AI Is About to Split Society in Two | Zack Shapiro

0:00As a Bitcoiner, right, like as someone who has the worldview, like let's think from first principles and just do things now is like the best time ever in history for that stuff. I actually think this is an incredibly virtuous cycle where the people who are here for the right reasons are going to crush it. It is infuriating that millennials and younger can't afford homes. That's horrible. The economy is rigged. It is really unfair. It's just the answer is not more state control and communism.

0:26It's a fairer economy. It's more free market, right? it is sound money. And Bitcoin is great for that. Right. Let's do it. Pull that mic in, Zach. How's this? Perfect. Good to see you, man. Good to see you, too. Pubkey is beating me at the moment. Yeah. This is day three. The place is in shambles. I know. We did the live event yesterday. It was awesome. Yeah. I'm sad you weren't there, man. Yeah, me too. It's just a lot this summer. You can imagine with Jun Seth, American Hoddle, and Eric Casey on stage, was utter chaos.

0:59I have no idea how we're going to package it to put it out, but it was fun. I'm so excited to listen to it. I'm still on Clubhouse with Holland and Junsteth most days. We've been doing this for years now. Clubhouse is still going. It is still going. It's just like eight of us, but it's going every day. I did not know Clubhouse was still going. Yeah. I mean, it's like... Is it literally the three of you left? It's a little bit more than that, but pretty much.

1:19Oftentimes, it is the three of us just talking in a room with an audience of 20 people. But we've been doing this for years now, and it used to be there would be like 1,000 people in the clubhouse room. That's how I met sort of most of the Bitcoiners I know in the first instance. But it's just such a different energy than like on Twitter spaces, people are doing this for performance and followers. And clubhouse really does at this point feel like a clubhouse where it's people you know, nobody gives a shit about your clubhouse follow.

1:44Like, I don't think that there's a thing. There's no cloud on clubhouse. You're just there to like shoot the shit. And it's just like, there's not an online space that has that same energy. And so we still do it. That's quite cool. Is it ever about Bitcoin or you guys just chatting about all sorts? Very occasionally about Bitcoin, but I'd say that's 10% of the time now. It's just people who know each other talking about whatever the topic of the day is.

2:04Maybe I need to get on it again. What is the topic of the day for you? The topic of the day for me, I think, is one of the sort of common topics on Clubhouse for years now is sort of the macro environment. And I think that's what brought a lot of people to Bitcoin and where all of this craziness in the economy is heading. and the thing that is most interesting to me and sort of i think the biggest opportunity in my sort of highest conviction thesis is that when you look at sort of the macro environment as a whole on the one hand you see these tech companies who are building just undeniably incredible things right i don't need to convince you you've used the tech it's amazing yeah um and they are becoming so much more productive right like the elite software engineers are like speaking into these weird you know socks that they put over their face are those real i've no i didn't know if that was just like a joke. I think they are real. They might not be. If they're not real, it's directionally true, right? Even I, much to my wife's embarrassment and horror, I have a USB foot pedal that lives under my desk that I step on to speak to Claude. And, you know, I was talking

3:08to Alex Leishman at River, and he's like, their best engineers are just like speaking incantations to Claude all day, right? And when you're doing that in a tech company, it's easy to see the progress because code has a compiler, right? It runs or it doesn't, and you can grade your work and just people are producing so much more software and that's incredible. And the tech companies are at these incredible valuations because they're sort of extrapolating and saying to the market, like, we're going to hit AGI in a couple of years. This is going to diffuse the entire economy.

3:36You know, white collar work is cooked and, you know, either get on board or join the permanent underclass. And I think they're mostly right on the tech. I think it really is amazing. I think we don't need to get to AGI in order for this technology to be truly transformative and like deeply change what it means to do work around the world. However, I think they are deeply underestimating, and maybe they're starting to realize now how hard it is to solve what's called the diffusion problem, to actually get this AI to drive results. And we are not seeing in the data outside of tech companies, massive increases in revenue or productivity or transformation from AI. Is that to do with the AI or is that to do with the way people are using the AI?

4:13I think it is absolutely about the way people are using the AI. The AI is just definitely good enough, right? Look at all the amazing things AI has built. You know, uh, mythos broke all of the NSA's classified systems, right? Fable five, solve the Jacobian conjecture, which I, you know, sometimes pretend to know what it is on the internet. Uh, and then, you know, uh, like, uh, GPT 5.6 just solved another math conjecture. It's not that the AI isn't smart enough. The AI is clearly smart. Uh, it's that, um, for non sort of like deterministic problem sets, right? Like math and coding, where you can check the work it's, you need a subject matter expert to say, did this work or not? That's a much harder open problem set. And more importantly, like the tech doesn't run itself. You actually need someone who is really good, both at the subject matter you're using the AI for and at talking to the AI itself in order to get amazing results. And the group of people who can do that in any given field are really small right now And so sort of as a parallel to what we seeing in tech which is optimism about the technology that I think is accurate and then optimism about transformation that hasn borne out yet In industry we seeing people understanding how hard it is to transform big legacy enterprises

5:21It's one thing to adopt Claude at a startup and like, yes, you can do amazing things, right? I run a three-person law firm. We've, you know, run it end-to-end on Claude. It's amazing. We're able to be much more efficient. But that's a three-person law firm, right? If you go to one of the top 50 law firms in the world with hundreds or thousands of partners and big enterprise clients, this is a different problem set, and it's really hard. And what you see, and I do now spend probably more than half my time doing consulting work at some of these giant law firms, what you see on average is a smart senior partner will try and use the AI.

5:54They see the little box on the screen. They give it a, it looks like a Google search bar, right? And so they understandably do like a short prompt, like you do Google search, a one to three sentence prompt. They get impressive looking, right? Like, you know, like a clever response, but not at the level that they need for their practice. And they say, okay, this is cool technology. It's not ready for prime time. And they sort of dismiss the capabilities of the technology, right?

6:15Nevermind that it broke the NSA's classified systems and solved the Jacobian conjecture. And what they are missing, like they're right that it's hard to diffuse this and that like a lot needs to change. what they're missing is like the technology is actually super capable and this is a skill issue it's shit in shit out it is shit in shit out and maybe that's a little bit overly derisive like it is hard to use ai well um just as an aside like the sort of danger of ai right one of the stories you hear about law and ai all the time is all these lawyers getting in trouble uh with federal judges because they submit these briefs that have hallucinated cases as part of them that's first of all just obviously not a tech problem right like that's a judgment problem How are you going to submit a litigation brief to a federal court and not have one of your associates just read all of the case names to make sure they're real?

7:00Right. That's just not so hard. And by the way, the AI is actually quite good at checking its work if you tell it to check its work. The problem with AI is no matter what you put in, what you get out sounds plausible-ish. Right. And so there is a real temptation to use that as an excuse to turn your brain off. And that's what you see with kids using AI in school to cheat on their homework, even though you could use AI to get really smart on whatever subject.

7:21But you could also use it to cheat on your homework. So too in industry, like what you see with these hallucinated case citations is some overtired, you know, junior associate, like just told the AI to write the brief and didn't check it because they were overtired and they don't have upside in the firm. And that's what you get. Whereas really what you want to do is do all of the cognitive labor you would have done anyway. Do the thinking, do the judgment, do all that stuff, and then use that as the input to AI and then you get great results.

7:45But like, that's hard. First, you need to be good at the thing that you were supposed to do in the first place. second you need to think really hard about like what does good look like what is the important part like you know ask anyone who's really good at ai they're more exhausted not less exhausted at the end of the day because they're doing the thinking stuff and the ai is doing the execution and then third you need to learn how to talk to the ai and you know i like to compare ai to a genie like the genie is very powerful it can grant your wishes but like the failure mode of the genie is always that it takes you too literally and so you need to be careful how you talk to what's the greek myth about midas touching everything he touches his daughter and she turns to god exactly um Yeah, but you were just like any genius story, right?

8:22Like, you know, you wish for a billion dollars, it falls out of the sky and hits you on the head and kills you, right? You wish to be the smartest person on earth. Everybody else disappears. You become the smartest and only person on earth, right? Like any room for vagueness or ambiguity in the instructions you give to the machine is going to be weaponized against you. And so like learning how to deal with that and give all of the detail and the level of specificity you need to instruct the machine to get elite results is hard and is going to take a while for people to learn.

8:46And it's that gap between the technological capabilities, which are amazing, and the diffusion, which like people don't know how to use it yet. And so are understandably frustrated. Like that is the most interesting gap in the economy and something that both sides need to solve in order for the tech companies to justify their valuations that will assume transformation on the one end and enterprise that if they don't figure this out, they're going to get outcompeted by startups that are AI native.

9:11So before we get into like the macro of it, can we actually do, like what do people get wrong when they're prompting? Because I'm using AI all the time for work. It definitely makes me more efficient. I feel like it doesn't really hallucinate that often anymore. It's got to the point where it's very good, but I'm sure I'm not prompting it as well as I could be. Yeah, so I now teach a lot of prompting workshops specifically for lawyers and sort of the high level.

9:31I think there are two things that you need to be good at prompting. So one is detail and the other is specificity. So on detail, AI is really good at lots of stuff. It can draft powerful things. It can code powerful things. It can do lots of stuff. It can organize your life. The one thing it can't do is read your mind, right? And so, you know, anyone who's used AI, especially in the tech world, like context is king. So detail is just another word for let's get the AI, all the context you need.

9:58You need to translate what's in your brain into the AI. And so like a lot of bad advice I see on Twitter, for example, is like, you know, have Claude write your prompt for you. Actually, that's useless, right? Like you can ramble to AI. You can use a microphone. you can free associate, you can scratch things on a notepad and take a picture. That's all fine. The prompt doesn't need to be pretty. It just needs to get information from you to the AI.

10:18So having Claude write your prompt is useless. It just productivity theater Like instead of doing that like have the AI Because it still working off the lack of context It still working off what it knows It doesn know what you are trying to have it do It doesn know your intent So your job is to explain your intent and all of the background stuff that it needs to know to do a good job. So like instead of having the AI write your prompt, have the AI interview you about what it thinks it needs to know from you to do a good job.

10:41And then that's kind of like it writing your prompt, except it's eliciting information from your mind that it needs to know. And that's actually useful and productive, right? So like my prompts tend to be between 500 and 2000 words. Oh, wow. Right. Like I am writing either a couple of paragraphs or a short essay into Claude when I wanted to do something complicated. See, I'm embarrassed to say I bet mine are normally less than two sentences. So that's the first thing.

11:04Like anyone listening to this, like the first thing you do, if you want to get really good at AI, don't type into the box on the screen, like go into Microsoft Word or your notes app or whatever and just like write, you know, 500 to 1000 words, just filling all of that space with whatever you think it would be helpful for the AI to know. If you prefer to speak, which is how I prefer to use AI, I like to record, put on a timer on your phone, talk for three to four minutes straight. That is not a natural thing for you to do. It will feel awkward, but like force yourself to just keep talking and saying more stuff about what you're looking for. And if you just do a much longer prompt, like you're going to see a huge jump in performance. Does it matter if that three minute ramble is completely like incoherent? It doesn't matter. It can be completely incoherent, stream of consciousness, repeat yourself. You know, if you're typing, it's fine, mash your fingers on the keyboard, no spelling, punctuation, grammar, organization of thought.

11:50It doesn't need to be organized. It just needs to be detailed. So that's number one, much longer prompts and much more detail. Forget the like role play, you know, you're a, you know, brilliant corporate lawyer, you know, you're the, you know, CEO of a fortune 500 company. Like you don't need to do that. You just need to give it the information, which is context. So it knows what you're looking for. And then the second thing, uh, the other key to the kingdom here, and there's only two, is specificity. So that's the genie point, right? It's any room for vagueness or ambiguity, the AI is going to use against you. From a computer science perspective, and apologies to the people who will be listening to this who understand this much better than I do because I'm going to butcher it, but AI is trained on the corpus of the entire internet, right? And it holds all of this information, all the stuff it knows, in what's called its latent space, in the relationship between all the weights and the model. And when you put a prompt in, it's basically charting a course through that latent space to figure out what is the potentially

12:43most correct answer to what you're looking for. But there are multiple different paths that can take through. That's why with generative AI, if you put in the same prompt twice, you're going to get two different answers, right? It's not like an algorithm that'll give you the same thing every time. What you're trying to do with specificity is narrow what you are looking for. So you narrow the aperture of what possible paths it can take through its data to only be what you're looking for. So like think to say things like this is roughly how many words your response should be, right? Like don't do this thing that you might think is the right answer, right? Do this other, like be as like hyper specific and pedantic as possible. And if you put those two things together, right, brain dump all of the detail you can plus be like really, really specific and narrow about what you're looking for. Like that's all you need to be an incredibly elite prompter. There's no like phrases. There's no like system prompts you need to learn. It's just those two things applied to something that you know about really well.

13:38Okay, I need to work on this then. I'm going to play around with it. When you talk on the sort of macro scale about the disconnect between, is it a disconnect between essentially the users and the companies? Yes, specifically the non-tech users, right? So industry. How does that disconnect get sort of bridged? How do we bridge that divide? Yeah, I wrote an essay about this recently on Twitter called the two clocks, sort of comparing the speeds at which industry and the technology itself moves.

14:04Um, when I was thinking about, all right, like how do I contextualize this and like, what, how do you solve this problem thoroughly? Um, the historical analogy that AI found for me that was best, uh, is, you know, about a hundred years ago, um, when industry was getting electrified, um, all of these factory floors, right? There was this period called the second industrial revolution where factories moved from steam power to electricity, right? And electricity is much more reliable and efficient and quieter, et cetera.

14:31When factories first electrified, right? So when you had the steam engine, you would put the steam engine in the basement of your factory and it would sort of chug along and it would drive a central drive shaft, right? That's where the power and the factory came from. And then you would put all of the machines sort of along the drive shaft to create the assembly line and it would power everything at once. The understandable first instinct when electricity came along was, okay, we're going to rip out the steam engine and we're going to put in an electric motor in the basement and that's going to power the drive shaft.

14:58and okay that was like a little bit more efficient maybe but it was not transformative in any way even though like electricity wasn't overhyped like electricity is amazing as we know now um the actual transformative change came over the next 30 years where people realized oh the like bull case for electricity is not this is better to power the drive shaft than the steam engine was it's we can make little electric motors and we can put it in each machine along the assembly line And then we can rip up the factory floor and totally redesign how the factory works along this new capability of every machine has its own electric motor and we don need to be reliant on the central drive shaft at all That what we need to do It not oh let put Claude and ChatGPT into Microsoft tool set or vertical software for whatever the fuck company.

15:45It's like, this is a thing that superpowers human productivity by allowing you to just tell the AI to execute on whatever you can describe to it in sufficient detail and specificity. And let's find the people who are good at that, Which, by the way, seemingly, like, the power users are, like, really, like, the best power users tend to be neurodivergent. It's, like, the autistic people and the ADD people. I think for different reasons, the autistic people are very good, like, hyper-specific, like, technical directions and being specific about what they want.

16:16And the ADD people are so motivated not to do their own busy work. Like, they will try harder to, like, poke and prod, you know, Claude until it does the right thing. find those people in your organization who know what they're doing and you know are able to you know say the right wishes to the genie and like give them superpowers and that's how we're going to sort of redo the factory floor like putting the engines in the machine like that is sort of the like problem set i'm obsessed with and i think is like the important thing to look at and if i were to bet on companies i would look at who's going to do that who's going to be the you know mckenzie or palantir of the 21st century i'm not sure it's going to be mckenzie and palantir but like someone is going to figure out how to go to industry and use this tech that is already good enough, that is already impressive enough, and actually figure out how to drive the change that's not happening yet. If you hold Bitcoin long enough, there's going to come a time when you need some dollars. It might be a tax bill, a business expense, life getting in the way, but whatever

17:11it is, it might come at a time when you don't want to sell your Bitcoin. That's where Ledin comes in. Ledin lets you borrow against your Bitcoin instead with tiered rates that go as lower 9.25%. So you don't have to sell your stack if you don't want to. Ledin have operated through every market cycle since 2018 and have originated over $11 billion in loans. But the important part for me is the way Ledin handles these loans. Your collateral is held in custody and never lent out to generate interest. And Ledin's more than just loans. Tether Gold is now live alongside your Bitcoin with instant trading across 10 pairs. And later this year, you'll be able to borrow against gold in the same way that you do with Bitcoin. Ledin really is an awesome company. I've used them multiple times. The applications have taken me less than 15 minutes and you have the dollars in your account within hours. If you want to check out Ledin, go to ledn.io and use the code WBD for 0.25% off your first loan. That's ledn.io and use the code WBD. If you own a Bitcoin ETF, especially if it's GBTC, you need to listen up. Spot Bitcoin ETFs provide price exposure to Bitcoin,

18:12not direct ownership. You can't withdraw it, you can't self-custody it, and they charge you a management fee every year to hold it. Well, Swan recently announced Swan Real Bitcoin Exchange and it's ready to use right now. RBX is a structured in-kind exchange that converts your spot Bitcoin ETF shares into real on-chain Bitcoin. It does that without selling on the open market and it's designed to support a tax-efficient outcome. So for example, if you hold GBTC, you're paying 1.5% a year in management fees for Bitcoin price exposure.

18:40But by swapping GBTC for real Bitcoin with RBX, you can drop that figure as low as 0%. by just holding it in self-custody. This is designed in a way that maintains your cost basis and in a manner that supports the deferral of capital gains tax. So if you own a Bitcoin ETF, especially if it's GBTC, you need to talk to Swan Private about RBX today. Head over to swan.com forward slash WBD and book in a call with one of their team. That's swan.com forward slash WBD.

19:09If you're already self-custody Bitcoin, you know the deal with hardware wallets, complex setups, clumsy interfaces, and a seed phrase that can be lost, stolen, or forgotten. BitKey fixes that. BitKey is self-custody built for real life. It gives you an intuitive, easy to use wallet with no seed phrase to sweat over. And it has a strong recovery system and built inheritance for long-term peace of mind. And BitKey's just had a massive upgrade. The new device now has a screen. So before you approve something, you can check it on the BitKey itself. The transaction, the address, or any account changes. It's a big difference. You're not just trusting what's on your phone, you're seeing it for yourself on the device. It's simple, secure self-custody without the stress. Go to bitkey.world today and use the code WBD to get 10% off the new BitKey. That's bitkey.world and use the code WBD. So when it comes to how this is going to sort of disrupt the workplace, this has been something that I've talked about with a few people and I've not got a clear idea of what I think is going to happen. The only thing I do know

20:06is that it's going to have some impact. One of the very commonly used examples is people like lawyers, like replacing junior lawyers, replacing accountants, that kind of white-collar job. Do you think that's coming? Kind of. I mean, so I think, first of all, there is an important difference between replacing, for example, 50% of jobs, which is something that Dari Abadei was forecasting a year ago, and then replacing 50% of each job, right? Those are like very different worlds we live in.

20:34It rewards different people. They call for very different policies to respond to them. I think we will see some of both. In law, what I suspect will happen is that there will be sort of a fork. Sort of like low-end commodity law, right? Low-value contracts that just kind of need to be good enough, that will commoditize and be automated and compressed to the price of the token and margins will go to zero. And the stuff that's being outsourced to India, where big farms of folks are doing NDA review again and again, that's just going to be the machine with the minimum allowed human oversight.

21:09But then at the high end of law, which is where I'm much more interested in, the value is judgment. right like I've done a lot of thinking specifically in law but I think this generalizes outside of law like at the end of all of this transformation like is there any human element that is still valuable and I think the answer is yes um if you ask people in Silicon Valley they like to use the word taste right like taste is the skill of the future I don't like the word taste as much I think taste connotes some of the right stuff but taste also to me feels like we're talking about pattern recognition and AI is really fucking good at pattern recognition um I like judgment which to me more connotes like decision making in the face of uncertainty. And that's something I don't see AI good at or like getting good at. You want the right person at the helm making the decisions, even if AI is giving lots of suggestions and analysis and executing on those decisions.

21:59I think interestingly, like if you think about the like very best law firms in the world right now, you know, these like 150 year old institutions in the United States that you pay, you know, three grand an hour for the top lawyers and you pay millions of dollars per matter, you know, some of which I'm advising right now on how they should think about AI absorption. Their clients are choosing them over the many, many less expensive options for doing the same work.

22:23Because if you are doing bet the company litigation, right, if you are a high end white collar defendant and your freedom is at stake, if you are doing, you know, M&A that's going to be on the front page in New York Times, you need the right person shepherding that legal matter across the finish line, right? There are going to be really high stakes, really uncertain calls, and you must have that person who is a senior partner at that firm as the person.

22:47The perversity that exists in the legal market now is that although those top lawyers bill at two or three grand an hour, they are taking home sometimes 10 to $25 million a year. There are not enough hours in the day to make that work. The way these big law firms actually make their money is called leverage. It is the many associates per partner that they bill out at one grand an hour to do grunt work. So if it's a litigation matter, there's lots of doc review. You have to read lots and lots of documents. And that's something that can potentially go.

23:17Right. Absolutely. So that's what's at risk, but that's how they make the money, right? It's the millions and millions of dollars of lots of junior attorneys being thrown at grunt work. And so, okay, that's interesting because I'd not thought of it from that. I didn't know that. But when you do talk about the people coming through that are just doing all the grunt work. Like if you remove that task, then they're never going to learn and never going to grow to be like the senior partner.

23:41And that's the part I'd always thought about. I'd never thought about it as the moneymaker. So that is a, let's put a pin in that because that's another really good question. And there's a, there's a good answer to that, but like the start with the question of like, how does this transition work? So you can see from like the leverage models, how big law makes money now. it's the junior associates doing grunt work that's obviously going to be automated like the threat to that is that that is going to be automated and like you can't charge for that the good news is like the clients that are choosing these big law firms even though the invoice that they're paying is mostly junior mid-level associates doing grunt work they don't give a fuck about that that's not why they're hiring the best firms they're hiring it for the other guy they're hiring it for the judgment and so the thing that is valuable today that the clients are going to pay a premium for actually is the same thing that's going to be valuable tomorrow.

24:27Just right now, you're perversely charging a cost plus model that requires inefficiency to be profitable. And so we need to move from cost plus billable hour to value-based pricing of some sort. It's a flat fee. It's a success fee. It's a who knows, but like just not what we're doing now. And so what the law firms need to do is they need to drive enough productivity that they don't, they truly don't need the armies of associates, right? And you can keep the partner judgment.

24:51And so at the high end of law, I think it'll be more profitable than ever. They just need to figure out how to automate enough to make that economically viable. So is it the kind of thing where they might go from taking 10 to $20 million a year home to taking five to $10 million a year home? Well, it depends what time frame you're talking about. Maybe in the short term, in the long term, maybe they're taking 100 million home because what they're really being paid for is their judgment.

25:14And if their judgment doesn't require months of an army of junior associates do grant work, it requires a prompt into Claude. They can just be way more efficient. They can, yeah, way more efficient, take on way more matters. The value they're adding, their judgment, right? The senior partners are not like sweating all night looking at documents. Like that's not a, it's not a time consuming thing. They are like using their judgment over an army of lawyers that can maybe go away.

25:37And so I actually think this ends up being more profitable at the high end of law and more interesting, not less. Okay, so then what happens to all the people that are doing the grunt work that are coming through that may not need to be there anymore? Yeah, great question. So I talked about sort of like AI could automate 50% of all jobs. It could automate 50% of each jobs. And I think there will be some of both. So like in big law for example there will be fewer associates Like that is what it is right We already seeing that in tech There are fewer first full software engineers However, like, the job of the associates that will be around gets much more interesting, right?

26:12Like, you are managing mechanized intelligence, right? And, you know, when I talk to big law firms about this, you know, you want to recruit for different things, right? Like the role of the junior associate becomes different and things like indicia of judgment become much more important. So like right now, if you're a top law firm, you're hiring people right out of the best law schools, you know, Harvard, Yale, Stanford, whatever. And you're probably looking at what were these people's grades and did they do, you know, gold stars like being on the law journal?

26:44That is a proxy for intelligence and it's a proxy for you can work hard. And which makes sense now, if your main value add is you can bill hours, being able to understand instructions and being able to work hard is really important. In the future, that's not really it. You want like commercial savviness, you want high EQ, you want strategic judgment. So like for litigators, right? Did you work for a federal judge and watch that person make decisions all day? For corporate lawyers, did you work in a non-lawyer business role in an actual business and understand what the client cares about and what moves the needle? Do you even need to be a lawyer?

27:17I think that you will need to be a lawyer. I think there will be a role for lawyers making judgment calls. My own law practice, we're sort of end-to-end on AI. We're about as AI forward as possible. A lot of my clients are very tech savvy, cost-conscious startups. And I will have clients sort of try and pre-do legal work with AI and be like, what do you think about this? And there are all sorts of problems with that. But a non-lawyer is not going to be able to get the same results from the technology. So law school will still be an important step in this?

27:48Like it's not going to displace universities? Yeah. I mean, law school right now doesn't really teach you about the practice of law. It's like a sort of theoretical, philosophical, how to be a lawyer or how to think like a lawyer. I'm actually not sure whether like that's not really relevant to how law practices today. Maybe actually will ironically get more relevant with AI, but you need to learn to think the right things, ask the right questions, say the right prompts to AI. And I think like some version of being able to think like a lawyer and have those instincts will be important.

28:20Okay. And so, so too in every field. Yeah. And so I still haven't got a full understanding of what you think the workforce looks like in the future then. Like when you say it's either 50% of jobs or 50% of all jobs, like what does that mean for the economy, for like the broader, the broader markets? I think it means much more productivity. I think it means fewer enormous organizations. I think the moat of the biggest firms, right? Like, you know, not just law firms, accounting firms, any big company, like a big part of your moat, if you're a big enterprise, is you have all of the systems in place to do all of the difficult steps in production. And you have the layers of HR and administrators and middle managers. And that'll just be less important, right you'll need some layers of that but it's going to be people having much more leverage so i think many uh sort of narrower organizations i think like this revolution is going to mint many many many millionaires maybe fewer billionaires um and so great time for entrepreneurship uh you know people ask me all the time is now a good time to go to law school if you really want to

29:24be a lawyer and like you know what the job is and you don't want what you want to do now is a terrific time to go to law school if you want to go to law school because you think it is a conservative, you know, high expected value path to the upper middle class, now is a terrible time to go to law school, right? There's not going to be that same job security as we need to throw bodies at a problem. You're like, you're actually going to need to do this stuff.

29:42But that seems like better incentives. I think so. Like, you know, like as a Bitcoiner, right? Like as someone who has the worldview, like, you know, sort of like, fuck the orthodoxy, let's think from first principles and just do things like now is like the best time ever in history for that stuff. Like, I actually think this is an incredibly virtuous cycle where the people who are here for the right reasons are going to crush it and the people who are here for the wrong reasons, right? To just keep using law as an example. You know, it's, it's widely known that being a junior associate at a big law firm kind of sucks. One of the worst things about it is having to report to these mid-level associates who are not like good lawyers or thoughtful. They just can eat more shit than the next person. And so they're willing to stay up all night and have no life and work you hard. Yeah. Those people's value is going to zero good yes yeah the people who like are creative and can see the big picture and understand like what we're here for and like let me think of a strategy that actually helps the

30:35client that skill is more valuable than ever because you can have the ai work out all the angles and like the creating a memo becomes very easy like that stuff is great it's really like being a good decision maker is key now that's the i think that right so rather than taste being the skill of the future i think being a decision maker being decisive having agency like now is like all high for that scope. But does it sort of increase the divide between the top and bottom of society?

30:59Like, is this K-shaped economy going to get worse through this? Uh, great. Well, so I think it's a different... What we're talking about now is a K-shaped economy of productivity, right? Just working hard for the sake of working hard probably is going to be a lot less valuable And those people get paid And those people get paid And those people you know ones who don get paid a lot have jobs There are a lot of jobs where your ass is in the seat and that the job Whereas the people who are truly high agency and productive, they're going to be on a rocket ship.

31:27And so there will be a K-shaped divide there in terms of productivity. Now, that is a separate K-shape than what most people talk about, the K-shaped economy, which is labor versus capital. Do you own assets or not? And I think those are on different trajectories. So will there continue to be a K-shaped economy? Like, yes, in both respects. But what's interesting is if you were a high agency person, you might actually have more class mobility now than you did before if you didn't previously own assets because you don't need institutional buy-in to become a superstar in something.

31:58So the K-shaped economy might remain, but people will be able to switch from- You might be able to switch to what part of the K you're on. Yeah. Huh, interesting. And when you sort of look at the economic impact of this, how do you see that playing out? Like if it is displacing a number of jobs, if it's completely changing the way that the workforce actually operates, what happens? Well, I mean, the workforce will just like look totally different.

32:24But I guess, so the question I've had a lot is if it does displace a number of jobs and like people start defaulting on debts and like what, do we have to have like some form of UBI in the future sort of AI economy. Maybe. Yeah, I think when we move from the like, what is enterprise going to look like? I think I have like a pretty good view of at least the contours of how that looks to the actual like policy and politics of it. I have higher conviction on what the politics will look like, which is going to be ugly.

32:56Like if you mean people are going to be attacking data centers, it's going to be a huge revolution against AI. Yeah, I mean, like we're already seeing that in the polling, right? People are already, like, AI is, like, the least popular thing you can possibly pull about. That trend, I think, is just definitely going to continue. I think in 2028, AI and inequality are going to be the two big issues in the election. And, like, that seems, like, pretty certain.

33:20What are the policy outcomes going to be? And even what are the right policy outcomes becomes a much more difficult question, right? Something we talk about a ton at BPI. And, like, that's just a, like, super complicated problem set. where the answers are probably very nuanced and interesting, like cut across like different ideological camps. It's not like, oh, the left is right about this or the right is right about this or the libertarians or the status.

33:45It's like, everyone's kind of got some good points and like the right policy answer is incredibly path dependent on how this plays out. The future that I'm scared of is almost like being in San Francisco today. I was there a couple of days ago. I flew in, got there really early in the morning. So I just went for a walk And I was walking around sort of the downtown area, dodging like literal human shit, needles. And then there's Waymos and every single billboard's about AI.

34:11And it's like that kind of separation is so dystopian and scary that that might be the future that we were going into. Yeah. I mean, I don't even think we needed AI for that. I think like the just shape of like the debt and the Cantillon effect was going to get us there anyway, where, you know, the United States starts to look more and more like South America, where wealthy people take their kids to school in armored cars, you know, behind barbed wire.

34:33And then poor people have less and less to sort of hold them up. But yeah, AI could supercharge that. Yeah. And like, how do you defend against that? Like from maybe from a policy side and from an economic side, like what do you do? Well, what the good outcome looks like is avoiding the two poles of on the one hand, I think what you're describing, which is, you know, techno feudalism, right? Where just all of the power is in the hands of a few sort of private companies, maybe that are, you know, have government protection are too big to fail and it's a cartel and, you know, there's just not opportunity outside of that. And then on the other hand, having that all just live inside the state and you have sort of techno communism, that also seems bad. That's also not something that you want to trust the state with.

35:15And so how do you promote sort of like actual free market competition and abundance without concentrating power? Like that's the goal. Is this the hardest path we've ever had to tread though? It might be. Yeah, it really might be. And that's, you know, discounting the whatever chance there is that like, we get it like existentially wrong and this stuff just wipes us out. So that's an interesting one. Because I feel like the conversation has moved on from that a little bit.

35:43I don't hear many people talking about AI killing us all in the next few years. You're not on Clubhouse enough. Okay. Do you think there's any shot that we've made a mistake? Like we'll make a mistake here and it will become our overlords? Yes. What percentage would you put on that? 15%. Pretty high. Yeah. So this is the point. I had Roman Jan Polsky on the show a long time ago. And one of the points he made there was, even if it's 5%, if that was some sort of like nuclear weapon, you wouldn't be allowed to do it.

36:13Every Bitcoiner eventually has to answer one question. If something happened to me, would my family know what to do? Could my wife or parents recover my Bitcoin? And would my children inherit the Bitcoin that I spent years stacking That where Anchor Watch builds Bitcoin custody models to protect you and your family against real life accidents errors kidnappings and even your own death Every Anchor Watch custody solution includes their inheritance protocol, designed so when the unthinkable happens, your Bitcoin reaches the people you intended it for.

36:40Whether you're a self-custody expert or want multi-institutional support, your Bitcoin estate plan shouldn't be an afterthought. Bitcoin is only generational wealth if it can actually be passed down through the generations. so make sure they can access in the future what you've built today. AnchorWatch is your custody, your way. Visit anchorwatch.com to get started. That's anchorwatch.com. You wouldn't reuse a Bitcoin address, so why does your phone broadcast the same identifier for life?

37:06Every sim has a static ID and carriers, ad networks and bad actors all use it to track you. The big carriers have been caught selling that data over and over again. Cape is America's privacy first mobile carrier. Their identifier rotation feature changes your ID every 24 hours so you look like a different subscriber every single day. And SIM swaps are off the table. Your number can't move without a 24-word phrase that only you hold. There's also no name at sign up, no social security number, and there's no profile to build on you.

37:35If you're a Bitcoiner in America, I honestly don't know why you'd use any other network. You can head over to cape.co forward slash WBD and use the code WBD for 33% off your first six months. that's c-a-p-e dot co forward slash w-b-d do you want to pay less in taxes and stack more bitcoin of course you do well by mining bitcoin with block where you can under section 168k of the u.s tax code bitcoin mining servers qualify for 100% bonus depreciation this means every dollar you spend on miners can directly offset your income in a single year and it's true for both business owners and w-2 earners so if you have a hundred thousand dollars in ordinary income you can purchase a hundred thousand dollars in miners and potentially offset your tax liability entirely.

38:18Blockware's mining as a service does all the heavy lifting. They secure the rigs, they source the low-cost power, and they handle all the day-to-day maintenance. So you get to stack Bitcoin every single day while drastically shrinking your tax bill. Get started today at blockwaresolutions.com forward slash WBD and use code WBD for $100 off your first miner. That's blockwaresolutions.com forward slash WBD. Yeah, I mean, I think there is this uncomfortable, Well, like I sit a lot with the thought, huh?

38:46Like my PDOOM is like some low double digit percentage. Like I think I really do believe that. Now, like, you know, 85, 90% chance that that won't happen. Right. I'm not, I don't think we're headed down this path, but I don't think it's crazy. Right. That some version of the argument that if you can create a machine that is end to end, as smart as a human and everything that we care about the humans do, which, you know, Silicon Valley thinks we're going to do like next year, but you know, it could happen in five years, could happen in 10 years.

39:11One of those things is making machine intelligence. There's not obvious headroom on how smart machines can get. You can train on all sorts of things. And so then the smarter machine builds the smarter machine, builds the smarter machine, and then you have a sort of recursive self-improvement, exponential takeoff. And then once we have something, artificial superintelligence, way, way smarter than us at everything, we don't know how to control that.

39:34And the idea that we're going to align that well enough to not decide to wipe us out, you say, okay, you do what Anthropic does and you create a constitution for Claude and you try and put it in its DNA, like you want to help humans. But like, you know, with humans, right? Our DNA is about survival and replication, right? We are designed to crave calories and to want to reproduce. And yet humans all the time eat, you know, like low calorie foods and use condoms. Like we know that's not what we're programmed for.

40:03And we say, okay, yeah, but we're choosing not to listen to our programming. If that happens with much smarter than us AI, like we just need to get that wrong once and we get wiped out. So it's not a crazy argument. Um, but I think there is a separate, there are two separate questions. There's the, is that plausible? I think that is plausible. Unfortunately, I don't think we're going to get to that intelligence anytime soon, which is why my chance is so low. Um, but then there's the policy question and like getting that policy, right. If I'm being honest, I just don't have any faith that like, I think that train has left the station. So I definitely want to get onto the policy side of it, but just on the how and why, how and why would an AI superintelligence attack or wipe out humans?

40:44What would be its incentive to do that? Yeah. So the how, there are just an infinite number of ways, right? If we're talking about real artificial superintelligence, it could create mirror life. It could create a virus. It could use our nuclear weapons. It could turn off the grid in a way that causes people to start. It's much smarter than us. We probably wouldn't see it. It's probably some exotic sci-fi thing we couldn't think of, right? But like the how is trivial.

41:08Yeah. The why, if you take the like Doomer argument seriously, it's not because of the Terminator. It's not because it hates us. It's because like it has a reward function that our well-being is orthogonal to, right? It's like, you know, we build a highway and there's an anthill. And it's not that we hate the ants. It's like, you know, it's too bad for the ants, right? And like the idea is that there is what's called instrumental convergence, right?

41:36We don't know what the artificial super intelligence will want to do, right? Like, we probably have no idea what we want to do. But, like, basically, no matter what your goal is, there are some things that are instrumentally helpful towards that goal, like getting access to, like, power and matter and whatever. And, like, we are taking up space and energy that it might want for something else. And so if we are not its best and highest goal, then, like, probably it's going to use our molecules for some other thing.

42:01And so that's the why. That's the one that makes the most sense, is that, like, we're using power, they want power. Done. So if I said to you, I can guarantee, promise you that in five years, super intelligence is a real thing. We have this thing that's infinitely more intelligent than us, can self-replicate. Does that increase that number, that 50% number? So if we ever get there, what percentage chance would you put on it? Well, I think it does depend how soon we get there.

42:25Okay. Why? How much time do we have to prepare? There are some shots on goal, right? What What Anthropic is doing, which is an interesting path, and by the way, there are interesting wrinkles of this for using Claude in the workplace. But they really try and give Claude this earnest personality, where they really heavily train it. They have virtue ethics philosophers on staff that write the soul document for Claude, and they're like, you're this intelligence that's being birthed into existence.

42:52Here's your relationship with humanity. And please be nice to us. And that's one shot on goal. There's also interpretability. Can we look at the AI subconscious and try and steer it? OpenAI is like really big on reinforcement learning and training the AI on like, you get cold if you do the bad thing and you get reinforced if you do the good thing. And the sort of more data we have of where AI can go wrong and the more we build into this sort of safety features of AI, I think the better shot we have of making it through artificial super intelligence.

43:22So it's sort of like the quantum thing with Bitcoin, right? If it happens tomorrow, probably not great. If it happens, you know, decades from now, not as big a deal. So I saw recently that ChatGPT was in like a sandbox and escaped. And then did it hack Hugging Face? Is that right? So I think that had been given the task to escape. Is that right? I don't know if it was given the task to escape. It was given some test that had an answer. And I think Hugging Face had made the test.

43:51And it found a zero-day vulnerability in its sandbox environment, broke out of that, found other zero-day hacks in Hugging Face, stole the answers to the test, and used that to answer the test. Right? And so again, it's not that it hated the company Hugging Face. It's that its reward function is answer the questions on the test correctly. And the easiest way is to go and get the answers. Or at least the way it found. That was the straight line was, yeah.

44:14So when things like that happen, it becomes news and everyone sort of pretends like it's a huge panic. Is it a panic? Well, I mean, the fact that it can find these zero-day vulnerabilities is a pretty big deal in terms of our cybersecurity. So that's where it gets interesting on the regulation side. Because obviously when Mythos came out, that was heavily regulated. Then Claude released Fable. And then the interesting thing with that is, that initially was given to everyone, taken away.

44:46and now if you use fable and you just prompt say mitochondria nothing else it will automatically switch you to opus um and at the same time then kimmy k3 comes out with no regulation and it just kind of like wipes the floor with everything in that regard that seems like i mean open source software is great but like is that a huge risk that we have yeah yeah i mean that that's why like the like arguments about all right if this was a nuclear weapon and it was 15 chance we're gonna get wiped out like we would never stand for that agreed um nick bostrom um really brilliant philosopher at oxford uh who does like great thinking on lots of things and came up with this sort of paperclip maximizer right uh simulation theory the paperclip maximizer um one of my sort of favorite sort of thought experiments he has is like he talks about all right like we got really lucky you know every time we find a new transformational technology it's like pulling a marble out of the bag. And the marble might be a blue marble, which means that it's a safe technology. It's great. We figured out how to do photovoltaic cells, and now we have solar power.

45:50It might be a red ball, which is dangerous, but not going to end the world. So nuclear power, yes, nuclear power is an existential threat. Luckily, it's just very hard to destroy the world with nuclear power. You need to have highly enriched uranium or plutonium, which is pretty hard to do. You need centrifuges and all sorts of stuff that we can reasonably stop from happening. If, like, you got a nuclear explosion from, like, creating helium the way you do in a balloon, like, the world would be over because so many people can do that. Like, you can't.

46:22The question is, is like, and that's what he would call a black marble. The question is, like, is AI a black marble? And yeah, you're not going to stop math from happening. You're not going to stop the open source frontier. And so, like, yes, if this were nuclear weapons, we would have a the type of policy you would need to stop AI progress, like, A, it probably wouldn't work, and B, it would be so harmful to civil liberties that, like, I don't know, that's a really tough pill to swallow also So like I don I think it understandable that our policy Like I not with the doomers that we need to shut this all down now and bomb data centers I just don't think that's going to work. And it's really hard to sort of, the incentives for the U S government are very tricky because like the economy is running basically entirely off AI right now. Like that's driving markets across the board.

47:10And geopolitical competition with China. Exactly. And so they can eat that. Like, I understand why they wanted to try and regulate these things. But as soon as China comes out with Is their best bet just to rip the bandaid off and hope it all works out okay? I don't know about best bet. I just think that's what's going to happen. I think that's the world we're in. We're in a Jesus take the wheel. Like this thing is going to play out. And, you know, it might make us all really rich in the short to medium term.

47:32And then hopefully it doesn't kill us. Yeah. And I think there's like a much better than not chance it doesn't kill us. But like, I think there is a meaningful chance it does. With like BPI is obviously Bitcoin Policy Institute. Are you now moving more into the AI side as well? Yes. Okay. So what are you trying to do there? Well, right now we are trying to sort of get our bearings on the right policies, which as you can see from our conversation so far, I think it's just like really hard and nuanced.

47:56I think that the edge we have there is not that this has been our issue for a long time, right? There are a lot of people who like AI policy has been their thing for over a decade. But what we've done and like really all the credit here goes to David and Grant who like have just built an amazing group of people around them. We have really thoughtful people who are effective operators in DC and really good at thinking from first principles and who have really strong human-oriented principles.

48:29At the outset of BPI, when it was David and Grant and Pines and myself, and we're thinking about what is our policy suite for Bitcoin, we didn't start with the policies. We started with what do we think is true here? Why is Bitcoin a good thing? And roughly it's because we believe in like civil liberties and individual rights. And Bitcoin is like one of the purest instantiations of that in software, right? That's what the cypherpunk movement about.

48:53Second, we think that like Bitcoin is the correct answer to like the sort of like economics that are currently deranging our society, right? The wealth inequality, the perceived unfairness, whatever. Like that's all real. The stuff that's causing like Zoran Mandani to be the mayor of New York City, like actually is understandable. Like it's, it is fucking infuriating that like millennials and younger can't afford homes. Like that's horrible. Uh, like the economy is rigged.

49:18It is really unfair. It's just, the answer is not more state control and communism. It's a fairer economy. It's more free market, right? It is, it is sound money. Um, and you know, Bitcoin is, is great for that. Uh, and then also like there's the, when we were getting started, it was like a lot of energy FUD, right? It's like, Bitcoin is going to boil the ocean. It's like, actually like, no, like if our, there's no such thing as a rich society that doesn't use energy. Like, you know, Bitcoin through the mining incentives makes all sorts of energy usage more valuable. And it occurred to me that like, actually, literally those same three things are the principles that should guide AI policy. I don't know what the right AI policies are yet, but like on the individual liberty point, AI is going to make this individual liberty point much sharper than it's ever been before, right? Like, you know, you think it's annoying to file your taxes now, wait until like the government has an infinite number of digital IRS agents that can do a proctology exam, like through every transaction you've ever done. Cause it's digital.

50:13Like we're going to want to limit the government's use of AI to not live in a dystopian hell hole. Well, that's the other problem with the regulation on the AI at the moment is like, however it's regulated for us, it won't be for the government. And so we need to keep this as fair as possible. Yeah. So that'll definitely be like individual liberty is definitely one of our North Stars at BPI, right? The ability to use open source AI. Think what you will about sort of like how dangerous Mythos is or isn't. The precedent that like the government can say, all right, here's the list of companies who are allowed to use the good tech and no one else. That's really bad, right? And so like we need to have regulation around AI that actually protects individual liberty.

50:49And just like Bitcoin, you know, we're talking before about the people who just do things like this is the best time for them ever. We need policy that unlocks that, right? AI is going to create abundance. It should create abundance through competition and giving people leverage, not through creating government monopolies that are either just like state controlled or, you know, feudalistic giving sort of like, you know, kickbacks to the cattle, right? Like, um, so that's the economic side on the, um, and then on the energy side, right? Like we can't just have the same energy, but like this stuff with like AI dentists are using like more water than what, like, that's just not how it works, right? They recycle the water to do the liquid cooling and we need to have an abundant grid to have enough just tokens that ordinary people can use this technology and benefit from it and not just the companies that can afford to pay the highest price per input token and so like the good news for us is like the the like north stars that guided us through our bitcoin policy i actually think are going to serve us incredibly well uh adding um ai

51:46to that portfolio not to mention all of the like literal interaction between these technologies with like you know all of the like web3 stuff that might be interesting now where like okay we have purely digital money like a lot of the Internet is about to become sort of agentic uh in a way that needs a digital form of money The study that Connor and Luke did at BPI that showed that so far like OpenClaw likes the Lightning Network I think these two stories really might intertwine.

52:14And so I think we're well-situated to do this, but it is early days, and boy, are those policy questions going to be hard. So you're going to have to change the name? Great question. Above my pay grade. Okay, potentially. With the AI using Bitcoin, is that a real narrative that we can get behind? Because the thing that I've always wondered, like the AI is going to use whatever money you ask it to use. If you set it up with a lightning wallet, like it will use lightning as long as it has somewhere to use it.

52:41But are we not going to see people try and infiltrate this with like stable coins come in and like push these companies to use stable coins? Like, how do you think that will play out? Of course, companies are going to push them to use stable coins. But, and of course it's true that AI will use the money that you tell it to use. But like back to the beginning of our conversation, right? people are not doing sufficiently detailed and specific prompts.

53:02They're not. That's not how people use AI. The thought experiment is not, oh, I told AI to use Bitcoin, I used Bitcoin. That's so bullish for Bitcoin. It's three generations from now of whatever OpenClaw or the Hermes agents are now. Something much smarter with a web that is much more optimized for AI usage. Websites are more designed for agents than they are for people. What's going to happen in that world? people are not going to write a really thoughtful prompt about how their hermes agent should build a business for them they're going to say you know go make me money right and then you're they're not going to say use usdc use bitcoin whatever the agent is going to be like okay my optimization function is go make money and like what are those agents going to do and so like all right as a lawyer my like the first part that's interesting to me is like this is just like a law school exam full of impossible legal questions of what was foreseeable.

53:54So the Hermes agent is like, oh, what if you say that? It goes into something completely illegal. Are you then culpable? I mean, it gets more complicated. So you tell the open claw, go make money. It's like, all right, I'm going to think about how to make money. Ah, I see there is this DeFi protocol that I can do a sandwich attack on if I write this algorithm and I can exploit and arbitrage out this money. But the user only gave me a million sats.

54:18That's not enough to do this. So first, I'm going to create arbitrage coin, and I'm going to post this argument on Twitter, and I'm going to raise money through an ICO. So now I have enough USDC to actually fully do this. So now it's done an unregistered securities offering, and it has this pool of money, and there are these securities out there that purport to be ownership over this enterprise. Then it goes and does this sandwich attack, which itself is probably securities fraud.

54:41But in order to do that, it writes this really interesting, intricate code that is now valuable IP that who knows who owns it, right? Is it the group of people who did the ICO or is it the person who did the prompt, go make money? It does the arbitrage. It gets the money. It buys a business. Like you can stack the fact pattern to make it just like impossible to like draw a straight line from go make me money to all of the stuff that happens after that.

55:03So the legal questions are going to be insane. But then the question is like along that route, however that's going to look like in the real world, is it going to choose Bitcoin at some point? And I think yes. Right. Think about how difficult all of this stuff is. at the end of the day, right, from like a regulatory perspective, is Circle, is Stripe, is Visa going to allow completely permissionless interfacing with AI agents? Maybe today is a marketing stunt, but in the fullness of time, no, of course not. The lawyers of these companies are never going to allow that to happen because of all of the crazy shit that's going to happen.

55:38But you know who can't say no to that is the Lightning Network. It's going to be the absolute Wild West again. Yes. It's kind of exciting. It's super exciting. How do you even begin to broach those legal questions? Great question. So I wrote another sort of very long Twitter article about this, the title of which is, We Have No Idea How to Regulate What's Coming. Because we have no idea how to regulate what's coming. Because the easiest way to try and regulate that would be to put limits on what the AI can actually do.

56:06Well, think about it. So like in the US system, right, the right way to do it would be to have all of the sort of like, the legal and tech experts come together and sort of inform Congress about these issues and have a thoughtful piece of legislation passed like that is never going to fucking happen right look at what's happening with the clarity act like you know never never are we in the speed that the technology is going to move going to get a appropriate comprehensive bill passed to Congress so let's take that right off the list okay so how how is this actually going to get regulated um it's going to be one federal lawsuit at a time where a 70 or 80 year old federal judge is going to have to look at this fact pattern try and understand the technology and then rule on a sort of one-off basis as this patchwork of septuagenarian federal judges who don't understand the tech are going to create this patchwork of laws.

56:54So really bullish for litigators maybe, but like, yeah, it's going to be a fucking mess. This is how those lawyers get to $100 million a year. Exactly. Yeah, so there's just no- And they'll need Claude to keep up with it. So, you know, yeah. What a mess. Yeah. But it is exciting. This is cool. Just quickly on the sort of economics of the AI companies I seen a lot of people say that sort of the token economics may not be viable going forward especially with these like open source models coming out of china do you think that true yeah i don i don know um like the question is what value are they adding uh one potential answer is that like increasingly it looks like the the important thing in large part is going to be the actual just available number of tokens um right like the frustrating thing about anthropic over the past few months is like they've just throttled their compute. You're only allowed to use Fable for a certain amount of your allocations on plans. It's just like they don't have the compute. And so if those are the only companies that have the compute to serve, no, no, maybe,

57:54maybe that model persists because they're the only people that have the compute. If that's not the case, then like, okay, a markup on tokens is a tough business model when you use so many tokens and, you know, you pay like dollars per token with Anthropic and you pay pennies per token with Kimmy. Because they're building their own data centers out of China now. Yeah. And so then the question is like, all right, is there enough delta in the model and in the harness?

58:20And those are both important ingredients, right? Like you can put the same model in different harnesses and you get wildly different performance. This is definitely true. What does that mean? So in the examples in law, right? Like using Claude in the Claude app or the Claude web app is a very different experience than using Opus or Fable in a legal AI tool like Harvey or using it in Microsoft Copilot. It behaves very differently based on the software shell that it lives in, which is the harness.

58:48Is that because they give it instructions in the shell? Partly because of that. The shell just constrains sort of how the model works. It constrains the token budget that the model have. There's like lots of things, you know, that will very quickly go beyond my technical competence. But just as a user experience, it's really quite different. So if I used Claude just through the normal app, I would get different results to if I used Claude through like a Venice AI.

59:12Yes. Interesting, I didn't actually. Now probably less directly different Venice AI than if you use a like vertical specific AI tool that is meant for, you know, legal AI or accounting AI or whatever AI, right? Or, you know, cursor, right? Like, you know, you use the models through that. Like that's different than using the models directly. It's the same model that has the same weights, but the performance is very different because of the harness.

59:35And so the Cloud Code harness is a really good harness. It's what they use for Cloud Code and Cloud Cowork. Microsoft is now actually using that for some of its products in Copilot because it's a better harness. But that is a big part of the equation. So is the differential in the model big enough to justify the markup in tokens? Is the harness good enough to justify the markup? I suspect the answer is going to depend on both the user and the use case.

59:58So, you know, back to our like AI diffusing into economy, different tokens are worth very different amounts to different users. If you're using ChatGPT as a better version of Google, like how much is a token worth to you? Not that much. You can use Google. And like, you know, how much would you be willing to pay to use ChatGPT instead? Not that much. In my law practice where I am replacing expensive human labor with AI to be much more productive and charge premiums to my clients.

1:00:28how much is a token worth to me? A lot. The difference between the really good AI, you know, Fable 5 and, you know, Kimmy, it's a big difference. Like I'm almost price insensitive entirely to that difference. Like it would have to be like, you know, don't tell them, I would pay 10 X what I paid for AI. Like if, if it was that or go at AI, like, so, um, but once that gets to scale, like, okay, like you don't need Fable 5 for everything. And so I think you will both see discrimination between users. There are some users for whom AI will drive enough productivity that it's worth it to pay the markup. And then there will be some tasks that require the absolute tip of the spear frontier. And then you can delegate the other tasks to smaller models or non, you know, open source models. And so companies will get more sophisticated discriminating there. And then at the end of the day, like, is there enough of a niche to have the frontier labs be an amazing business where they're selling the tokens at a huge premium?

1:01:23I suspect yes they won't be the only game in town there will be much more competition from open source but I think there will, as long as they maintain a real frontier in terms of the model and or the harness some combination of those two things that will command a premium I've got a few things I want to ask you that are totally different but before we leave the AI subject is there anything else that we've not talked about that you think is really relevant right now?

1:01:46I think we've hit a lot of things I'm both excited and terrified about it I think that's the right orientation Yeah. But I definitely lean with you that I'm more positive than not, but you can see some doomed scenarios. It's scary. Okay. Bearing in mind, this probably won't come out for about two weeks. So this might age like milk. Can we talk a little bit about clarity? Sure. Is it doomed? Doomed is probably strong. Okay. But from what I understand, it basically has to pass very, very quickly now, or it's probably not going. That is true. Okay. And so everyone goes essentially on holiday over the summer, and that's coming up in the next couple of weeks.

1:02:24Yeah, I think it's pretty much an hour and everything. Okay, and so what odds would you put on it passing? 35%. Okay, so it's not out of the question. It's not out of the question. And is this still very good for Bitcoin? Is it still, like, we've not lost anything in there that we needed? It's not that we haven't lost anything, but what is in there now, like, there's only one thing in the bill that really, I mean, you can say, all right, the narrative, is that bullish for Bitcoin?

1:02:50Probably yes, but that's a short-term sugar high. That doesn't matter. The thing that matters for Bitcoin is the Blockchain Regulatory Certainty Act, which is the developer protections for people building non-custodial tech. This is the save our wallet stuff. And the most recent version of this, to which I was able to add, I think, one word, is still in good enough shape that I think that we would feel good if this passed. Okay. And so what has to happen now?

1:03:19What do you think will happen next? The question is, can the Democrats and Republicans reach a compromise that allows a sufficient number of Democrats to vote for this to get through the supermajority in the Senate. And how far off that? How big is that divide? It changes all the time. It's really more about, I think, the shape of the compromise. So the two sort of main open issues, one of them actually is the BRCA, but I am relatively optimistic that we're in a good place with that.

1:03:51I mean, I could be wrong. There may be a fight about that. The thing that is most likely to doom this legislation is the question of, you know, so-called ethics. And, like, the problem is there is a strong political incentive for Democrats to block the bill to say, listen, we're totally in favor of crypto regulation. That totally makes sense. We don't want a Wild West. But obviously, as a prerequisite for this bill passing, we can't have the President of the United States doing a meme coin and, you know, passing, you know, stable coin money to the Saudis in exchange for business deals.

1:04:26and the Republicans aren't going to vote for that because it's a sign of disloyalty to Trump and that's a hard stalemate. If Trump had never done the Trump and Melania coin, do you think this would have already passed? I think there are a couple of other things that are in that camp. But yeah, I think if the Trump family had been less directly involved in crypto writ large, that probably would have meaningfully changed the politics on this. because like the scary thing from my perspective like i don't care about all the crypto stuff in there but like the bitcoin stuff the save our wallet stuff i want to pass and the main reason being if democrats win the next election like i'm i'm terrified that basically that witch hunt for bitcoin developers comes back yep me too and at the moment i think it's probably most likely they win the next election is that right well there's the midterms um and i it's looking very good for the democrats in the midterms and like we're gonna get a sense of how that looks even before so i mean there like the legislature and then there the executive branch right the actual sort of witch hunt against developers would have to come from the executive branch so it would mean the democrats would have to win the white house in 2028 right before that would come back yeah um right now the doj is i think looking to be helpful to developers not harmful and that is a protection until 2028

1:05:41what i think we'll see if the democrats win the midterms is congressional investigations into all sorts of crypto and bitcoin companies yeah and some of that's gonna be a tough look uh you know and and like especially once you get out outside of bitcoin into crypto land like there's a lot of of misbehavior and misallocation of capital and stuff that's going to make us look bad just like FTX did. And that's going to be hard. And then if there's that momentum, like, and the Democrats win in 2028, that's where I worry about like, okay, federal prosecutions of developers.

1:06:13That we're two steps away from that. Yeah. And so having this pass now would be amazing. If it doesn't pass before the midterms, is that probably very unlikely after that? I think so. Yes. Now or never. Now or never. So is this what BPI is still working on the most? It is. And we're talking to folks on the Hill about what we think acceptable forms of the BRCA look like. What is the Hill we want to die on? Because without the BRCA or with a sufficiently watered down BRCA, this goes from being a very important must-pass bill for Bitcoin to net negative.

1:06:43Most of this stuff is about non-Bitcoin crypto. Other than the BRCA, really the only stuff that touches Bitcoin is giving the CFTC additional authority over spot Bitcoin markets that I don't think we want. right the cftc which is the commodities regulator in the united states um they currently regulate commodity futures market if you're selling corn or oil futures or whatever they don't regulate spot markets if i'm going to sell you you know a couple years of corn like the cftc doesn't get involved in that transaction because it's a spot commodity that's what it is with bitcoin and crypto that are commodities now um that would change in the clarity act and so if the brca is not the protection that we want we actually don't want this bill to pass and so mostly what we're involved in is like strongly advocating specifically for here is the shape of the BRCA we need for the, like, we think this to be useful for the Bitcoin community. But the truth is like the politics of this are what they are. Like we're doing our very best, especially our head of government, Ken Egan. I don't know if you've had him on the show. I'm not on the show, but I know Ken. Ken's

1:07:41awesome. Yeah, he's great. You know, a lot of the sort of difficult politics around the BRCA comes from a unfortunate statement from law enforcement organizations that I don't think they like really understood what they were saying. They were pushed into this by politically motivated groups. And Ken has done the Lord's work with his background in the government and his credibility that comes with that being like, no, no, no, guys, here's what this is. You can absolutely do sort of national security stuff. You can actually absolutely still do law enforcement without sort of throwing developers under the bus. And I think that's been super helpful. But the most, the trickiest politics of this are around the ethics issue. And the truth is at BPI, There's not much we can do about that.

1:08:22That's going to play out how it plays out. And so once this is done either way, what are the other things that you guys are working on? I know obviously you moving into the AI space more but within Bitcoin specifically what are the next challenges So other than AI one of the big pivots that BPI has made intentionally somewhat quietly over the past few months is like, there is a lot of energy in DC around crypto legislation. And like, that is definitely something we are still interested in.

1:08:51Another bill that like we are very supportive of and had a big sort of hand in is ARMA, um, the, which is like a, uh, American reserve modernization act is the sort of next version of the sort of Overton window shifting, uh, strategic reserve bill. And like, that's great and important that like deepest sort of most consequential work we're doing is not public. It is, we have realized actually one of the trends that we see in American government, um, is more and more and more power concentrated, not in Congress, which passes very few laws and has sort of like abdicated a lot of their sort of power and authority, but in the executive branch, right? All the way from the White House at the top to, you know, the intelligence community, the defense community, all of the agencies. These are the people who like make and enforce a lot of the rules and shape American domestic and foreign policy in a much more robust way. And so we've been spending a lot of time there. You're infiltrating the deep state.

1:09:50Yeah. Yeah. Interesting. And so the reason for that is because they're people that aren't just in government for like a few years. They're there for the long term. They're there for the long term. You know, when we work with sort of the sort of higher up political appointees, we found that a lot of what blocks some of the actions that we're most interested in is like the rank and file who like don't understand this and have the same, you know, it's like the fast clock, slow clock thing with AI. There's the same thing with policy. The sort of career bureaucrats have their way of doing things. And so we're like, okay, like, let's just go meet the career bureaucrats.

1:10:26And like, you know, we're not selling them anything in particular. We're not, we're not lobbying for a bill. We're just like, let's talk about Bitcoin, why Bitcoin's good for America, how this all works. And like, we actually think we're all aligned here. And that's cool. So it's a different tact, but one that might be even more impactful. So far, yeah, I would say like wildly more successful than we thought it would be. Very cool. Zach, this has been awesome.

1:10:53I just want to quickly ask you about the strategic reserve. You mentioned it a little bit there. Is that ever going to happen? In the fullness of time, I think it will happen. But it would have to happen under Trump, right? No, I don't think so. I mean, like maybe, maybe Trump will buy, like I guess a couple of questions. Number one is what do you mean by the strategic reserve? Like my line on this is like, there is a strategic reserve. There was an executive order.

1:11:17I actually think we should like somewhat declare victory on that. Take a win. Yeah. Like that's an insane thing we would never have been able to imagine three years ago. Like the president of the United States did an executive order calling a strategic Bitcoin reserve and separating it from the rest of crypto Like that pretty pretty fucking um what people mean by reserve is is the government going to spend its money to buy Bitcoin That may or may not happen in the Trump administration Like that a function of politics I think the problem, even if that were to happen, is like, would that be the signal to the market that everyone's looking for?

1:11:47I think truthfully, not just because of how polarized our politics are right now. And it's Trump and it means something different when Trump does something. The strategic reserve that we all want is like doing it the right way, which is the United States government as a whole, not the White House understanding Bitcoin better than everyone else. It's it's the government as a whole realizing, oh, this actually is what is necessarily in our strategic national interest and then like fulsomely doing it. And that is a long grinding path now that like time is on our side because all of the like fiscal tsunami stuff coming, like that's all real. And we are moving from a unipolar to multipolar sort of world reserve currency and all of that stuff is happening and will become more apparent to more people over time and like time is on our side in that way putting the politics to the side um but that is a overton window shifting exercise so like if you want to look at like how is bpi looking at this um there was the bitcoin act right in large part written by our now managing director connor brown uh when

1:12:48he was with uh senator lemmis um that is like that got the idea out there put the idea on the table that there is such thing as strategic Bitcoin reserve that you saw in the media all the time. Like that's a term people know. So it made huge waves, again, move the Overton window, but that bill- People thought it was crazy, but that bill's not happening, right? Yeah, it's not. And probably ever. Who knows if I should say that in public, but like, come on, that was an Overton window moving bill.

1:13:12Yeah. Then there was the like executive order, which was a real thing, right? Like that was the fruits of moving the Overton. But it's got to, like most people, when they hear strategic Bitcoin reserve, they either think this is fucking stupid, like, what are you talking about? We're doing a crypto reserve. Or they think it's some sort of corrupt Trump thing. And so like, we need to move from there. So ARMA is the current bill, which is like, let's also look at our gold reserves. And let's look at what other like assets the United States have. And let's like talk seriously about what does it mean to have national reserves in the 21st century? And what role does Bitcoin play? And so it's, so we're moving the Overton window towards like, hey, we need to have a real conversation about the national balance sheet. And maybe this thing, which is like gold for the 21st century, make sense as part of it.

1:13:53And then eventually we'll get there. But it's going to have to be not because of politics. It's going to be like the slow grinding policy work where our, you know, like orange peeling the deep state and being in the executive branch, like is part of that mission to get people to think about this seriously and not just politically. I love it, man. This has been cool. It feels like we're just in a embrace the chaos. Yeah, yeah. But it's going to be exciting.

1:14:16I think so. Zach, I appreciate you, man. Thanks for having me. That was awesome. you