Beff Jezos on doxxing, e/acc, and why crypto will save AI

Like, our whole movement is about freedom of information, freedom of speech, freedom of thought, freedom of compute. It's very simple. And the fact that that was deemed dangerous enough, or I don't know, like, to want to suppress and provide the public leverage over me, I don't know, that's a huge red flag for me. I think that we were becoming a voice that was going to be a problem for the executive order. And there are very special interests behind that executive order. And they wanted to, you know, send me a warning shot, I guess. I think the keyword that was like, set off alarm bells for me was public interest. And, you know, I had just crossed 50K followers. And, you know, because the law is you can't dox people unless you're a public figure. Right. And I guess I crossed into— I guess there's a threshold, right? So is it $50K? Apparently, it's $50K flat, right? But to me, it's like, okay, if they think they, you know, if they think it's of public interest, and they're gonna go with it, they're probably gonna go with it. And so I'm kind of screwed here. I gotta make, you know, I gotta make a move.

So wait, is that, is that the journo group chat decides when you become public interest? Or do we like— I'm very curious, like how they rationalize Because to me, that if you want to be anonymous, you should be able to be anonymous. And doxxing you, what gives them the right?

This week on Upstream, I wanted to share a conversation I had with Beth Jezos, originally released on our other show, Moment of Zen. Beff, whose real name is Guillaume Verdun, is a founder of Effective Accelerationism and was recently doxxed in an article in Forbes. We're also joined by Bayes Lord, another founder of Effective Accelerationism, and Nathan Labenz, host of the podcast The Cognitive Revolution. We discuss anonymity and doxxing, the intellectual roots of Effective Accelerationism, the intersection between crypto and AI, and more. As a quick disclaimer before the episode, I'm an investor in xTropic, The stealth AI hardware startup founded by Guillaume. Please enjoy.

Let's go. How is that? Yeah, well, super happy to be here. It definitely feels weird. It definitely feels very weird. You know, it's been kind of a compartmentalized part of my life. You know, felt like a video game on my phone, more or less, right? Like having an alternative Twitter and having a very professional life day to day. You would never guess that I'm Beth from just like day-to-day interactions, frankly. And now the two have kind of bled into one another and they are one. And so I'm still pricing that in mentally, frankly. So it's been an interesting couple days, obviously a lot of inbound, more DMs I can ever count. And yeah, I mean, it's been a very Surprising reaction. Well, I guess not that surprising. I guess like, you know, tech Twitter kind of hates the establishment media at this point. And, you know, came to my defense, which felt quite good. But, you know, I'm happy to go into any part of it. You know, how it went down. You know, why I started Buff, anonymity. Anything, the startup, let me know how you want to structure this. Yeah.

So can we start with the sordid details of how they figured it out and what voice they sampled and how that whole sordid tale played out?

Yeah. I mean, there was a couple instances where there's reporters that, you know, think they have my docs. And I'm just like, you know, you can't, you can't dox people, you know, I, I do not want you to do this. And usually that kind of kills the story to some extent. But, you know, it's kind of been, you know, I've gone, I've been in the SF social scene in person. I know a bunch of investors, for example, they, some of them know, some of them have correlated the two. But obviously, I spent a bunch of time in Twitter Spaces. I have some online lectures from my days in grad school in quantum computing. And really, that's basically what they've what they've correlated. They use some— I don't know, I call it CIA technology as a joke, but, you know, they use some government-grade technology to identify my voice, you know, just because I have a 50K follower account that speaks truth to power. I guess it tells you that, you know, we're doing something right and that some people in the establishment, you know, want to have leverage over the leader of the grassroots movement that is for freedom and against top-down control, which is very scary because that's what they like.

You spoke truth to power and now they're going to try to speak power to truth, so to speak. But I'm curious, someone must have been tipped off because they didn't do an all-against-all voice sample. They started after this pairwise interaction because there was nothing online that would have suggested it. And then they confirmed it via that, I imagine.

Yeah.

So see, they must have like asked around, you know, I mean, it's kind of like, people know other people's doxes, right? Like there's kind of Twitter parties in person, and people go by their alternative names. So they know your face, and eventually they correlate things. But there's kind of a, yeah, there's an unwritten rule of like, you don't, you don't like talk to reporters, you don't share people's dox without their consent. I'm sure some people, you know, broke that rule. But, you know, I don't know what drove them to really break the story now. I think there were some latent variables there. But yeah, essentially, I think it was on Thursday, I get a text from some of my investors. They identified some of my investors. So this podcast is going to drop after we announced the round. So I'll just mention investors. But yeah, so some of our big Big investors get a text like, "Hey, I think this you know over this first reporter. I think it was Conrad has started correlating your identity with Beth." They didn't correlate all the the company because we had a you know I had a company changed the name now works Tropic. They hadn't correlated everything perfectly, but then they sensor fused across reporters internally, so so they didn't have enough to ship it on Thursday. But then on Friday, a different reporter Emily. Joint force filings, the track name changes. They went to my personal Facebook to, you know, to identify a photo I shared, like, on my friend, you know, friends-only Instagram of the party, you know, when we were on stage with Grimes and stuff like that. And they correlated everything, right? And so they just had me in this sort of checkmate, like, we have this, this, this, this, this, this. Talk to us, we're going to ship it. And I was like, all right, I got to get in front of this, right? Again, you know, I've been doing this deep tech startup for nearly a year and a half. And, you know, as the bio says, I come from Google X, you know, we're taught to be very secretive about what we do, because in deep tech, that's kind of the MO. And, you know, we want to be secretive for longer. But for all sorts of reasons, including national security interest reasons, right? Like our technology is pretty out there, and it kind of forced our hand, right? Like it correlated all identities, and also it also kind of doxxed the fact that I had founded the startup because I was kind of I hadn't identified that on my main account. So for me, it was kind of a you know a moment of panic because you know I want to do what's right for my company, right? I've been working really hard on this company. We were planning announcements in a couple months from now, and now we have to rush everything, right? Which is not great. But what I ended up doing, getting on the phone, getting in front of the story. I already knew exactly, like, that's the thing with reporters, like, they're so low entropy. They are all in the same typical subspace of stories. You can predict exactly what they're going to write just from the prior that they're trying to get you. And so I just disarmed every typical attack that they would try to do on EAC, trying to have second or third order guilt by association to some idea that's a derivative of another idea and neutralize that essentially entirely. I see we have Bazelord joining. There you go. What's up, Baz?

Can you guys hear me? What's up?

Yeah, yeah, yeah, we can. Awesome. Oh, we have Nathan as well. Oh, all right, here we go. What's up? What's up, guys? What's up? What's up? Yeah, good to see you again.

Great to see you.

I guess now with real names in my case. So yeah. So, so yeah, it's been an interesting, you know, I guess 48 hours dealing with this. I think overall, I mean, we can get into why I was anon for various reasons. But overall, it seems like it was positive to the point that where some are like, have a conspiracy theory that this was planted. But anyone—

I had that theory, Guillaume. I thought we were all getting conned and you were actually dropping this because I was gonna ask the question, do you think you would raise at a higher or lower valuation now than you did this round? I would bet the valuation would go up, actually.

Definitely higher.

Yeah, exactly right.

But I mean, I didn't— I wanted to raise round, you know, raise around just uncorrelated. I didn't necessarily initially disclose to my investors. I didn't want that to be part of the price or anything like that. I didn't know where it would go, right? I didn't want to correlate the 2 identities. It was really like orthogonal. And now my hand was forced to correlate the 2 and own it. And so I got in front of it. And of course, I'm going to harness any hand I'm dealt. I'm just That's that's what you're supposed to do, right? And now now we've harnessed it, you know pretty well. I I would say we'll see how it plays out, of course. And and now there's this conspiracy that it was it was planted. But you know in my case again, I just wanted to control the narrative. I knew the story was going to be really bad if I didn't get on the phone and and you know I actually I wanted to stall them. I wanted to give them enough content so that we had a couple more hours. We adjusted our website. to have a pseudo-launch the day of. So, you know, shout out to the team for being so adaptive. And, and, you know, as of, as of the moment of airing this podcast, we'll probably have announced the round for Extropic. So we rushed that over the weekend. So yeah, definitely not planned. But at the same time, it was always sort of a trap, right? In game theory, you want to make it like, you want to make sure that Like the incentive was to not dox me, right, for most people, because they knew that if they did dox me, now I could go on these podcasts with my real face, with my credentials, I could talk to potentially politicians, and I have credentials that have some, some firepower. And now more of a problem. So they kind of messed up. It's their fault, really. It was a trap to some extent. But, you know, I tried my best to, you know, stay anon. But, you know, again, voice identification technology, I'm not going to use a voice changer on Twitter Spaces like every day. Are you kidding me? Like, I'm not going to do that. And I would imagine maybe the first podcast we did, they used that voice and correlated it with my YouTube lectures. But it's probably how they got me. Yeah.

Can I still call you Beff?

Sure. That's easier. Gil or Beff. Gil is like the Americanized version of Guillaume. Very French, French Canadian. So you got me.

Super simple question. Did the reporter, journalist, whatever we want to call this person, did they give you a rationale why they thought it was a moral imperative to dox you?

I think the keyword that was like, set off alarm bells for me was public interest. And, you know, I had just crossed 50K followers. And you know, because because the law is you can't dox people unless you're a public figure, right? And I guess I crossed into. I guess there's a threshold, right? So is it fifty k? Apparently it's fifty k flat, right? But to me, it's like okay if they think they you know if if they think it's of public interest and they're gonna go with it, they're probably gonna go with it. And so I'm I'm kind of screwed here. I gotta make. You know, I got to make a move.

So wait, is that, is that the journo group chat decides when you become public interest? Or do we— like, like, I'm very curious, like, how they, they rational— because to me, that if you want to be anonymous, you should be able to be anonymous. And doxxing you, what gives them the right? Because, because if I was to go put the, I don't know, the address of that reporter online, and maybe now we're arguing over whether or not an address is too far Like, what point is too far? And then who gets to decide? That's my question. And I don't think anyone has given an answer outside of, I'm a journalist, so I have a special badge, used to have blue checks, now they don't. And then they get to decide who can be anon versus not. But if you were to do it to them, that's harassment, right?

Right. Well, I mean, to me, I was, you know, mostly focused, eyes on the ball, you know, I have to— my prime responsibility is to my company. And I was just concerned, like, hey, we're, you know, getting announced. We have to announce and get our stuff together, you know, without preparation, without heads up, only a few hours. So, you know, it seems like we were— it really seems like we planned this, but we truly didn't. So, but I'm just glad we adapted given the circumstances. But yeah, overall, I think like, I mean, this was clearly wrong, uh, what they did. And, you know, I, you know, there's not that much to have leverage over me. It's like, okay, cool. I used to, you know, work in quantum computing and, and have, you know, a normal-ish background. Um, but, you know, maybe some other people that are trying to speak truth to power and have their voices heard and want to use anonymity as a tool to speak truth to power, uh, you know, because there's a sort of power asymmetry, so you kind of equalize that when you're anon. Uh, they can't, they can't do that anymore, right, if they, they fear getting doxxed. And so it's kind of like they want to make an example out of me, right? And I mean, I think there was, uh, I think there was a congresswoman at a, uh, con— uh, conference, uh, a defense conference that Explicitly named out EAC as a sort of dangerous movement that needs to be suppressed with AI. That's the tweet. I'm not sure if it's correct. I still have to watch that clip. But to me, that sounds very dystopian. Like our whole movement is about freedom of information, freedom of speech, freedom of thought, freedom of compute. It's very simple. And the fact that that was deemed dangerous. Enough, or I don't know, like, to want to suppress and provide the public leverage over me. I don't know, that's a huge red flag for me. I think that we were becoming a voice that was going to be a problem for the executive order. And there are very special interests behind that executive order. And they wanted to, you know, send me a warning shot, I guess. That is my theory.

But one thing I couldn't understand, and maybe, Beth, I just haven't read all your tweets, I couldn't understand, A, you obviously have a very distinguished academic background. In fact, I was reading your quantum tensor flow paper on the way in. Ages ago, I worked in experimental quantum computing, like 15 years ago, back when it was in its total infancy.

No way.

Yeah. And I knew a lot of the theorists back then, like Michael Nielsen, and I actually dated Julia Kempy. Anyway, long story. Back in the early days of computing, but obviously wasn't really smart enough to continue it. But in any case, I was reading it with a lot of pleasure and it's it's odd that they would consider this a bad thing. It's obviously given a lot of intellectual credibility to the e/acc movement. Your background is pretty good. It's pretty distinguished. You've done a lot. A lot of co-authors. That's why I thought the conspiracy— and I don't really believe it, to be clear— but I thought maybe he actually provoked Forbes to do this because this is a brilliant thing. You bring actually so much positive social capital to the movement. You come off as this edgy, hip, poster who kind of had, you know, was pulling the world's tail. It's like, yeah, what's the downside to this? Like, I don't understand. Yeah.

Yeah, for me, I can let you know why I was anonymous. I mean, originally, I was at Google X working on some pretty secretive technologies. I had access to, you know, the top of the food chain there. So I, you know, I couldn't necessarily tweet anything like political and stuff like that. My tweets were watched, which is, you Normal. But, you know, I wanted to have just a separate account where I can really let myself go, kind of like offload thoughts. And also, I reached a point in quantum computing where I was respected enough that people would cheer me on or they approve of my ideas or whatever. Oh, so smart, whatever. And that was kind of like, kind of annoying me that like, do they like my ideas or do they just want a job or something? And I just wanted to put my ideas in the arena, right? It's like I just wanted people to evaluate my ideas for their own, not look at credentials, not like weigh my opinions dependent on status. I just wanted my ideas to be evaluated on their own. And, and to me, like growing a movement and having ideas resonate from scratch, uncorrelated from my original identity, Like, I mean, of course that felt like, you know, it felt like New Game Plus, as we say in video games, right? You restart the video game from scratch, but you have kind of all your intellect and memory and your gear, but you're trying to rank up again. So I started from scratch, an account with like no followers, and grew it to 50K before. Now, now my— there's a stat boost from the status signals of credentials. I personally very much hate credentialism. I think I would have been an entrepreneur 5 years earlier if it wasn't for credentialism, especially in deep tech. I got very frustrated when I tried to do a startup at 25, and, you know, I would get like, who are you? And, you know, I just went through big tech, the gauntlet of grad school and big tech to prove myself and prove to others that I can do things so that I can raise the capital to realize the visions I had. And so, you know, I've always— I think Bayes is on the same page here. You know, we always wanted sort of like EAs is all about like gatekeeping, gaslighting, status signaling, at least to us from our perspective. And, you know, EAC was about kind of bottom up. Everybody has access to opportunity to build, no gatekeeping, no sort of status signaling. It's all about building, right? And we want to maintain that. And I think that, like, I was afraid that if I, you know, used any sort of creds, that would dissuade people, you know, from participating or feeling like they can participate. Because the thing is, everybody starts somewhere. And I was once, you know, that kid that was really smart, had a lot of ambition, but didn't have the creds. I was just straight out of school, and nobody gave me a chance. And I want, you know, the next, you know, genius to be able to, you know, start building their dreams right away. And I just personally very much hate gatekeeping. Even though nowadays, I've gone through the gauntlet, I've paid my dues. Now a lot of doors are open for me. But, you know, I'm just trying to kind of pay it forward in a way that, you know, young— my younger self would be appreciative. So really, that's why I was anonymous. I just wanted to get, you know, get my ideas evaluated in the arena in an uncorrelated fashion from my credentials.

Hey, we'll continue our interview in a moment after a word from our sponsors. Real quick, what's the easiest choice you can make? Taking the window instead of the middle seat? Outsourcing business tasks that you absolutely hate? What about selling with Shopify? Shopify is the global commerce platform that helps you sell at every stage of your business. Shopify powers 10% of all e-commerce in the US, and Shopify is the global force behind Allbirds, Rothy's, and Brooklinen, and millions of other entrepreneurs of every size across 175 countries. Whether you're selling security systems or marketing memory modules, Shopify helps you sell everywhere, from their all-in-one e-commerce platform to their in-person POS system. Wherever and whatever you're selling, Shopify's got you covered. I've used it in the past at the companies I've founded, And when we launch merch here at Turpentine, Shopify will be our go-to. Shopify helps turn browsers into buyers with the internet's best-converting checkout, up to 36% better compared to other leading commerce platforms. And Shopify helps you sell more with less effort thanks to Shopify Magic, your AI-powered all-star. With Shopify Magic, whip up captivating content that converts, from blog posts to product descriptions. Generate instant FAQ answers. Pick the perfect email send time. Plus, Shopify Magic is free for every Shopify seller. Businesses that grow grow with Shopify. Sign up for a $1 per month trial period at shopify.com/tornberg. Go to shopify.com/tornberg now to grow your business, no matter what stage you're in. That's shopify.com/tornberg. What's next for for Yak? Where do you see it going from? from here?

Yeah, well, now that I guess I am doxxed, right, one thing that was kind of stopping any sort of like organizations or incorporations of any kind of any sort of org or institute was that, you know, we're like going to register with our real names somewhere, these institutions, and that would have doxxed us. So now we have that opportunity. Will we take that opportunity?

TBD.

But personally, I mean, I think right now there's sort of a lack of theory and educational material for people to understand complex systems and self-adapting systems like capitalism. And our whole thesis is that bottom-up self-adaptation is superior to top-down control. It's very easy to convince a crowd like, put me in charge, I will do this action, it will have this impact. It's much harder to convince a crowd of like, hey, if we all act in these microscopic exchange laws, the emergent behaviors that like we have a highly functioning society, right? And the more sort of research you create there, the better and more educational material. So, we might start a research institution there and fund some grants. I would love to fund some open source hackers. Obviously, we believe in open source software for AI. We think that open source software is a hedge against the oligopolization of AI, right? There's a couple players right now that obviously, now that they're in the lead, they have all sorts of interests in closing down AI, outlawing open source models. So, you know, sort of like ensuring freedom of compute is something, uh, we might create orgs to, to, um, sponsor hackers, but also, uh, uh, potentially, you know, have some influence in Washington, right? Those are some next steps. Obviously, like growing the movement on Twitter, that's going to keep going, also known as X. And overall, yes, it gives people an attack vector now that I am doxxed. That is unfortunate. But I do very strongly believe in what we talk about in EAC. And I am willing to go all the way, whatever it takes, to defend these ideas. And I won't let sort of pressure or reputational pressure affect me. I think I'm quite robust against such attacks and if they want to come after me, then bring it, I guess.

Can I ask one question, Erik?

Sure, please.

You mentioned one thing there, Guillaume, that I found sort of— sorry, I'm a Francophile, so I'm going to use your full French name.

Yeah, that's good.

And also mentioned the Prix Goncourt completely out of context. Sorry, that's one of the running gags on the podcast. You mentioned one thing about the open source. So it's funny. Dan and I come from the crypto world where, as you know, decentralization is like this religious mantra that I think is often overexpressed actually. And in fact, I think Dan, I think you quoted it in Farcaster, right? Of sufficiently decentralized, which is almost like sufficiently a trinity or something like that. It's enough of a certain theological concept, but no further. But I always found AI, and I'm very much seeing that world from the outside to be clear. I'm kind of a tourist, but it always seems to me actually to be totally centralizing. And I think the hijinks you saw at OpenAI with literally the little junior high school drama between 3 people somehow wrecking the cutting edge of AI is something that literally couldn't happen in crypto for a bunch of reasons. I mean, you might cite the example of SBF, but then Dan would instantly throw a fit and say, but that was actually a centralized exchange, not a decentralized exchange, which is true, actually.

Right.

And in the fullest decentralized version, speaking of anonymity, so I work in a crypto company. We We've had an anon employee that I didn't know who they were, and I would just pay them, right? And we work with the anon founders of a protocol, and they never turn their video on. We have no idea who these people are, and yet we still do business with them and transact, right?

Wow.

So that's— yeah, yeah, yeah. And it's not even that weird. And then I've met them also in person. Like, it's a similar thing. Like you mentioned earlier, how, like, you know, it's really weird that these journalists are such losers. I did this tweet interview set up that's like, these people are losers, right? Like, they never actually go to the parties where you meet. You know, Beff or who— like, I know a bunch of Twitter anons. I think we all do, right? And you meet them in person, you know who they are. They introduce themselves. This isn't a big secret. It's almost like that meme where like the loser's in the corner and saying, hey, do you know that Guillaume is Beff? And like everyone dances. He's like, dude, we know. Shut the fuck up. Like he actually cares, right?

Exactly.

Like, it's like, it's like such a loser thing to say, right? Um, but don't you think like AI goes the other way, right? Like, just as an example, like I'm, I, I was using ChatGPT today. Okay. I'm pro-AI. My best friend is an AI, right? But the fact that OpenAI instantly does a deal with Microsoft and instantly bakes itself into literally the most octopus-like corporation in human history, it's a little weird, right? It's a little bit like pinning to the other side of the spectrum instantly. So I'm just curious if you feel that that binarization of it is valid, and feel free to say that I'm full of shit, or if there is another AI vision that you're sort of stretching towards that maybe is a little bit more decentralized and not so naturally sort of centrifugal to—

Yeah.

I think that self-organizing systems tend to organize themselves in hierarchies. So, it's sort of like you can imagine a tree or a sort of fractal sort of structure. You have your cells form your body and then you have groups of humans form a family and then a corporation and then a city and then a nation and so on. So there's a hierarchical structure to organization, and there's this hierarchical structure to sort of control, right? There's, there's some control system at the head of your body, it's called your brain. There's a control system at the head of the family and at the head of the corporation, nation, and so on. You know, it's all about the balance between, you know, it's all about engineering fault tolerance to corruption, right? If you have a one-to-all connection in terms of control, you have a control system that affects all the nodes, in the system, that node has a fault, the whole system has a problem. Just like OpenAI, you can decapitate the leadership, not physically, but suddenly you're in control of the organization. And now everyone that were using their APIs have a problem, right? Because they have an ideologue that they didn't vote for in control of a product that they depend on.

Yeah.

Right. And so decentralization is about sort of diffusing, obviously having decentralized loci or plural locus of control so that you have fault tolerance to corruption of these control nodes. Of course, having a fully greedy algorithm where everything's disordered is not optimal. So it's all a balance between the two. It's a balance between centralization and decentralization. That is what is like fundamentally optimal. The reason we're fighting for decentralization is because we think, you know, right now there's a tendency towards over-centralization of AI, and we're very worried about that. And so we need to push things in the opposite direction at the moment. I think that fundamentally right now there's a lot of alpha in just scraping the whole internet, centralizing it, And having centralized training. I think at some point that alpha will be saturated, right? Most companies will have a model that compresses basically most of the data that's on the internet. And then you're going to have AI that seeks to capture data from the real world. And for that, you have to be decentralized. You got to, and you want to have the intelligence perhaps at the edge. We're not there yet. There's still alpha from like centralizing data and soaking it all in and having one big model that compresses it all, because intelligence is more or less compression. But I think over time, we're gonna see decentralization in the form of sort of like, first of all, you're gonna have a personal assistant, you know, like Humane, like Tab, or I don't know, there's gonna be all sorts of AR assistants, I'm sure. At first, like, you're gonna have, you know, the intelligence is gonna be in the cloud, but the data acquisition is gonna be on the edge. But eventually, people are gonna want their own Compute that they own and control, I would imagine. And maybe there's compute in the robot directly, right? There's no need for a connection to network. I think that's going to be sort of the decentralizing effect right there. And you can have federated learning over a fleet. I think people are working on this. Obviously, right now, in order to train the biggest one model to rule them all, Uh, the centralized approach, centralized compute and centralization of data, has a huge advantage. I think a big problem as well is that right now the centralized approaches, they scrape the whole internet, uh, which includes your data, and then they rent it back to you, right? Bit by bit. And I think the future is about, uh, crediting people for the data they contribute to AI systems and sort of distributed ownership, right? I mean, they're starting to do this with sort of GPTs. OpenAI, but it's just an early start. But I do think there's going to be a very interesting wave of startups right now, like from crypto migrating to being infrastructure to align incentives for AI. We're not a crypto company, by the way. Just people thought we were going to launch a coin. We're not a crypto company at the moment. I have nothing against crypto. I love it personally. But, you know, I think that crypto being sort of the value exchange network and programmable incentives for kind of collaboration in AI, right, to have sort of decentralized research labs, I think that's gonna be a very potent application of crypto. And it's just something I fundamentally want to encourage because I think that if we have a future where there's only a few companies that are like government, that have the government mandated monopoly or oligopoly to serve AI models, then you have a sort of like single point where a few people get to control the cultural priors of these LLMs, what they're allowed to say. So it's an information supply chain attack because people won't ask each other what is the truth. They're going to start asking the LLM what is the truth. And if, if you, if you change what it says, then you're controlling people's sources of information and you're controlling people. So it's cybernetic control of the population through this information supply chain attack by proxy, by, by, uh, you know, saying that, oh well, we're responsible, we should be put in control of what these, uh, LLMs are allowed to say, right? Um, so we definitely want to fight against that, and one solution is to erode their market power by having alternative solutions that are just as good or, or nearing the same, same level. Of course, that's not going to happen if we don't leverage sort of capitalist-like technologies for value, for incentive alignment, right? And yeah, very bullish on this space. Again, not personally involved at the moment. We're building a hardware company for AI that's fundamentally new. We can get into that at some point today. But yeah, overall, very Very worried about the over-centralization of AI, started making my voice heard online. And I think, you know, there was an event where I got to interact with the chair of the FTC. Maybe that got noticed. Maybe that got me in trouble. I don't know. But, you know, there are ways that our voice is being heard in Washington. And, you know, our point is that Uh, this sort of, uh, fear-mongering, uh, and, and doom is really sort of, uh, a very nice cover for very subversive regulatory capture by the incumbents, right? Like, oh, AI is dangerous, put us in control, we're the only ones who are responsible, uh, you know, you're not allowed to have more compute than we do, you're not allowed to have open-source models that would, you know, erode our market power and our pricing power. Uh, because they're dual use, they're dangerous. Uh, that's all, that's all bullshit, right? They, they cooked, uh, they cooked those, those proposed, uh, regulations, uh, for their own advantage, right? Um, and so we got— yeah, we gotta fight. I mean, they're gonna push, they're gonna try to push these regulations through. Um, and so that's why we're not, we're not stopping the fight. And I don't think, I don't think my doxx is stopping anything in terms of like In fact, it might accelerate things.

Hey, we'll continue our interview in a moment after a word from our sponsors. Have you ever wondered where your donation could have the most impact? In 2007, a group of donors had that exact question. But when they sought out information from charities to help them answer this question, they instead received cute pictures or unhelpful stories. Their experience led them to create GiveWell, an organization providing rigorous, transparent research about the best giving opportunities. GiveWell has now spent over 15 years researching charitable organizations and only directs funding to a few of the highest impact opportunities they've found in global health and poverty alleviation. Over 100,000 donors have used GiveWell to donate more than $1 billion. Rigorous evidence suggests that these donations will save over 150,000 lives and improve the lives of millions more. GiveWell wants as many donors as possible to make informed decisions about their giving. decisions about high-impact giving. You can find all of their research and recommendations on their site for free. You can make tax-deductible donations to their recommended funds or charities, and GiveWell doesn't take a cut. If you've never donated through GiveWell before, you can have your donation matched up to $100 before the end of the year, or as long as matching funds last. To claim your match, go to givewell.org and pick podcast and enter Upstream with Erik Torenberg at checkout. Make sure they know that heard about GiveWell from Upstream with Erik Torenberg to get your donation matched. Again, that's givewell.org to donate or find out more.

I mean, I'd love to deepen into the crypto-AI overlap, Guillaume, thing that you hit on, but if we want to move on to the regular— because I, you know, again, I'm seeing the AI world from the outside and obviously I use it and I've been watching, you know, some of Karpathy's videos and stuff. But like what you just said, right? This business of paying data owners for their training sets, like Fortunately, we do have a public ledger of ownership that's natively financialized with an underlying value model that does this very well. And in fact, some people are even working on blockchain attribution solutions that figure out where this thing came from because this other person used it and it's worth X amount. So I've often thought about if there is some sort of crypto-AI collision, which I think is inevitable, just to shoot the idea down just for one second, why wouldn't you say pay Reddit for its data literally in a direct deal rather than all this crypto craziness? Will it ever be the— this is the classic, not to say this is a bad idea, Guillaume, but the classic bad ad tech idea from Web2 was like, oh, pay users for their data. Well, Brave does that, and it turns out that data is worth $3 a year. Brave is a great product and lots of people use it. I use it, but they don't use it because of the $3. They use it for a bunch of other reasons. The data that you actually own that you express with your browser just isn't worth enough. Even if you could get things down to literally the micro ETH, would I care that I'm getting paid because the model's getting trained on my Substack? Even if you could figure out literally What is the actual value per query that my data contributed to? And I'm sure the incrementality of that is super hard to figure out, but you guys would know better than I would. But even assuming you could figure it out, that's going to be literally worth like 30 cents a year. So would it even make sense to wire all that together?

Well, yeah, I mean, I've had some various ideas in this space. I guess I could just broadcast them, YOLO. Essentially, I think I think you can price the value of data according to how much information gain the system gets from your data. And there's some very specific mathematics for that. And that can give you a share of a model's future profits, right? So similar to how eukaryotic cells own a fractional ownership of the success of the greater organism, right, through DNA. And that's better than prokaryotic cells like bacteria. I think that the future is people owning fractions of a model, right, according to what they contributed to it. And I just realized that, you know, there might be 10 different tokens now that spawn using this idea. But I don't know, for me, it's like I have more ideas than— I can act upon in one lifetime. So I'd just rather broadcast them. And, you know, this is something Bayes and I have been talking, you know, we considered doing something in this space, but, you know, I think at the moment we have our hands full with changing the entire AI hardware software stack from scratch beyond the transistor, right? Which is a significant undertaking. So, um, but yeah, uh, I, I'm, I'm really— I think, I think there's really gonna be some, something interesting that comes out, um, and hopefully it can erode away the power from sort of the centralized players. Um, and not, not that like, you know, they're necessarily nefarious, but, you know, every, every metaorganism's control systems act in the metaorganism's best interests, right? It's like the realpolitik, right? And I think that's the thing about IAC, that is that we, like, we cut through the bullshit, right? It's like every agent and subsystem is going to act in its own best interest. It's going to do whatever it needs to do in order to secure resources and/or utility towards its own growth, right? And acquisition of resources, period. That's just how everything works in nature. That's just reality. And it's like, okay, now that we have this reality, how do we create the system that harnesses this to create a sort of emergent altruism where we reach greater prosperity, find new optima of the techno-capital machine that, you know, allows us to support more humans on Earth and to scale civilization to the stars, right? And so, Anyways, I got into my typical Twitter Space ramble there. But yeah, I do think that, again, super huge opportunities in the interface of AI and crypto. And that's partly why there's kind of been an informal handshake between crypto and IAC. Brian Armstrong, you know, they put out an ad for Coinbase. It was basically an EAC ad, frankly. Really, we're kind of fighting the decels and the centralizers, right? The incumbents, those that seek to control everything and to cause inflation, to secretly tax you. And they're scared of sort of bottom-up decentralized revolutions. that they don't control that causes deflationary pressure, right? And so, you know, there is interests in deceleration. There are kind of interests of the control systems that are kind of greedy at the cost of what they are controlling. And it's kind of— we're kind of the autoimmune response, if you will, to the control systems, right? And, you know, we're causing a bit of inflammation to the brain now, or the brain of whatever this whole thing is. And it seems like, you know, they tried to apply a Forbes anti-inflammatory pill, if you will. So, yeah.

You know what IAC reminds me of, Guillaume? Have you read John Perry Barlow's cyberspace essay from back in the day, like in the '90s, way before your time, probably? Okay. But you have Baseline? Do you remember it fondly? I remember reading it when I was young and getting into the internet and I found it was the most amazing thing.

I think I was only shown it like 2 months ago, but very based and definitely has vibe overlap.

And at the time, the internet was forming as cyberspace, which now seems almost cliché or cringe almost, was this edgy space that you would meet. And he has this line in which he basically addresses it to the weary giants of flesh and steel, i.e., the industrial giants to which the internet represented an alternative. And I think a lot of EAC reminds me of that same rebellion, of course, except the weary giants of flesh and steel are actually of silicon now. It's actually the old internet that has gotten kind of old and boomerish, to which this is a rebellion. Part of the reason why I'm in crypto, I came from the sort of FAANG world and working in all that world, and I felt that that was all slowing down and becoming the man, actually. And crypto was like the only thing that reminded me of the early Web2 days, in which it was like, Like, if you don't have people coming after you and getting severely pissed off because you're building something, you're building shit. I mean, to be— or something that's like not important, right? Like, if you read the story of Uber, like literally every taxi commission, people in Paris were kicking the shit out of Uber drivers. You know, all of Spain shut down Airbnb. Like, you know, if you don't have major governments pissed off at what you're doing, you're actually not building anything particularly important, right? And there wasn't a lot— there isn't a lot, in my opinion, in internet consumer outside of some of the things we discussed that meet that. So anyway, it just reminds me of that vibe. of the cyberspace vibe. And I think, obviously, I think we need more of that. So it's cool.

We're definitely the most cyberpunk movement out there, hence the Arasaka Tower vibes here. And we had the party with Grimes after the OpenAI Dev Day, and the point was keep AI open, right? And You know, it's got to feel like a bit of a rebellion because it kind of is, right? And then you have the engineers from these, these big players, they come to the party and they're like, man, this is freaking cool. I kind of want to join the rebellion. I don't want to work for the Empire. What the hell? How do I join? Right? And so that's how it starts. So, um, we'll try to keep that going in many ways. But, um, yeah, I mean, it's definitely, it's definitely very— it feels like we are in the cyberpunk future. It's, it's kind of been, uh— surreal how we've gotten here, frankly. Um, and, uh, yeah, no, I'm just grateful to, to be here, uh, at this point in history. Frankly, it's an exciting time.

To quote the Steve Jobs line, why join the Navy when you can be a pirate? Which is something you used to be able to say about Apple, but I don't think that's the case anymore. And I have a little bit of experience there.

That's the thing, right? Like, what is, what is the sort of mind virus that infects organizations, that decelerates everything, adds way too much process, way too much bureaucracy, and then it grows kind of like a cancer. It's kind of like middle management kind of grows and eventually takes over from the founders. And now it's kind of like this, you know, Borg of some kind that, you know, these corp— these huge corporations, you know, have decisions by committee for everything. There's tons of process. It's just very hard to get anything done. And really the answer to that is disruption, right? You got to have a free market. You got to have free market competition. And if an incumbent is too slow, it gets disrupted by a startup. We saw that with OpenAI, right? Frankly, I don't— Yeah, I can't say too much about Google, but clearly there was a sort of slowdown in its speed and the talent eventually saw that and sort of migrated to startups, right? At least for AI. And now they're getting disrupted and they're kind of in trouble. But if they remove that mechanism where bottom-up challengers can dethrone and compete with incumbents, if they remove that ability, then we don't have this kind of self-correcting mechanism and we'll just live with kind of monopolies that are, that are not shipping great product, and then everybody, the consumer suffers. And we don't want that, right? And so to me, it's kind of like, you know, you have sort of like the doomers pushing for centralization and control, and then you have the, you know, EAC pro-freedom folks that are more about antitrust, right? Like it's about monopolization. And so I think there's an interesting political landscape shaping now. I think, you know, it seems like some Democrats are more on the side of like AI safety so far. And allegedly Trump has said he would cancel the Biden executive order on AI. You know, EAC is not partisan per se, but we're kind of like, we have an issue that we care about, which is the freedom to compute, the freedom to do to advance technology. I don't know if you guys want to get into 2024 discussions, but—

Oh, we never talk about politics here, Gio. Never, never, never, never. I'm totally joking. We do it all the fucking time. Go ahead.

Just on the question of why big tech is so bad, I won't relitigate the whole thing. The episode we recorded earlier today and the piece that Nadia and I put together talks about some of what I think about this. But I do think part of the issue here is just simply, do you have actual improvements that you can build onto things in the world, onto systems in the world? And I think there was kind of this growing lack of capacity to actually do things that were sufficiently ambitious. And then also when you have the effect of a bunch of capital accruing, Companies naturally end up becoming kind of inward-facing, navel-gazing. People are incrementalists and it's just kind of locally optimal for people to kind of be a little bit lazy, L+1 until you retire. People are obsessed with doing FIRE and they want to ride around in a van or whatever instead of build cool shit. Yeah, I think part of the shift here is just that We have a bunch of new technologies coming online and new capacity with AI especially. And people just see that it's time to try to dissipate that alpha and do the work. There's real work to do.

By the way, if you'd given me the betting odds of which FAANG company came out with the biggest open source dedication to AI and it being Facebook, I don't know if I would've bet on my former employer.

To be honest.

I don't know if you agree with that characterization.

It makes a lot of sense.

These Frenchmen, they're so based.

Yeah, all the AI Frenchmen, they're all for freedom. But I mean, look, every agent acts in its own self-interest. I think that if you're number 3 or 4, I think right now the leaders are really OpenAI, Anthropic, I would say. I think that's not a controversial statement. You know, if you're kind of competing for number 3 or 4, like, I think the point is you want to kind of equalize the playing field and kind of band together to try to beat, you know, the top 2, right? And so I think that's the thing about open source is it kind of groups everyone together to collaborate on iterating on the open source architecture and tooling and products to compete and erode away the market leverage of the top players, which they're eating, I don't know, 90% of API calls, if not more, if you include Anthropic, frankly. And so, I think, yeah, I don't know, it just makes sense from a fundamental standpoint. Facebook has a lot of data and they have a lot of compute. They have a lot of great researchers. And, you know, I'm really happy they're contributing to open source, but hopefully they don't, they don't stop. But at the same time, you know, it does cost them a lot. It does cost them a lot. And it's kind of like they're, they're giving to the community this, this value. At the same time, people are open sourcing models. They're not open sourcing the training code, right? So it's not fully open source AI, right? So they still have their moat of like how to train these things. And we saw Uh, you know, couple engineers that did Llama left and then started Mistral because they have that sort of artisanal know-how of how to train these beasts. And, and then they raised hundreds of millions right out of the gate. So it's very valuable knowledge. So they're still keeping some alpha there, uh, which is good. Uh, that makes a lot of sense. Um, but, you know, I, I do think that, uh, it's kind of like communist AI to some extent to have like just open source everything, everything's free. Like, data costs money, compute costs money, engineers that are talented cost money, uh, more and more, in fact. Uh, and, uh, until we have a sort of proper, like, decentralized/centralized system of incentive alignment, I think, frankly, with— through crypto rails, uh, I, I don't— I think it's gonna be tough for open source to, to really, uh, compete. with the big centralized players, right? Um, just like how much capital is being injected in API calls towards, uh, the top players, you know, pales in comparison to the rest. But, um, you know, if there's actual capital flow like that gets reinvested then in improving open source models, uh, beyond just us scrolling on Instagram, then, then I think that's, um, I think that has a shot.

So you think crypto is actually critical to the survival of open source models?

That's interesting.

You think that's—

Yeah. I haven't been very public about that, but it's kind of a thesis I've formed over time. I mean, I don't know if it's necessarily crypto. I would say crypto-like thinking. You could just use Stripe if you want, but some sort of way for people to collaborate on models and pooling data, compute, and capital to train these things and know-how.

Well, if I could interject, the greatest accumulation of GPU that is not in a data center was the Ethereum network until they moved to proof of stake. And that was a crypto economic system that worked pretty damn well. Maybe Bitcoiners would tell you different, although they have their own set of compute, a little less useful for AI. But I do think we don't even have to imagine it. It's not science fiction. It did exist. That's actually what Nvidia's stock price was originally bumped up on was crypto before obviously AI has taken it to new heights. So I would imagine a world where everyone's gaming PC, and maybe you just don't get to the level of compute that you can with the H-series in data centers, but if you were to take every GPU around the world and then be able to kind of do something like SETI@home in a kind of decentralized manner, maybe you do actually have this kind of— you can't go drone strike the data center because I am the data center. Come and take my GPU, right?

Yeah, I think that is the dream. I think that at least right now in the way we're doing these big models, you need parallelized high-bandwidth multi-GPUs. It's very hard to shard the models without a significant slowdown over the network, right? And if an alternative is like 1,000 times slower or more pricey, then than the centralized incumbent. Like, in the free market, like, people want to support decentralization, but if the product is not, like, competitive with the centralized player, like, at the end of the day, people pay for what works and has the right cost-benefit analysis. But, um, I think that, you know, there's gonna— there's gonna have to be sort of algorithmic breakthroughs. But you can imagine where people, instead of having just a gaming PC, they have maybe, you know, bigger boxes that have beefier GPUs and they can run maybe a whole single node.

Yeah. Like this is what George is working on, right? George Hotz, right? Yeah.

So George Hotz is working on kind of the hardware infrastructure for that, right? Selling beefy GPU boxes of several GPUs. I don't know if he's—

What does he say?

It's like a petaflop at home. Is that the idea?

Right, right, right, right. And that seems like an attractive kind of like node to run potentially a future protocol. I would encourage a lot of people to do the research here. I think there just needs to be a lot more players in this space. And I think the AI people, you know, are gonna have to talk to the crypto people. And there's going to be kind of going to be a moment there where there's going to be, there's gonna have to be bridges built there in terms of the language. But, you know, I'm I'm optimistic. I don't know. That's my prediction. I think the merging of crypto and AI, you know, this kind of anti-centralization of AI movement is going to kind of combine forces with crypto. Again, I don't own anything in any protocol right now. So not talking my book, just like my prediction. But yeah, no, I think it's exciting. Personally, I'm really worried about You know, okay, cool. Like, great. We've decentralized the algorithms and the sort of like distillation process of AI, like distilling data into neural weights, right? That is the software process that we're talking about. That's what OpenAI does. That's what Anthropic does. Great. Maybe we figured out how to decentralize that. You're still buying your GPUs from the same supply chain that is down to, you know, NVIDIA, TSMC, ASML. Right. ASML is, for those not familiar, is, you know, the, the machines that do the extreme ultraviolet lithography. So the most advanced process nodes to, to, to, to create the GPUs used with NVIDIA. NVIDIA doesn't build their own GPUs. They, you know, work with TSMC, which is in Taiwan. And, you know, that's where they're built. And so again, you know, I'm just thinking about fault tolerance of the system. And right now, our supply chain for AI hardware is absolutely not fault tolerant, and might be co-opted by totalitarian leaders from, you know, the CCP, right, and might cause a major global conflict because of that. So, you know, what, what, what I'm working on is sort of like decentralizing the AI supply chain by fundamentally changing the substrate on which we run generative AI completely, right? Beyond transistor-based compute, beyond digital compute. And once you have this fork in the tech tree, there's all sorts of opportunities that pop up for far more energy efficiency, for far more speed, and eventually far more density. And that's what we're going after. So, we're still kind of in the— we're still living our values. It's just we're going after the much harder Problem of hardware engineering.

Production is really expensive right now. Model production is very expensive for all the reasons that you listed. And I think that crypto, probably crypto across AI, has a lot of potential in the nearer term with the one-to-end part of this, where you're diffusing the capabilities of the model in the same way that you see open source already doing without crypto. I think in general, Yeah, this is a broad thing that we've talked about a lot, which is there is so much work to do to put intelligence into every corner of the world where it's needed. And the idea that you're going to do that with a few thousand people at a couple of centralized companies is probably wrong. And I think you can do this with APIs, but a lot of times you need more privacy than that for your data. Maybe you don't have internet connection. There are a number of constraints that come up in the real world where you don't want to have these centralized APIs. And I think there's kind of a natural balance there. Nathan brought up this paper, I think this DeepMind paper, which is they recently came out with some innovation in doing distributed training, some kind of federated learning scheme. I think It's probably worth emphasizing the main problem here right now, which is just the network latency, as Guillaume said. I don't think anyone has figured this out. As far as I know, nobody has solved this problem, making it cost-effective. And I think it's probably worth underscoring that. Yeah, right. So if it takes, let's say, 100 days to train a frontier model, if you are off by a factor of 2 or 5 or 10, in your distributed scheme, that pretty much kills it. It's a really long horizon. By the time you finish, it will already be not state-of-the-art anymore. This is really problematic. I think it's also just about the DeepMind paper. It's using data parallelism, which a lot of people have figured out. I think BitTensor figured this out. But you also need model parallelism. You need to shard the model over nodes.

Yeah.

But you're screwed if you can't fit the whole model in one computer that you can plug into a wall outlet and not have to have a home nuclear power plant, it's really hard to do model parallelism or it's impossible to do data parallelism for a big enough model. And if you're doing model parallelism over the network, you're also screwed from the interconnect. And so kind of creating the hardware substrate that's going to allow for decentralized AI, we have to solve from first principles, how to increase the density of intelligence in space-time, in terms of space-time and energy, from the first principles of physics. And that's sort of like what we're, what we're building, what we're trying to enable. So, you know, like, that's what— that's why I think like, if there are decentralized crypto protocols of all sorts, if we have the best AI hardware that has the highest density and runs most energy efficiently, obviously, it's in our, you know, we will have a lot of customers, right? Similarly, we could be the NVIDIA for Ethereum in that case, right? We don't, again, we don't have a crypto protocol. But I think that that's a very hard problem. You need to assemble a team that's basically like a Manhattan Project-like team. And, you know, we came from Google X, Google Quantum, AWS Quantum, all sorts of institutions, IBM, Google, Meta, et cetera. And we assembled quite the team and we're going after the hardest problem in AI right now, which is how do you embed AI into the physical processes of the world the most efficiently? So you got to really understand the duality between physics and AI. And that's what we're That's what we're after. And so it's kind of like, to me, that seems like, you know, okay, I would love for there to be a protocol that is competitive right now, but we need to solve for the density of compute first. Look, maybe George builds crazy boxes that are water-cooled and you use like 2 plugs in your house, and maybe that's just enough to run certain models. That's great. I think, I think we need to go much further. I think there's orders of magnitude, there's still orders of magnitude to go in terms of energy efficiency. and density for compute, especially for AI. And that's what we're solving.

What is this mysterious—

I just wanted to say thanks for coming on the podcast, but you guys continue the conversation.

Thanks so much.

Cheers. Guillaume, I have to ask, what is the hardware though? I mean, are you going to bring quantum computers to market doing AI? I mean, I have to ask.

It's definitely not quantum computing. Yeah, we all got jaded. by quantum computing being kind of like nuclear fusion. You know, the timelines are very long. You know, fundamentally, a quantum computer, you have to cool it to absolute zero, ideally.

Yep.

That's obviously physically impossible. And so, what you have to do is this process called quantum error correction, right? So, you have to identify faults from, you know, the universe jiggling and screwing up your computer, the computer's operation, and you got to identify faults and filter them out. So, you're pumping entropy. So, it's basically a fridge, right? But this sort of algorithm that is your fridge occupies like 99.9999, well, not that many nines, but quite a few nines of your computation overall. And to me, that seems very inefficient. And a lot of us were kind of full-stack architects or engineers, the software, the hardware, and the compilers. for quantum computing. And we'd look at the roadmaps, we'd look how long it would take, and we kind of got depressed to some extent. And so a lot of us were like, actually, maybe there's ways to use this noise instead of it being a hindrance. And so we set out to do a different type of physics-based computing that is not quantum mechanical, that is specifically focused on generative AI. And for now, we kind of got our hand forced with this whole doxxing situation. So, you know, we're going to be still nebulous about what exactly it is we're doing. But, you know, we have quite a few scientific publications in preparation, right? So, but yeah, overall, you know, we think there's a different path forward, a fundamentally new way to compute. It's going to be like quantum computing, but a new type of physics-based computing. And ultimately, we learned a lot from quantum computing in terms of how to program, how to have programmable matter, how to have, how to integrate, you know, these sort of physical systems into a deep learning program. You know, that's what we pioneered with, you know, the software I did at Google with my CTO now, Trevor. We did TensorFlow Quantum, right? And so now it's about how to really have programmable matter and figure out the tightest embedding of AI in the physical world, which is exactly what the doomers fear most. And so, you know, as a joke, we kind of say Bayes, for example, is one of our principal FOOM engineers. And we just announced that today, that Bayes is part of the team. And ultimately, I think that there is no path forward where the ultimate form of AI isn't built. And I think that we could talk about human augmentation and sort of the transhumanist path forward. I'm very bullish on that, and I would love to find ways to fund more efforts in sort of human-machine collaboration and augmentation. But yeah, overall, the EA thesis has always been like, hey, I don't think banning GPUs is going to do much. The tech tree is going to mutate around whatever your restrictions are and is going to adapt similar to a virus kind of mutating. The techno-capital machine just finds a way. It's kind of a system that's almost alive, right? It's entropy-seeking, it's exploring all sorts of configurations, and it finds one that allows it to grow. And, you know, might as well build it and try to make it technology that's, you know, helping humanity scale, right? Our goal with EAC and really with the technologies we're building at our company is to enable sort of AI, the ability to perceive, predict, and control our world, you know, at all scales, including the nanoscale, such that we can you know, tackle the real problems that are in the way for us to scale to Kardashev Type 1, which is a scale of civilization in terms of its energetic expenditure. You know, it's kind of like, to us, it's like the ultimate denomination of like societal progress is like the Kardashev scale, because every other measure like GDP, or like it's based on dollars, you know, you can like fudge the numbers, right? It's not anchored in like physical reality. Similarly, our cultural thesis is that you should evaluate your actions in terms of how do you think it's going to contribute to the growth of civilization down the line rather than subjective measures of utility like hedonism, like how much pleasure is this giving people on average, which leads to kind of spurious optima. wireheading or TikTok and whatnot. So, yeah.

You've got me in total suspense though, Guillaume. If it's non-transistor-based computing, you've got me thinking what the hell it could be. I'm guessing.

Yeah, yeah, yeah. I mean, we're working on it. We still have to remain somewhat secretive. And, I guess there's going to be a lot of interest now. Again, we want to be in stealth until we had more to say, more to release. But for now, we were keeping things pretty close to the chest. But I think there's certain discoveries that you make that, you know, you're like, okay, I don't think I'm gonna be able to unsee this. And if I saw it, someone else will see it. And we should just move as fast as possible to bring this forth. And for us, it's like, okay, how do we make this the most impactful to the advancement of mankind. Like, let's tackle the actually hard— the hardest problems that most people don't dare to tackle. Um, and, you know, it takes some— takes some courage there. Uh, it takes some fanaticism to some extent. I, I think to me, the fact that we have this framework of EAC, it's, it's a really powerful motivator, right? It's like, what am I contributing to, right? Like, we're kind of in a Well, you know, I went through a sort of whole phase of, you know, I grew up Catholic and then I was not a Reddit atheist, but I was the typical atheist, studied math in undergrad and then went through a nihilist phase. But then I think like understanding that the whole system seeks growth and entropy production and that's kind of the way things are and that process is what created life, civilization. and the technologies we enjoy, like, okay, we want to contribute to that. So, so having the knowledge that you're contributing to something greater than yourself gives you that sort of like infinite dopamine well to grind through the long nights, to skip the holiday dinners, to just file more IP, right? Like, and, and put in the hours. And, uh, you know, I, I think like scaling, you know, everybody has their own cultural or religious framework that, that provides utility to them. For us, you know, For the EAC community, I think I think it has had utility for a lot of people to get out of this sort of nihilistic rut that they were in. That you know the world was going to collapse. There's only doom and gloom. Everything's going to get worse. Put us in charge. We're going to fix it. Maybe oh we didn't fix it. It's because you didn't give us enough power. And like we're just like you know it's kind of like well you can make parallels to SF politics. I'm in SF right now. How's that?

How's that, by the way?

Oh, it's Night City. I love it here. Uh, you know, uh, it's, it's, it's truly Night City. I, I like to say our office is in Arasaka Tower, uh, so we went with this sort of cyberpunk vibe. Um, it's, it's very much like, uh, Night City the video game, right? Um, no, but you see the sort of like, if you let the decels in charge and you let the movie play out, this is what you get, right? Um, you know, you get— and these cells are kind of, uh, uh, um, like they have much more power if they're attached or in control of something that's very prosperous, right? And generating a lot of value, like similar to big tech, right? Uh, you know, there's money printers and, and then these sort of, uh, uh, folks that seek power kind of take over take over, uh, but, you know, they, they can, they can cause a lot of damage. And I think, I think there's a, there's a cultural turning point. And hopefully, you know, it doesn't matter if you're Republican or Democrat, like having people that are anti-tech progress, anti-tech-first solutions, and, um, you know, uh, sheathing themselves in, in under the COVID of virtue to gain more power and doing things out of self-interest and, and LARPing that it's for the good of, of many. Um, you know, I think people have had enough of that and are ready for a change. And, um, you know, of course, like, as technologists, like, we propose technological solutions, but ultimately, like, like, we believe in the power of technology. We believe in people having agency and not accepting that this is just how the way things are. You got to accept how they are. Things are just going to get worse. you know, screw your dreams, kid. Just accept it and give up. Right. And it's like, no, fuck you. We're not— we're going to make the better future happen. We have agency. We can build. Get the heck out of our way. You know, we're going to make the better future we want. And you need a sort of like, fuck you optimism. Right. And that's, that's EAC in a nutshell. And Hopefully, it keeps growing because we think that's what the world needs in many ways. And, hopefully, we can accelerate SF and then we can accelerate the rest of the world.

Can I ask you one follow-up? And then, Nathan, I know you want to jump in and I want to give you time to jump in. But, I did want to follow on one question that Kio mentioned. When you mentioned being raised Catholic, I was also raised Catholic. I eventually converted to Judaism and kind of rejected this sort of default secular majority of society. That's a whole other story you've done a bunch of episodes on. I do think it's one of my pet theories that religion never goes away. It's almost like energy or momentum. It only gets transformed and you get worse, crappier versions of it.

Yeah.

The reality is you can't actually navigate the world without a sense of metaphysics of some form. And in fact, if you don't have an explicit metaphysics, you can't even do empiricism well, because then you have to change reality to suit metaphysical ends. I mean, to cite the COVID example in which they denied the lab origin hypothesis because It serves some metaphysical end. And so they stopped being able to do empiricism as well as they could have. And then that whole COVID story, if there wasn't actually a way to have a conversation about, do we save the kids and screw the old people or vice versa, you couldn't have that conversation. So instead it was about the science, but really it was a moral conversation that nobody could have because we weren't all morally or even religiously on the same page. That's a whole separate conversation. But I do think it's interesting, and I do think it's one of the things that's a little bit lacking in tech. I mean, you see these homespun religions like I would say Burning Man is a little bit culty. Startups are obviously a little culty. And I'm not saying this is a bad thing, by the way. No, it's good. Yeah, I know it's a good thing, but it's a little incomplete, right? In the sense that like, well, it doesn't quite tell me how to raise my kids or who I should marry. I don't know if— and I know I'm putting you on the spot a little bit to like give me the gospel, bro, but like what would be— and direct me to a set of writing if there is one, but like what would be the EA gospel or like that? What is the true, the good, and the beautiful in this world, other than obviously building for building's sake, which I think we all get? What is the bigger picture?

Yeah.

So at a high level, right, EAAC is all about figuring out whatever sort of cultural framework yields the maximal expected growth and scope and scale of civilization. And we don't want to be prescriptive. All we're giving is a loss function, and you can have your own hyperparameter settings, you're like, okay, I have a cultural heuristic. I think this is an optimum of this sort of loss function. This is how I want to do things within my tribe here. It's essentially a sort of like, think of like a very thin framework from which you can have like subcultures, right? And the idea is to have sort of culture be more or less like code, like on GitHub, you can kind of have a base framework and then you can add kind of commandments or add things you believe, or you can diff it, right? You can make forks and you can kind of keep track. And so we've seen, for example, Vitalik forked EAC and made some changes, right? And then he's like, this is what I believe, right? So really, EAC is sort of a metacultural framework. We don't prescribe too much. We try to prescribe the minimum. But, to us, it's very clear that the engine that keeps progress going, maintaining the sanctity of its sort of momentum and maintaining malleability, adaptability, and so on, it's crucial to maintain freedoms and maintain entropy in the system and accept variance rather than constraining things, crushing entropy, crushing variance. Because that leads to crystallization and catastrophic failure. And so, at a very meta level, we try to maintain variance in most parameters, but it's not all variance, no restraint, because that just doesn't work. It's kind of like running a system at very high temperature. It's just pure disorder. So, it's kind of always about finding the optimum balance between order and disorder, between entropy-seeking behavior, uh, and, and novelty seeking and sort of constraint and conservatism. Um, and so we don't have any one particular prescription of how to live your life, but some people like to make forks and have more particular prescriptions. And to us, it seems— at least my personal thesis on how cultures get kind of mimetically post-selected for— it's like whichever culture either confers its adherence, uh, and a better ability to grow. Or if a culture is more sort of viral, then it's going to be more likely to exist. That's just like by probability theory. Um, and so, you know, to us it's like, okay, if we have this sort of metacultural framework, people have all sorts of forks, and all these memetic— all these memetic forks with different parameter settings can compete, uh, in the— in a sort of cultural setting. But we should explore space of cultures and heuristics of, of how to live your life. So for example, another, uh, friend of ours is, is, uh, Bryan Johnson, and he has his own life heuristics, and he's trying to create his own cultural framework, and he has much more prescriptive ways to live your life. And he thinks we should, you know, lifespan should be much longer than what it is, and you should have, you know, longevity as a priority, right? Um, and so, uh, I'm, I'm all for this sort of renaissance of exploring all sorts of Neo-religions, neo-cultural frameworks for how to live your life. I think it's much needed because otherwise sort of parasitic mind viruses like the ones we've seen do all sorts of damage recently, including the decel class of mind viruses. That's a whole category. Yeah, they come in and they kind of, like you said, kind of fill in this gap in people's hearts, if you will. So, I think we're on the same page. And I think, you know, much more, you know, Lindy religions, right? Like, in a sense, like, that have been around for a long time, like, they're very robust, right? It's like a code base that has been through hell, you know, and back. And like, it's just, it's very robust, right? It's been robustified over a long time. And by the proof that it's lasted a long time, it's It's a good heuristic. So, you know, that's great. But I think like some people want kind of, you know, modern variants, right? And they want to contribute to shaping new subcultures. And so we encourage people to form subcultures. So really, we're kind of like only setting the hyperparameters of the whole thing and people can set like more finer-grained fine-tunings of how they want to live their lives. So, there's no one way to go about it, but we do talk a lot about building because we think that obviously fundamentally building technology is a very high leverage way to use your time on Earth to impact the future scope and scale of civilization. And so, encouraging one another and helping one another in building technologies that have a positive impact, we think that's a good heuristic that should be in most forks. And so we encourage that. Some people think that's the only message, but it's not. It came from kind of a higher-level thinking. But yeah, cool.

Well, I think you're in the right city for exploring religions. I mean, the way I see San Francisco, it's really a petri dish for literally exploring every weird-ass thing that society wants to do, whether it be autonomous vehicles, whatever weird computation you're cooking up with, not having rule of law, for example, psychedelic drugs. Whatever, bring it. It's going to be cooked up here, and then from here, expand and diffuse whether we like it or not to the rest of the world. But I want to let Nathan in. I know he's probably been biting his tongue this entire time, and I think I'm the last host left standing. And so I'm going to invite Nathan to go ahead and comment and provide maybe the other side of it if he wants to.

I mean, my angle on the whole thing is I describe myself as an AI scout, and I'm getting more and more putting more and more emphasis on let's really try to figure out what is today, what exists, what can be done with it. If we are going to extrapolate, can we extrapolate, first of all, in a high confidence way into the short term, and then use those discussions as kind of the foundation for figuring out what we should do about the bigger picture questions, where I think inherently there's a lot more uncertainty. And one thing I haven't really heard from you guys, and I've pieced a little bit of it together from Twitter, but I don't have a great sense of what are your near-term expectations? Do you think we are headed for AGI in the next couple of years? Metaculus has it at just over 2 years. The leaders of Anthropic, for example, say that the leading developers in '25, '26 timeframe may create such a lead that no one will ever catch them. Are you kind of in that same headspace of thinking that we're going to see pretty radical transformation on just a few-year timescale is like my very first question?

Yeah. First of all, I don't like the term AGI that much. I think it's human-level AI or human-like AI. I think like calling AGI general intelligence, like human-like AI that was distilled from human-generated data, I think that's like very anthropocentric. And I think, you know, I work in physics-based AI, you know, inspired by physics and to understand the physical world. I've done so for 10+ years now. I think that intelligence is a much more general concept than just human-like intelligence. And frankly, I'm not scared of FOOM because, you know, again, I've worked on AI for engineering matter, drugs, simulations, biology, all sorts of stuff. It's much harder than people think. I do think it is disruptive for our economy based, you know, our knowledge economy of sort of human-like white-collar intelligence. I think there's still going to be a 10-year gap for physical intelligence, right? Robotics. I think our motor intelligence is much harder. It takes many more parameters. It took billions of years to evolve rather than, I guess, like, I don't know, 100 million for the neocortex. Probably got those numbers wrong, but it's ballpark. But yeah, essentially, I think I think there might be a disruption to our economy, but I do think that people will adapt. People will learn to augment themselves out of self-interest, right? And it's like, where will the system go? Every corporation is going to do what's in their best self-interest. They're going to maybe be 80% AI, 20% human, and each human is going to be augmented and control a fleet of AIs, right, that are doing its bidding. And that's okay. Maybe the human employees become the control systems for a fleet of AI to do most of the work, right? Maybe they become more kind of like the capital allocators or the executives of companies that are mostly AI for the execution layer, right? Maybe that's the future. I think we're heading towards interesting times, but I don't think there's going to be cataclysmic effect. I don't think it's going to end humanity. I think we're going to adapt. and the system will adapt. And this sort of— EAC is all about the— in the main, like, it's a faith and, and, and, and sort of, uh, worship of this adaptation of the Homo techno-capital memetic machine, the whole thing, right? Like memes, technology, capital, humans— it's all coupled, it's all adapting, it's always shifting. It's all— the only constant is that it's always changing, and it should be always changing, and that seeks to grow, right? Um, And so, you know, we have a faith that the system will adapt. It might be abrupt. Um, and so, you know, personally, I'm not trying to— uh, personally, I'd rather augment humans in orthogonal directions to human-like intelligence rather than try to replace human-like intelligence. Obviously, in an economy that's already shaped to take in human, you know, intellectual labor and do all sorts of produce products, having human-like AI is what's going to grow the fastest. So that's what's being built first. But I'm already pricing in AGI personally. And the technologies we're building in hardware and software all assume there's probably going to be a human-level AI within 2 to 3 to 5 years. And that's been priced in for me mentally. And it's like, now what? What's the next thing? And to us, it's like, We're going to seek to grok and perceive, predict, and control matter at the nanoscale. And then, we're going to seek to, again, increase the density of intelligence in terms of the substrate. That's what we're working on. And so, yeah, I think it's coming. I don't think it's going to be cataclysmic, but I think that people should start preparing and start integrating their business processes, integrating their personal, uh, life and their personal workflow with AI, right? There's gonna be the class of sort of the tech-forward people that embrace AI and, and sort of do well, right? They integrate with it. And those that refuse to use it that don't do so well, right? Uh, but, but the important thing is that if people get to own a piece of the system in which they're contributing At least they own a piece of the future as it grows. Whereas if there's only centralized players from which you have to pay rent constantly and maybe they'll give you a sprinkle of UBI at the end, that seems pretty dystopian to me. So that's the future we're trying to prevent. I don't think the disruption can be stopped at this point. It's coming and the only question is like, who's going to own the future?

I think it's a lot more questions than that actually.

I'm gonna have to hop too, but I think you need to go. Can I ask you one last question? It literally is a 30-second thing, by the way.

Yeah, yeah.

One of my employees, he was digging up some of your old videos. He saw that you had a 405-pound bench PR. Is that actually true? Yeah, yeah, yeah. Because he was very impressed by that.

Yeah, yeah. I mean, I played college ball, college football. I have a friend who's a doctor in the NFL, played with him. Great friend. And yeah, I mean, I just kept lifting after football. And, you know, to me, it's just been like, I was a mathematician and I was a powerlifter. And to me, it's just like engineering signals in order to have neural adaptation. And it's all one and the same. And so, yeah, I mean, you know, the Beth Jesus character, it's all about mind and body. And, you know, I do like to cultivate strength and, you know, push myself to the limits to some extent. And yeah, that is a— I'll just post the video if you want.

I'll tweet it. Well, he found it. I have no idea how he found it on Vimeo, but I'm impressed because a lot of this transhumanism business is very agnostic and kind of denying the physical, but you're actually embracing it. Anyway, that's also another conversation. Thanks for making time. I'm going to have to hop off. I assume everything will upload when we all hop off. I'm not sure what— Erik has never disappeared before, so I literally have no idea what's going to happen. All right, to be continued, guys. And Guillaume, I— yeah, definitely to be continued. And Guillaume, one of these days I want to talk crypto and AI with you when you're over the whole hump of fundraising and all that stuff. Because I've been thinking about this whole thing for a long time and I'm very— yeah, I'm impressed. Okay, cool. Awesome, man. All right, see you guys.

Cheers.

Upstream with Erik Thornberg is a show from Turpentine, the podcast network behind Moment of Zen and Cognitive Revolution. If you like the episode, please leave a review in the Apple Store. Turpentine is a network of podcasts, newsletters, and more covering tech, business, and culture, all from the perspective of industry insiders and experts. We're the network behind the show you're listening to right now. At Turpentine, we're building the first media outlet for tech people by tech people. We have a slate of hit shows across a range of topics and industries, from AI with Cognitive Revolution Econ 102 with Noah Smith. Our other shows drive the conversation in tech with the most interesting thinkers, founders, and investors like Moment of Zen and my show Upstream. We're looking for industry-leading hosts and shows along with sponsors. If you think that might be you or your company, email me at erik@turpentine.co. That's erik@turpentine.co.