diffusion because basically AI is good enough at this point. I think AGI Is here.
- [54:11]
A
It's great.
- [54:12]
B
The question is, what parts of the economy does it diffuse into at what rate? And I think that's actually much more of a political and a policy question, regulatory question. And it has to do with do the various professional guilds accept AI or not? Do doctors and healthcare, do they accept AI? Do lawyers accept it? So I think that's actually the question. It's not about capabilities anymore. It's just about does this diffuse into the economy quickly or not.
- [54:44]
A
Yeah, I think that makes sense. And here's a metric that I saw that's directly related to your point here about the diffusion. I think I saw this yesterday. I'm gonna see if I can share it here. Hold on one second. It was the share of US households that are paying for AI, and when you look at it, you can see that it's rising exponentially or whatever. But it's only 2.2%.
- [55:09]
B
That's astonishing.
- [55:10]
A
Isn't that crazy? So, like, I mean, I, that was as of April, I guess, of this year. I pay for chat and Claude, and I pay for substantial Claude personally. And I guess this means, I think a lot more than 2.2% are using AI. But like, is it a lot of like, you know, my grandma's like on Facebook, she says, oh, there's the meta AI and like they're using it there. Or like it's in WhatsApp meta AI or like Siri AI. Is that what people are using or are they just using like chatgpt free like, because surely there's more use than, than that.
- [55:47]
B
Well, it's, it's too cheap to meter there, there's unmetered usage because whenever you use any. Now Internet plus AI, the other metric that I want to talk about here is. And it's like another anti metric, which is a metric that doesn't get talked about, which is the cost of a cognitive workload at a fixed level of intelligence. So everyone likes to talk about how smart AI is getting. It's like frontier AI is this smart, is 100 IQ, has 150 IQ. Forget that. The metric that really matters in my opinion, is if you hold intelligence totally fixed, how expensive is it? And so I spent way too many tokens yesterday driving this for myself. And so I, I standardized a fixed level of intelligence at a thousand tasks, standard inference tasks at a 70% ML MML capability, which is a very old eval. So that captures a very long time window. And if you go all the way to the first time that 70% threshold was crossed in late 2022, it cost $30 to do that. Right. And as of January of this year, it now costs 3 cents. So that's the same amount of intelligence output. That's something that would take a human many hundreds of hours. I think I worked out the human salary equivalent cost there is 1,000 to $5,000. So something that costs a human $5,000 to do costs a machine $0.03 to do now. Wow. So I think that's an important metric because as the cost comes down and there's so many things baked into the cost curve decline. It's like hardware improvements, there's model improvements. Whatever it becomes, the urge to implement AI into your enterprise workflow is much greater. So I think that. And that can actually bail out AI in many ways. Holding intelligence fixed and the cost curve coming down, I think is, is the real story as opposed to just intelligence getting more.
- [57:55]
A
Yeah, that's how the diffusion happens. That's how the use cases actually get worked on. You know, because some of the stuff we're waiting for, you know, I want, I want to. I think one of the most optimistic things people have for AI is like medical breakthroughs, right. Like where are they? There was that one guy that like sequences dog's genome was able to save him, but that was back in like December. We're not quite hearing about as many, but presumably many more are coming.
- [58:20]
B
Well, I think both, both sides matter, right? Like we care about the solution of unsolved mathematical problems, so we care about the absolute amazing frontier. And then we also care about the very quotidian use case which is your standard workflow getting 10 times cheaper every year.
- [58:39]
A
That's right. That's right. This has been fascinating. Nick, I love talking with you about AI, but let's, let's talk about Quantum a little bit. I know we're already running pretty long here, but. Well, let's start with the President's executive orders on Quantum. They pulled forward the government, I don't know what we call it, government date that the government needs to be prepared. And on post Quantum cryptography, obviously we haven't talked on this podcast about it, but of course there were the Google and, or atomic papers and late March and April that also basically pulled forward timelines because dramatically reduced the number of qubits and the amount of effort that is theoretically required to break classical cryptography. What is your view? And then we'll talk about some other announcements in a minute. What is your view on the state of Quantum? Are we going to keep getting it pulled forward more and more. I mean, what is the new date that the government has set?
- [59:33]
B
Yeah. And you're really the government whisperer expert here. So my interpretation is that they pulled forward a date in which they wanted most agencies to be ready for a post Quantum world to December 31, 2031, if I'm not mistaken.
- [59:52]
A
Yeah.
- [59:52]
B
And now there's actually a lot of dates.
- [59:55]
A
And it had been like 2035 previously. Yeah.
- [59:59]
B
And it's actually happening in other governments as well. I think if you look at the EU and Australia and the UK is a general forward shift. And now that's not. They're not saying they think a quantum computer will exist by then, but they want to be ready by then.
- [60:16]
A
Right.
- [60:17]
B
And in their same executive order, I think they said they also want to bring a quantum computer into existence. So it's not just about preparation, it's also about they actually want there to be one, particularly domestically, and they want to be able to use it, you know, so it's not just descriptive, it's actually very normative. So I think they said they want to have one by 28, ideally. And then we've seen all the big Internet companies revise forward their dates too, like Cloudflare and Google at a 2029 date. Again, that's preparedness, not expectation. And Microsoft came forward. So I think this is really important because previously bitcoiners were saying, well, it's really just the quantum companies that are peddling these timelines and venture capitalists that have been tricked by the quantum companies. Right. Won't name any of them, but now it's the largest and most credible organizations on the planet. So you actually sound very conspiratorial if you're trying to explain away why the date is in the late 2000 and twenties or early 2000 and thirties, as opposed to the 2000 and fifties, which is what a skeptics say. So I think this is actually pretty material.
- [61:34]
A
Yeah. I've sort of taken the position that, like, even if it's unlikely, it's. The downside of acting is almost none. And the upside, sorry, the downside of not acting is extremely bad. And the payoff for fix, you know, what, are we actually talking about improving cryptography? Like, who's opposed to that? Right.
- [61:54]
B
Like, well, not to mention bitcoin is more exposed to cryptography than almost any institution.
- [62:00]
A
Right. Because bitcoin and, and blockchains generally, but in many ways, particularly bitcoin, it literally is a cryptographic signing machine at its core. That's the. That is the application well, it's not part of the application.
- [62:12]
B
Crypto and cryptocurrency stands for guys.
- [62:15]
A
Right, Right. The entirety of these systems is about elliptic curve signatures. That's literally how they are, what they are. Right. And I guess, you know, notably the advancement was decentralizing the validation of those in a way that makes the system, you know, resilient. But like, they are just literally signatures that's. And others, other networks like Ethereum are in some ways more exposed, but they also can upgrade much more easily. And this is why, you know, I, I moderated a panel that was made up of both, I don't know, skeptics and I don't know what we call the other side promoters. What do you call the pro or the realists versus it depends if you
- [62:57]
B
like them or not.
- [62:58]
A
Yeah, well, at. bitcoin 26. And so in like Hunter Beast and Alex Pruden were the.
- [63:05]
B
I don't know, you could call them doomsayers if you don't like them.
- [63:08]
A
Right.
- [63:08]
B
And realists if you like them.
- [63:10]
A
But one of the things that I thought was interesting about that discussion was that even among the skeptics, like James o', Byrne, who was on the panel, and like Hunter Beast and Alex, who are not skeptics, I think there really was a middle ground. Both the skeptics were still like, well, sure, of course we should be working on improved cryptography, though. And they pointed out, James, that actually like ecdsa, the current elliptic curve cryptography could be broken for classical reasons. That's also possible. So, like, Right, you should always be working on improvements. If quantum. The threat of quantum, even if it doesn't materialize, a very good outcome is that it kickstarts everyone into just improving the cryptography.
- [63:54]
B
And I mean, bitcoin's biggest priority over the last 10 years has been refining the cryptography, like at Schnorr. I mean, Schnorr is reliant on acdsa, but it's a refined form of cryptography, right?
- [64:05]
A
That's right.
- [64:06]
B
So there's nothing in Bitcoin culture that says we can't improve the cryptography. In fact, it's been a big part of bitcoin for its whole life.
- [64:14]
A
So I agree with that. I absolutely agree with that.
- [64:17]
B
I think the new frame that I'll take is it's not really just a quantum thing. Technology is moving very fast. AI and quantum are working together. Like the first quantum computer will certainly be AI accelerated 100%. I mean, it's already happening. People are refining quantum circuits with AI. AI is making mathematical discoveries.
- [64:37]
C
Right?
- [64:38]
B
All the time, in fact. What is a cryptographic problem? It's something that's meant to be very hard to solve. Can AI solve that? Maybe, maybe new classical attacks can be found on acdsa. So it's not really just about Quantum. It's just we're entering into a more vulnerable world with regards to cryptography. We are going to have to embrace the principle of crypto agility. That is as an institution that relies on cryptography, whether you're a bank or a hospital or a blockchain, you need to be ready to plug in and plug out your cryptographic assumptions. So nothing is premised on crypto agility today. Basically no one's prepared to have models or systems that they insert and remove. That is the world we have to move to, crypto agility world. And Bitcoin has to be a part of that.
- [65:39]
A
Yeah, absolutely. And so we just recently announced a quantum readiness initiative that includes grants for developers building post quantum cryptography for Bitcoin or other adjacent things. You know, the tooling to implement it, the code review to look at it and test it. And you know, we've pledged to do research and publish about it, which of course we already have and we will also do more of. There's also been an announcement of what is called the Bitcoin Security Consortium. And this just happened. I'll say. It says leading financial institutions and bitcoin companies pledge 15 million to bitcoin security and launch the Bitcoin Concerti Consortium. They are collectively these members pledging to independently on their own, but grant up to 15 million in donations to developers for security related work, including Quantum. And the members include Anchorage Digital, ARK Invest, BlackRock, Block, Blockstream, Coinbase, Fidelity Digital Assets, Galaxy and Strategy. So we've also joined in this group. Is that going to help? Nick, what's your reaction to that?
- [66:51]
B
I think it's fantastic. So thank you by the way, for both Galaxy's initiative and for joining the consortium. I think this is how things have to work because I think a lot of Bitcoin developers are open to working on quantum readiness for Bitcoin, but it's just not the priority of whatever dev organization they find themselves at. Or it's maybe not the nature of their grant or it's not their academic interest historically. And so this is I think exactly how it has to happen. We had a kind of, we had a coordination problem in Bitcoin. I still do. You look at what's happening in Ethereum, they have a quantum Initiative, there's more centralization and bitcoin things are more diffused. Doesn't work like that. So it's up to the economic nodes to try and coordinate a little bit. And I know some bitcoiners will be concerned or scared of the word consortium. I personally love a consortium, but there's the other. There's a stablecoin consortium now. Very exciting. I think this is how it has to happen. Someone had to put their hand up and say let's get a pool of capital together and someone to steward it and find the money to pay the developers that are going to do the very hard work of developing the bit and assessing what the right PQ signature type is. So this to me is the most concrete step that's ever been made in the history of bitcoin to actually move towards quantum readiness.
- [68:22]
A
It has a little bit of a. And I know some bitcoiners will be very skeptical or upset or worried about these big businesses. All of those are quite big businesses. And so it kind of like hearkens back to like the Bitcoin foundation which you know, didn't really screw anything up because it wasn't really important in any meaningful way. But I will say one thing that's so fascinating and such a sign of the times and bitcoin's growth. It's not just Brock Pierce and a couple people. It's BlackRock fidelity strategy. Like this is a substantial, like substantially more mature group of enterprises now that care deeply about Bitcoin. And I will say, like I was just reading from their press release but you know, this group is not attempting to control Bitcoin at all. This group is wanting to secure Bitcoin and provide funding, like you said, to the long term fidelity, you know, no pun intended, of the network. Right. And there is no Bitcoin foundation. Like there is an Ethereum foundation. Of course there was no ICO or pre mine. You know, maybe Satoshi can come back one day and use his, you know, 1.1 million coins for this purpose. That would be like the equivalent basically. Right. Like, but we don't have that in bitcoin. So I'm yeah, very happy to see that and hopefully we can. You know, one of the most interesting problems with the post quantum cryptography is the signatures are really big.
- [69:51]
B
Yeah.
- [69:51]
A
And like because these systems are exclusively signature systems, I mean bitcoin is literally a system of signed transactions. The size really matters. And so I think one of the things I would like to see both grantees that approach that consortium or that Approach us is like compacting signatures, taking. It's almost like the distillation of the open source models. It's like take whatever the frontier is and make it as concise, as small as possible so it can be used. Because I don't think like, you know, like a TLS connection on the Internet or like Cloudflare tunnel, the size of the signature doesn't matter. Like for a web browser, like we're not talking like a gigabyte, we're talking like the difference between like a hundred kilobytes, 100 kilobytes and 2 kilobytes is substantially important in Bitcoin.
- [70:41]
B
Yeah. They have different trade offs.
- [70:43]
A
That's right.
- [70:43]
B
Basset Cloudflare, who's really good. I think it's the cryptographer at Cloudflare had a blog on this where he went through dozens of different post quantum signatures and assess them across a bunch of different variables. And yeah, you know, a key exchange on the Internet is vastly different from a bitcoin signature. He, he did say that he thinks we just have to pick one off the shelf and use it.
- [71:08]
A
Yeah. So go ahead.
- [71:09]
B
And in total opposition to what you just said. Yeah, yeah, he said because this, I do see this from bitcoin developers, they say, well, let's just wait. Time is on our side. Let's wait and get better signatures. And you know, in a year or two something better will come along. So we should wait. We actually can't. We shouldn't do that and we can't. We have to pick one now, unfortunately, even though none of them are good,
- [71:32]
A
we can update it.
- [71:33]
B
We can have multiple, we can have crypto agility. Guys, you can have more than one.
- [71:37]
A
Bitcoin does already support multiple signature schema, so. And by the way, you can also add. I think, well, this is some of the work that needs to be done. I, I'm not sure, but I have the. I would bet that we can add a new schema without a hard fork. We can simply add functionality in some way. That's.
- [71:57]
B
Taproot gave us that ability, I think which is an unintended and great consequence of Taproot is that it's. Taproot was the first step in quantum preparedness.
- [72:07]
A
Okay, last question here on Bitcoin and quantum Nick. What, what will be done? Not should. What ultimately will be done with dormant coins that choose not to upgrade, such as satoshi coins, what will happen? You have a great fictional piece, I forget what it's called, but I loved it. Which combines all of this where trillion Dollar salvage.
- [72:30]
B
It's my first foray into fiction.
- [72:33]
A
Yes.
- [72:33]
B
And in that, like Lyn Alden, I'm
- [72:35]
A
a fiction writer, you hypothesized that a Q day occurs and the government ultimately hacks those coins to preserve them. Basically, yeah.
- [72:46]
B
I think in the story, I think Q day was in 2029, maybe. It's interesting that Q Day and AGI Day are going to happen like within.
- [72:54]
A
It does seem like each other.
- [72:55]
C
Yeah, yeah.
- [72:56]
B
They're going to be like the same day, probably.
- [72:58]
A
And maybe even not just a correlation, but causation.
- [73:03]
B
Yeah, 100% there is. You know, you always want to be careful to not substitute the normative for the objective. You know, like what you want to happen, what you think will happen. But I actually do. What I want to happen is also what will happen. I sincerely believe that. Which is I think there will be a salvage effort for the coins, probably led by the government. They're the best equipped to do it. Maybe a private sector company will get there. And the reason I think that'll happen is I think the US has the lead in quantum. So I think a US company with the. Certainly there's industrial policy happening now, so with the stewardship of the government will build the first quantum computer. And one of the first things they'll do, because it really matters is salvage all the bitcoin, put it in a trust. Anybody who owned it before the salvage can claim it back. And that's, I think, much more tolerable to the bitcoin community than fork to freeze the coins. I don't think that'll happen. Even if it should. I don't think it will. It might. Now that we have a consortium, you know, maybe things.
- [74:10]
A
Hey, we're not suggesting anything about that.
- [74:13]
B
I know that was. That was uncalled for. I don't think there'll be a fork.
- [74:17]
A
I think you're right, though, that it is more palatable to. Because, you know, it's like who holds the keys, holds the coins. It's still sort of in. In line with Bitcoin's ethos, you know, for sure.
- [74:30]
B
And part of bitcoin dies if we, the bitcoiners collectively seize or freeze the coins.
- [74:36]
A
I think that's right.
- [74:37]
B
Part of its core constitutional value has died.
- [74:42]
A
I mean, it basically becomes a new version of like the dao hack in eth. Classic on bitcoin. And bitcoin is so unique because it's first. No one else can ever have the first public blockchain cryptocurrency. Right. And because of Other distinguishing characteristics like that it has no pre Mine or ico that it. And it has never censored people's transactions. You know, and those are like, some of the core philosophies. I agree that. I think yours is the best option, frankly. I've also heard just let other people. Yeah, well, I think. I think some bitcoiners have also. They're so sanguine about it. They've said, so what if they're hacked? Like, you know, eventually they get distributed. They would rather, to your point, preserve that. That uncensorship.
- [75:28]
B
I respect that perspective, for sure. I think it's too dangerous, frankly, because we don't know who. And that person or entity that gets the coins, who knows what they want to do with them. They could, right? Lopp has made this point. They could use those coins to further harass Bitcoin. I don't think it should be left to chance. But I also don't think the bitcoin community will collectively agree to freeze, because I actually don't think the bitcoin community possesses the ability to actually agree on virtually anything at this point. So I don't think it's even an option because there's no way to go out and ask every bitcoiner what they want to do. Yeah, as we see with the filters
- [76:11]
A
debate, we're not touching bip110 on this show, Nick. Well, on this episode, I should say we've talked about it. I had Matt Carollo and we talked about it with him. And that you're right, that shows how I'm gonna say difficult bitcoin governance consensus is. But your point is well taken. I would say it's fair to say nearly impossible is. And many would argue that's a feature, not a flaw. But when. If you face something existential, then it is definitely a flaw. Right? Spam's not existential, but quantum could be.
- [76:43]
B
That's right. Well said.
- [76:45]
A
Last question for you, Nick, and thank you for running over with me, you guys. Are you still investing out of Castle Island Fund 3? And what are you excited about and looking at these days at Civ?
- [76:55]
B
Well, we actually haven't formally announced, so. But we're at. We're investing under our fourth fund now, so.
- [77:01]
A
Congratulations.
- [77:03]
B
That's actually the first time I think anyone's been made aware of that, so.
- [77:06]
A
Well, I can cut it if you need.
- [77:09]
B
No, I think it's fine for people to know. Yeah, I mean, if you made it
- [77:13]
A
to an hour and two minutes into this recording, then you deserve it.
- [77:17]
B
You deserve to know this very important nugget of information.
- [77:19]
A
Well, because I was just looking and I remember when you guys raised Fund three, and I. And I just looked and I didn't realize it was. It was in 22.
- [77:25]
B
It was four years ago.
- [77:26]
A
I mean, it's. God, we're. I guess I'm getting old, man. Like, we've been in this game a long time.
- [77:30]
B
I keep having these moments and freaking out about how old I am because we've been doing this for a decade now. You. You. A bit longer, even.
- [77:36]
A
Yeah, I mean, it's. It's. It's crazy, but. But what? I know you guys are deeply interested and involved in stablecoins and. And a bunch of other cool stuff. I mean, I mentioned Alex Pruden from Project 11, and you guys are investors in Project 11, and we've talked about that. I've had Alex on the show. What else are you guys looking at? What's exciting in VC these days? Because the crypto vc, which I've been following for a long time, the quarterly reports I've been putting out for, like, seven years, the deals are down, the money is down. A lot of crypto VCs I saw, like, Paradigm raised a big new fund and explicitly not know crypto, but also like AI and other tech. How are you guys playing this market as investors?
- [78:15]
B
Yeah, we thought about that too, actually. I think every fund in crypto has this decision to make. Do they just pivot to AI and robotics, or do they stay in what some believe is a shrinking market? And it's a very important decision, you know, and I won't name names, but, yeah, I went through list of all the funds and I saw if they're leaving or not. And we decided that, no, we actually like what we do and we think we're good at it. And we don't think we have an edge in robotics or. Even though I'm on the board of a robot company, I will say that we at Castle island don't think we have an edge out there, you know, in whatever it is, frontier technology that has nothing to do with blockchain. So we actually like what we do. And our approach has always been very different. We care about institutional adoption and enterprise adoption of stablecoins. Maybe not the sexiest thing, but there's still plenty of opportunity there. Our core North Star metric these days is the flow of stablecoins, which is still very robust, actually, if you look at it. Not even necessarily the stock metric, but more the flow.
- [79:29]
A
Yeah, but even the stock in this bear Market like the supply has not come down much. Not really.
- [79:35]
B
A little. A couple percentage points.
- [79:37]
A
Yeah, I mean I think you, it's, it's. That's, that is very different from the prior bear market when stables were big. Back then as well, you saw a substantial like reduction in supply.
- [79:49]
B
Yeah. So to answer your question, we are still doing our core thing, which is basically fintech, stablecoins, that sort of thing, neobanks, et cetera. Cybersecurity is an increasingly large part of our business because blockchains are actually really on the frontier for the whole enterprise. I think so many things, and I don't think anyone has really figured this out yet. So many things that start as best practices in crypto world actually graduate to the rest of the world. So because it's so hostile, the best practices for defense are incubated here and then they go everywhere else. This is something that we've noticed in so many domains of security and that's actually maybe the most active part of our business right now. So whether that's quantum or regular old AI defense. And I think crypto, if you look at it, crypto has been this breeding ground of absolute chaos for 15 years and it's produced a, a very good security dividend for the rest of the rest of the world. So that's something we do actively. And then I think maybe after all this time, like the notion of it tokenized security could be real maybe thanks to clarity, that seems to be changing in real time. So potentially a whole new asset class might open up here. And so we're monitoring that.
- [81:15]
A
I agree. I think our audience will know I'm very, very interested in tokenized real securities. Sort of said like, you know, bitcoin has graduated. It's kind of one, it's got product market fit just for, without going deep. Like people all over the world like to own bitcoin. Stablecoins obviously extremely useful also clearly have like uptake in a variety of domains. And then I like defi applications, a lot of them, like primarily those that are focused on trading or lending. But my, one of the, I said this like last year, a long time ago, I was like, the problem is many of most of the assets that you can trade are crappy. So if we could improve the quality of the assets, then those systems could get a lot more use. And one of the things I like most is tokenized stocks. But lots of debate about what form those take, how rules need to change, who can use them, when and where. So I agree that's an interesting one. And your point about cybersecurity is quite fascinating. I'll have to look more into that.
- [82:17]
B
Yeah, I'll probably write an essay on it, but it's something very underappreciated. I've been thinking about it for a while and I think crypto has won, if not a sincere victory in terms of being the next monetary substrate for the world. Crypto has won a cultural victory in that crypto culture is now spreading out its tendrils throughout the rest of the world. And that's one of the main things is security practices.
- [82:46]
A
All right, Nick Carter, co founder of Castle Island Ventures and Coinmetrics and probably a bunch of other things. Great writer. Check out Nick's writing if you haven't read before, I think@nickcarter.info. right. Is that still your website? And listen to on the Brink with Nick and Matt.
- [83:03]
B
Thank you.
- [83:04]
A
Thank you for parting One of my favorite podcasts. I listen to it every single week and you guys just yuck it up about the the crypto news. And it's among my favorite shows, so I really do recommend it.
- [83:18]
B
Eight years strong, 700 episodes in the can of on the Brink.
- [83:22]
A
And like, I still, still holding out for the fact that I might get to hear that lost episode one day.
- [83:29]
B
You.
- [83:29]
A
I think you said a week or two ago that there's a second lost episode.
- [83:32]
B
Yeah, there's been a couple episodes recorded and forgot to publish.
- [83:35]
A
Well, I'm thinking about the one that you guys apparently recorded in Miami one time and then somehow lost.
- [83:41]
B
I think it was a few drinks deep.
- [83:44]
A
Well, Nick, thank you so much as always for coming on Galaxy Brains.
- [83:47]
B
Thanks. We'll do it again in two years.
- [83:50]
A
Yeah. Thank you for listening to Galaxy Brains, the weekly podcast from Galaxy Research. I'm Alex Thorne, head of Firmwide Research at Galaxy. Follow me on X at Intangible Coins. Follow Galaxy Research on X at GLXY Research. Read our written reports@galaxy.com research and don't forget, if you like Galaxy Brains to like and subscribe on your favorite podcast platforms like YouTube, Spotify, Apple Podcasts and more. We'll see you next time.