The billionaire paying AI researchers to stay out of Big Tech

This episode is brought to you by Accenture. When your advertising operations fall out of sync, everything else follows. Spotify and Accenture are working together to reinvent the rhythm of ad sales. Using automation, analytics, and smarter workflows to simplify campaign delivery and access better data across the business. The result? Less time spent on operations, more time connecting brands with the moments and fandoms that matter most. Learn more at Accenture.com slash Spotify. This episode is brought to you by Google Chrome. You think you know a browser, but Gemini and Chrome? That's new. It can help you with practically anything on the web, like restoring a vintage motorcycle from a 50-page restoration block, or finally break down that long article you've had open for weeks. Gemini and Chrome is here for it. Ready to make anything online make sense? There's no place like Chrome. Check responses, setup required, compatibility and availability varies 18 plus. There's upsides and there's potential downsides. And that's true about technology in general. It's really true about AI. And I think we can cure cancer. Perhaps most types of cancers completely by the end of my life. What happens when every top AI researcher gets an offer they can't refuse? A few companies end up owning the future of the tech industry. Andy Konwinski, co-founder of Databricks and Perplexity, is trying to stop that. His new effort, LOD, is putting over $100 million behind AI researchers to keep their work in the open, instead of disappearing. Big tech. Actually, LOD sits right there and kind of paving the path to help students know how to capture the world's attention. It's that much more important in the time of AI, because AI researchers have left society behind entirely. This week on the show, we discuss how the top researchers actually feel about where AI is headed, Anthropic's recent controversy with its new model, Fable, and why Andy thinks big tech should fund the open alternative. I'm Ellis Hamburger, founder of Meaning, a storytelling studio for tech companies. And I'm Alex Heath, founder. And I'm Alex Heath, founder of Sources.News, your home for scoops about the AI race. Welcome to Access. Andy, great to be with you here in an undisclosed location on the east side of LA that could only be described for people watching on video as Hogwarts. I don't know, although there's a bust of Ben Franklin behind you, so maybe Hogwarts isn't fair. Maybe Ben Franklin was an alum of Hogwarts. I think for anyone in tech who thinks they are a renaissance man, this is where they would aspire. This is where they would aspire to do their best thinking and writing in Obsidian. So we have Ben Franklin behind us. We've got some beautiful artwork. It is genuinely inspiring. And your beard is also inspiring. Oh, thank you. This is apparently like nearly every issue of science that was ever published in bound form. Behind us. Yeah, behind us. Wouldn't Claude love to be here too? It'll go into Alex's Claude after this. Yes, it's already here in spirit. So Andy, we want to talk about Laud and the AI world and your view of it because it's really interesting and unique. But I want to start by actually positioning to you what I would describe Laud as, which you're trying to keep the future of AI and the concentration of power that's happening in AI out of big tech's hands, or at least diffuse it more. Is that fair? Is that the mission? That's not the mission of Laud precisely. Precisely, but that is certainly one thing that I care a lot about right now and always is diffusion of power. I also care a lot about diffusion of innovation. So I would say the root of Laud for me goes back to being a PhD student and then observing a type of research, which, by the way, research as I've internalized it in my life and I hang out with researchers, I did research, I still do research, is discovering and disseminating knowledge. So being here, sitting amongst these libraries and wizard-like settings. Yeah. Which is how I think of academia is that, it is that effort. And my experience as a younger person coming into that, you think of it like the church of academia, like pursuit of knowledge and discovery of light for our species, there's a corner of that effort that is all about translating the discoveries into impact for society. And of course, what happens in academia is the breakthroughs happen in the open. That's like the, we have the secret sauce that millennia old tradition, going back to the monastic years of how do we make breakthroughs and publish them? Like you, how do you cure diseases? How do you discover things about electricity, refrigeration, like the ideas that get distributed out to the world, move humanity forward and that captivated me and I loved being part of that. And in particular, the part where we translate the ideas into impact. And so a lot is, is an organization. It's an organization that aims to help researchers have impact out of their research. Now what's happened in the AI paradigm, which aligned with my creation, creating LOD and LOD kind of catching stride in a very meaningful way now, as we come up on our year, celebrating our anniversary is we've seen a consolidation of power in the world around this technology paradigm. I'm a student and a fan of technology paradigms from history, going back to railroads and electricity and refrigeration and, you know, more recently cloud computing and. Mobile and the personal computer, you know, like the internet, these are all things that moved humanity forward, but throughout all of those academia and more generally open research was able to produce breakthroughs and ideas that were disseminated out. And also it was able to do it because of dissemination, a scientist talking in their living rooms and in libraries and in classrooms and at conferences, this paradigm, the most important one of all of them, the one that's accelerated to grow. 10 or a hundred times faster than any other one is different in that for the first time, the power dynamics and that the shift has occurred from how much of the breakthroughs might be able to happen in the open versus in behind closed doors. The frontier research happening in AI right now is at closed labs. And that's, that's a problem for democracy. It's a problem for my daughter's futures. And so a big part of what I'm kind of championing with LOD is. Is impact out of research still that benefits and moves humanity forward. And I believe for that to continue happening and for human humanity to prosper for all of humanity to benefit from AI open research needs to continue to thrive. I know a lot of people point to the military innovations, you know, that begot cell networks and, and so many other things, or even for, uh, for, you know, going to space and whatnot is how, how do you frame that within the narrative you just shared? Is it that once those things were kind of out or things were over? Did they made their way into more or less the public domain? Yeah, there's always a balance between secrets and discoveries being disseminated, uh, industry, which as you know, I've co-founded successful companies. I'm a capitalist at heart industry is, uh, the exercise of taking knowledge and creating advantage and, and strategic secrets, intellectual property, trade secrets that let you translate the, the breakthroughs happening in open science to. Product and solving people's problems that is just as important as the knowledge discovery. And, um, that's true about the public good for through government as well. So the national defense sector, all of those things, uh, there are many, uh, government initiatives. If you go back to the Manhattan project, countless examples where you need secrets, uh, to build on top of the discoveries that happened in the open. But the problem. The problem now is that all the secrets are behind the closed labs, right? Yeah. And that the, and that the rate of discovery will actually long-term diminish if we extinguish the tradition of open science, the academic institution itself, if that, if that implodes, then who's going to train the next generation of research scientists. If you were to just buy, you know, you were mentioning capitalism, the, the drive of capitalism that these research researchers can now get plucked out of academia and. make millions of dollars a year at anthropic or open AI. Wouldn't that also mean that the research will continue and there will be new ideas and new breakthroughs that will just happen inside open AI or anthropic and not in academia. Like, isn't that still going to happen? It could, um, there are breakthroughs still happening, but the type of research happening in the closed labs is very different than the type of research that traditionally could happen in the open. Open research, uh, is able to take advantage of this diffusion idea where scientists talk to each other, collective intelligence builds on each other. It's why we have so many conferences and peer review, and the whole mechanism itself is about collective intelligence. This Gestalt principle or phenomenon, the sum is greater than the parts with discovery and invention, um, that can happen behind closed doors. But one thing I know for sure is that for the history of this country, anyway, there has been a balance between things that let's say DARPA and NSF and, um, even industry arms that do research and publish it. transform architecture was published in the first place by Google. And that was during a time where they were at containing a tradition where industry published and that has shifted now. The amount of publication that these labs do has gone way down, way down, uh, in some cases, much closer to zero, depending on the lab that you're talking about. And, uh, that, that, that shift could lead to the end of academia. And the academia is consists of all this institutional knowledge is memory. Like we have advisors and advisees. When you go get your PhD, some of the most profound inventions, let's say what John Shulman did with deep reinforcement learning, one of many Berkeley and Stanford and MIT and Carnegie Mellon PhDs that have gone on to change the world and change industry with their inventions. What these folks did is they trained for five, six years at the right hand of another luminary researcher in this advisor-advisee relationship. It's one of dozens of institutional mechanisms that academia has to train up the next generation. In fact, the primary goal of academia, even though it is discovering knowledge and disseminating it, is teaching the next generation, like self-propagate the machinery to discover knowledge and move humanity forward through it. These labs will have to recreate all of that. They'll have to have their own conferences. They'll have to have their own advisor-advisee. And the capitalist objective function, while I. Believe in it and I benefit from it and society benefits from it, isn't a great one for reproducing the successes of research culture. I think there's also a tendency to try and get closer and closer to the applied once you go internal, maybe, or correct me if I'm wrong, but I'm curious about the motivation because I don't think I ever understood professors and PhDs and academics to be primarily motivated by money. Is it just that the money is so crazy or is it that they're being told, hey, do what you do internally. And then it just doesn't quite work out that way. You mean the motivation that would get them to leave academia? There's two things. One, when we built Databricks in 2013, the salary difference between staying in academia and going to industry was maybe two to three X. Now it's 10 to a thousand X. So you're some PhD students, even first, second, third year PhDs are turning down two million, five million, ten million dollars a year in equity for that. That. Now, if you look at the trajectory of this equity, you can imagine that's more than ten million dollars a year. Fast forward six months. I talked to an anthropic researcher a few months ago who had just left. No, he's not public. There's people don't know who he is. He's making 30, 40 million a month in equity gains. Yeah. So so the thousand is not an exaggeration. There there are people with that many, you know, that that many zeros, if you multiply what these PhD students will get if they stay in academia, PhDs at the top programs right now get paid less than one hundred thousand dollars a year to be. A PhD, it's like an internship basically to learn the discipline, not to make money. You get to hang out in these kinds of rooms, though, right? It's the primary perk. It is one of the best perks and the association of, you know, inventing an impact. That's really what motivates this type of person. Whereas when you go into tech, you have to live in one of those places you'd see on Reddit slash male living spaces. That's like a mattress on a floor or something like that somewhere. That is a badge of pride. But yeah, that's. That's different than the the gilded halls of this building. But it's not just the the compensation difference, actually, that there are many researchers who would who would say no to that. It's access to resources to do frontier research. Yeah, that's it's those two things combined. One or the other people. I wouldn't be nearly as worried about existential problems for academia or just open science and open research. But because of the two things you need one or two orders of magnitude more dollars and GPUs. Then you used to to stay doing research at the frontier right now. Do we have speakeasies now of these researchers who have to go underground to talk about how cool their shit is now that they're I haven't heard of a speakeasy, but informally, I think they want to talk about this stuff, right? And I do talk about, you know, I mentioned diffusion of innovation, this idea that scientists getting together and talking collective intelligence, the ideas build on each other. That's how citations work and related work sections of your paper. The idea is building. Each other. I've witnessed that myself with like Mesos, we built it at Databricks are at Berkeley and then we built spark and it kind of was born out of missiles and messes itself was born out of Hadoop and there was billion dollar industry around Hadoop and then eventually billion dollars. You could be saying Lord of the Rings characters, but you're talking about technology. Sorry. Those are big, important projects from a past technology paradigm from a distant age of 2013 and I list the names of the projects because they they built on each other. And that, that concept of diffusion of ideas is so essential to the science and yes, what is happening in industry is scientists go spend time at one top lab and they a year later go spend time at another one of the top labs and then six months later they go spend time at the third of the top labs and they just do this rotation and bounce around collect equity like infinity stones, the equity goes up a little bit each time and they get to stay up to speed on what's going on. It's happening. That is a form of diffusion of innovation actually. It's partly why the labs can compete with each other, but it's happening at such a tiny scale compared to what we've done in open research. The culture that we have here at universities, what does this look like in terms of trying to address it? You guys have a couple of programs at law where you try to fund and give grants to researchers to basically say, don't go to meta, don't go to open AI, whatever, keep, keep doing your thing in the open. We'll give you money. So a big part of LOD is absolutely grant writing. LOD is at its heart of what we do operationally is we give the right resource to researchers at the right moment to help them maximize their impact and what researchers who want to have massive impact do is they have the idea, they prove it out with some research experimentation, and then they want to capture the world's idea, the world's imagination with that idea. So they want to ship that research as a tweet or it could be a paper that they post on archive or a blog post. Ideally, they want to go jump over to industry and monetize it and actually not just like it's not the monetization that drives them primarily, even though that's nice. It's the getting their idea into people's hands and their strollers and their cars and their bloodstreams. Right. That's like what these people love, because that's how you measure impact strollers, cars and bloodstreams. So my favorite products to invest in use cases right there. Speaking from experience, I imagine a quality stroller or YouTube, I think. Right. Oh, yeah. I would keep investing in innovation for stroller technology. Yeah, I think we need more innovation on the wheels. I've got one of those babies and yo-yos and when I walk it down the street, it's like just one wheel like that. We got AGI, but we can't do stroller wheels. It's just too difficult. It's the hardest problem of all. Yeah. I need to ask Claude to start a company to improve strollers for us. So this what they're doing is hopping around for the definition, but it's not happening at the same rate. Yeah. Give the right resources to the right researchers just in time. That could be when they're at their startup already. So we have a venture arm and it could be while they're still doing research. So we have Lott Institute, which I put $100 million into to do grant writing. And we at the Institute, we invest in labs, projects and people. And those are three different grant sizes. So for labs, we have a program called the Moonshot Program. That's $10 million for three to five professors at a top university. With let's say 20 to 50 PhD students for three years, it's a big effort focused on societal problems, AI's impact on healthcare, scientific discovery. And then smaller, next smaller is projects. So individual researchers, PhD students, usually who have a cool idea, they want to ship it to the world. We plug in what we call those slingshots and we plug in to give them either API credits to do inference for their research. Or an engineer or communication support. So how are you going to do your amplification strategy when you launch your project? What's your readme look like right now? Or just straight cash or GPUs. So that's how we help those individuals. We have a residency program where top PhD students at Berkeley and Stanford and MIT spend time sort of embedded outlawed and get even more access to GPUs and this infrastructure that I just talked about, like communications. And the communications is one of the bigger parts of it. I actually would love to spend a little bit of time talking about that in the context of podcasts and talking to reporters and journalists and what that I believe AI needs a whole lot more communications than we've had. You're speaking to the choir here, man. I think so. Yeah. I think that's partly why we're speaking because I'm so passionate about it and I know you are. Obviously, money is where your mouth is. You've chosen those. Every startup tweet in your Twitter feed, I wrote. That's Ellis' claim to fame. I'm just joking. Thank God. It is important though. There's so much work happening and it's like, how do you stand out? Yeah. Yeah. And just the basics of having researchers speak to the world is not obvious. It's not something that over the millennia in academia we've optimized for. We discover the knowledge, but the road from the path from research to impact or research to startup is actually not that well trod. That's actually a lot sits at right there and kind of paving the path to help students know how to capture the world's attention. It's that much more important in the time of AI because AI researchers have left society behind entirely and that's like- You mean in terms of just the wavelength they're on left behind? Yeah. Yeah. Out of touch with the problems that people perceive when people have data centers built in their backyards. Yeah. How do they feel about that? My cousins in Illinois who are police officers, how do they feel about AI? I don't think most of my friends who work at Anthropic or OpenAI or startups, they're like, "Oh, I don't know how this works." They're like, "Oh, we don't know how this works." a lot of technology especially software over the last you know decade or so is that people love to build for themselves and what we like to build for ourselves is new to-do apps and notes apps and tools to make crms more efficient and to fix all of our internal tooling or to make a quick buck and vertical sass or something like that and you know you i feel like you've spoken a bit about public goods uh you don't see a whole lot of companies at least the ones getting the headlines um in terms of the quantity at least who seem focused on that type of stuff and i mean i wonder if if some of the researchers can you say specifically what you mean by that like what type of stuff um i would say software for other white collar people doing image generation or some other white collar work as opposed to let's say you mentioned health or something like that it's just not quite as glamorous reskilling or yeah trying to manage reskilling is an interesting one where you actually do see researchers focusing on working with um blue collar folks on how to use frontier ai for taking pictures of plumbing fixtures and you know there's sort of the embodied helping repair people i think one of the moonshot seed grant recipients is a stanford team that includes eric brinjolfsen and d yang and they're doing reskilling and they've run uh workshops that include um people responsible for uh things all up and down the stack so the reskilling focus in particular is more plugged into the sort of full spectrum of potential job displacement long term yeah and i think uh i would probably argue that even when there are uh good hearts trying to build very valuable things they still kind of suffer from that intellectual issue as many democrats do of just getting very much into the weeds and over explaining as opposed to having some high level message i mean you see the success of someone like obama's message as opposed to what we've seen since with you know a whole outline bullet list as opposed to just one word it really is that simple you know when it comes down to communication a lot of times and uh yeah i'm just not sure how many big famous loud voices we have out there who are who are trying to make that stuff more palatable you know and i do think we need more of that yeah i do think some of the problems maybe the biggest one in this sphere that i care about around the consolidation of power and the crisis that we find with ai leaving a lot of society behind is about an absence of leadership it's a vacuum of what i think of it like you mentioned obama like somebody who speaks and the message resonates yeah and it's uplifting yeah and it's inspiring because they're speaking the pope's coming on next episode so we'll ask him i mean because they are speaking but they're dario's saying half the jobs are going to go away that's the problem it's like a lot of technical leadership and the incentives are a little bit wrong for are actually completely wrong for ceos and vcs to speak to the world and bring the world along with them and kind of speak truth the incentives are to paint a rosy picture and a picture of of prosperity and panacea when claude uh gets to agi and um although i guess the the narrative was is more altered in the last week uh that dario's been talking about but um you know the incentive is one to tend to want to talk about the positive outcome of this technology and that's not the perception of people in my kindergartners right class is and the teachers at my daughter's preschool so this episode is brought to you by accenture when your advertising operations fall out of sync everything else follows spotify and accenture are working together to reinvent the rhythm of ad sales using automation analytics and smarter workflows to simplify campaign delivery and access better data across the business the result less time spent on operations more time connecting brands with the moments and fandoms that matter most learn more at accenture.com spotify this episode is brought to you by google chrome you think you know a browser but gemini and chrome that's new it can help you with practically anything on the web like restoring a vintage motorcycle from a 50-page restoration block or finally break down that long article you've had open for weeks gemini and chrome is here for it ready to make anything online make sense there's no place like chrome check responses set up required compatibility and availability varies 18 plus yeah i mean a lot of the people i talked to um maybe not at y'all's elite level um i'm just not sure very many people are able to describe what the better future looks like for the most of us and i'm curious you know what what you think the the best everyday version of that is uh that you've heard is it about is it about quality of life is it about health is it about the reduction of pain is it about uh prosperity more entrepreneurship anything in particular well um i'm really worried about civic discourse and uh the polarization of it in our society and one of the topics in the moonshot grant program was to have researchers use ai to help people who disagree come to a common ground and to identify their uh their their things that they have in common and you get people on two sides of gun control who in left to the internet's devices might just flame at each other uh you can use ai which maybe put ai sycophantic uh side to to good use and have an ai that might care a lot about the future of the world and i think that's a lot about brokering common ground so there's lots of studies out there by done by academics that have shown like with with kind of rigor that you can use ai to moderate conversations and have people agree more uh who would otherwise disagree well i hope that it you know play it out if we do it right uh where you end up with less polarization and less uh tumultuous times in our democracy even existential times for our democracy well and to elsa's question about the positive view i know i mean i talked to some researchers not as many as you and i i know that some of them have positive views but to be honest the thing i've found you know in the private conversations is a lot of them actually have a fairly dystopian view of what could happen and they understand the gravity of all of this and they're preparing for it people read the headlines about you know sam altman's bunker and all these things right um but the researchers themselves i mean they talk about automating things and they're preparing for it and they're preparing for it and they're preparing for it their jobs first right this this is the whole concept of recursive self-improvement and you know take off and all these things is like and the labs are starting to talk about it very publicly even in recent weeks right you had altman have that note about rsi and all of that and delaying the ipo because of how weird that could be if that gets unleashed so i i think there may be an element of this and i mean correct me if i'm wrong but maybe the researchers sometimes shouldn't be saying what they feel because it's scary to people to show it should not be saying to hear how people who are building this stuff at the frontier actually feel about world where the world is headed it can be for me it's i've had some moments where i'm like oh wow like you really believe all this and maybe the maybe the other side of this is the world needs to hear this the world needs to hear how these people view this yeah because they're building it i think the world needs to hear what i think researchers need to think about it more than we are yeah i think we need to talk about it and i think we need to talk to the world about it uh i i i feel strongly about that and uh if we're worried let's talk about how we're going to be able to do that and i think if we're worried let's talk about how we're worried if if people think we shouldn't be built it they should shouldn't be building it they should say that um i personally think that the scientists i know all nearly all of them find identify as very being very pragmatic so there's upsides and there's potential downsides and that's true about technology in general it's really true about ai we can mess everything up um i do worry about my daughter's future for sure keeps me up at night and i think we can cure cancer i think we can cure cancer many types of cancers perhaps most types of cancers completely by the end of my life uh i think that's what's on the table i'm not saying for sure i'm not a lot has to go right um but i want to be able to nudge it and so do most scientists most researchers most people wish they could have a little bit of nudge uh for the good for their kids right so um that's my sense is that they're pragmatic they get that there's downsides they believe there's massive upsides and i felt some of the upsides my parents have had cancer and i used clod actually to do deep intensive research about my parents cancer i learned about my mom had leukemia and i learned about uh blasts and the specific uh drug trials for multiple specific drugs i could never have become an expert to the level i was anywhere it's it's science fiction what i was able to do in two weeks when she was in the hospital after her pre-leukemia advanced leukemia it was a profound experience for me to be to be unlocked on that knowledge to have that knowledge be made accessible to me so i do personally have even like the felt experience of the some of the positive impacts even in that way even before we cure cancer empowering families to ask the right questions to the doctors is one thing that i felt personally yeah uh my dad who's probably listening right now had to uh a very long time ago comb through the early internet for hours and hours and hours god knows how many hours on our very on a dial-up motor with cancer yeah i think it probably was on dial-up and i can't imagine how scary that must be to feel like oh i'm I'm actually the one of the whole care team who's like spending the most time doing the research or something like that. That's how it is, for sure. And obviously, I mean, he's a lawyer, but he's not a researcher out there reading papers and who knows how many of them were even published at the time. Right, yeah. And so I'm a big believer in the health information angle, for sure. Yeah, there's another one of the Moonshot projects is digital embryo twin. So can we use AI to help make a full resolution twin of an actual specific embryo? And then you can, with that full resolution twin of it in the computer, you can simulate all sorts of things and understand genetic defects and prevent genetic defects early enough. So there's some very real statistics about the impact that being able to identify genetic defects early enough would have. How much do you feel like you're talking to people who are living, obviously, in the future, but just have this very grounded sense of where the world is about to turn that the rest of us don't have? And how do you bridge that in your mind? Because you're living in that world, you're talking to these people, and then you're also talking to people like us, and you've got your family and all this. So how do you deal with that? And how big is that gap? Is it as big as I think some perceive it to be? I feel like sometimes I feel like it's a big gap when I talk to these people and how convicted they are about AGI and all these things that are going to happen. And it feels like most people just have no idea. Yeah, I do think even within the technical community and the deep research community, there's different times at which people have their like, oh, crap moment. Like, for me, some of the stuff are my mom's, the research of my mom's cancer. And before that, coding agents is one where a lot of people in the last six months have had that moment where they're like, wow, something's different about this. Maybe we are riding that exponential, which is to say that the progress keeps happening. And now one very large. Important segment of our industry, which is software generation, has been fundamentally disrupted. And that plays it all the way back to universities where, you know, computer science enrollment has shifted in the last three, five years, you know, like flattened after 20, 30 years of being like the most steep increase in training. Now it's flat. So it's a real societal thing, percolated all the way through society. And people understand that when their kids drop out of the CS program. So it's touching their lives already in that way. Um, now go all the way up to the frontier. I think each of us has this moment or has a lot of researchers having this moment where it starts to feel like you're living in sci-fi. That's how I would say it feels. Uh, and I'm a big fan of sci-fi, you know, star Trek, star Wars, cyberpunk genre. And I'm just so used to reading about things that feel like science fiction. I would say living through when I was a kid and the internet came out or mobile, these were like, wow, my, my PhD advisor used to talk about how cell phones are like tricorders, like this sort of. Early thing that was in star Trek that was cell phone like, actually, it was like, look, we did it. We, you know, like a clamshell, uh, so, but now it's, it's bigger ones. Like, could we cure cancer? And, you know, you know, things like the matrix, people talk about getting AGI pill. That's a direct reference to the movie, the matrix. And I wonder how many people even know that that's the origin. Um, so even there at the frontier people, it's a complex emotional experience to, to know how profound it is and it does, I think in many folks stir something deep that, and for me, what it stirs is needing to raise my hand and stand up and point out that there's this absence of leadership and that we, like the world needs to be brought along. Society is transforming, is going to transform. And that brings me kind of very quickly back to this problem with consolidation of power and the threats to democracy. These are like, basically I have this long list of species level things going on. And, uh, I think the best minds in our species need to be working on the species level considerations right now. That includes economists and lawyers and doctors and policy makers and, uh, union leaders like you name it, religious leaders, the Pope should be weighing in on this regularly. So, uh, I think every part of society needs to be spending brain power on this. Yeah. You, uh, are, uh, an impressive student of all this stuff and I'm curious if you've thought about how, uh, literate. The leaders at any given time were in the technological revolution. Is it naive to say that it's like, oh yeah, with our, you know, uh, octogenarian leadership now, it is uniquely a moment where they are especially out of touch with what's going on. Or has it always been that way? I think the rate of progress has never been anything like this. And that's hard to get your brain around the quarter when you're talking about a century and a decade, something like that. That's real. Um, and so we've had balance. I talked about the balance between. Academia and the balance between pay rates. There's been balance between different sections of society and what's happened in the last five years and especially one year is capitalism and for-profit entities have just gone on a breakaway, just sprinted far away from at the speed at which they're moving, the progress that they're making, uh, including companies that I'm part of and that, you know, like that is what industry was built to do, to, to make progress, to, to invent and just sprint. Uh, it's good. I like to call it the oldest company there is. Indeed. It's like, can you imagine? And probably the most important one. It's like our startups are messy enough. Yeah. So I think the problem is more that this paradigm is so different than past ones and what worked in the past in terms of policy, being able to catch up over decades, isn't going to work where policy has to catch up over a decade. There's more of a buffer. There's more time. There's more time. There's less time. There's less time for months or weeks. So innovation has to happen at the intersection of the public good industry and academia. And you're seeing this coming to a head even in just the last week, like the week before we're recording this with Anthropic releasing Fable, nerfing it for research capabilities where you couldn't use it to help, you know, do AI research or train another model. Right. Silently. Silently. Did you pull an all-nighter to use Fable, by the way? No, I didn't. I should have though. I should have. I should have. I should have. I should have. I should have. I should have. Yeah. Yeah. Yeah. affiliated with law. And that's just wild. It's a wild situation to be in where the consolidation of power we've talked about has gotten to such a point now where these people in government who, to be honest, yeah, they don't fully understand it, but they're reading the headlines, they're getting told how capable it is, and they're putting the brakes on it in a way that is not ideal for anyone, right? And I'm curious when you saw that, just these events of the last even few days, your reaction and where you think this is headed, because it feels like it's going to get worse maybe before it gets better. Yeah. I think it's clear now that the consolidation of power is a risk, not a feature. It is not something that's good for us. And we're seeing that play out in real time. There's a difference between safety and centralized control. That's just, we need to take that as a given. The threat to democracy of centralized control is real. And so the debate between open versus closed, that's sort of actually ensued amongst researchers. What happened, if you are following along on the Twittersphere, which I'm sure we all are, is when Anthropic released Fable 5, they put these limitations on one for cyber and biology that if you use it, if you try to ask questions around those things, it'll bump you back to Opus and it'll tell you. But two, if you try to do AI research on it, things that they perceive to be maybe distillation or things related to frontier AI research and explicitly because they were worried about other people making progress as fast as them, then they would also downgrade you, but not tell you that. And so it was this- The shadow ban. The silent manipulation of the alignment, the like, what your experience as a user was no longer what you thought you were having. And it was buried in their terms and conditions, like deep in the model card, not in the blog post. That's what the scientific community reacted so strongly to, was like this secondary thing of like, one, it's clear that there are a lot of incentives for Anthropic. Dario talks about safety being the biggest, but they're the world's most successful company. Like the $9 billion to $47 billion in revenue. In nine months, that's real. And so there are clear conflicts of interest. If you're building the thing that's become the most successful product company that ever existed and saying only we get to have access. And just a week before, one of the co-founders of Anthropic published the blog post, this blog post about recursive self-intelligence, how Claude is using Claude to build Claude. And there was some graph of like all these, this graph that was full of Claude's, like Claude's calling Claude's calling Claude's. And it was like, oh, I don't know. I don't know what's going on. I don't know what's going on with Claude's. And literally within the span of a week, you're talking about how Anthropic is using it to accelerate all their progress and then saying, we're worried about this acceleration. So nobody else gets to do it. And it's not actually that this debate, I think, that I think is so worrisome, or that they chose to make this decision about policy and roll it back. That's whatever. I have an opinion about it. But the more interesting thing is a completely different narrative. It's that they felt the prerogative to make that decision in the first place. Intelligence is clear, especially at these levels with biological concerns, crypto, the cure to cancer. It's a foundational resource. These breakthroughs are different than past breakthroughs, where, sure, it's happening in closed proprietary labs, but it's going to change everybody's lives. It's going to change our daughters' lives, and our sons' lives, and our parents' lives. Dogs' lives. Dogs' lives, sorry, yeah. Everybody's life. And the question is, who controls it? And the thing that's common between open science and open research and democracy, is participation, and that legitimacy comes from broad participation. So that's true in meritocracy that you see at universities and beyond universities for open research. The ideas are vetted, peer review, dissemination, and all these different ways. It's the secret to the innovation, and it's the secret to keeping society benefiting and moving all of humanity forward. It's true for democracy, and that's self-apparent in the definition of democracy, that you have representation. You have power. Sovereignty, that's like the beauty of democracy, and why we've spent so much energy in democracy defending against centralized power, and why we left centralized power in the first place to build this nation. So when it comes, that's the thing that changed for me, and that was so profound in the last week, was this new thing entered the zeitgeist of a company and a person, Dario, exercising control over what I believe is a profound and fundamental foundation. It's a fundamental foundational piece of technology that every human has a stake in. So we need to figure that out. We have to avoid centralization of power and have a solution to safety. The answer is not to throw the doors open. It's not like, yes, publish everything. Clearly, biological threats are real. Cyber threats are real. But the answer is also not let somebody who has really good intentions hold the key, like the one key, or even three people hold the key. There's got to be a thing that addresses control. Well, because the key approach that we're on right now, let's be honest, it's going to lead to it could lead to riots. It could lead to, I mean, the wealth disparity on steroids, the power disparity, disparity that we're talking about. I mean, the end state of this is fairly dystopian, even if everyone are anthropic and they're a public benefit corporation and all of this, they decide to lay it down at some point. I don't know. I feel like the capitalism of all of it is too involved in these companies now. For the truism of how they set out to go actually being realized. I mean, I think maybe the cat's out of the bag. Well, I mean, they set out with a capitalist intention. So in spite of there being a mission, like non-profits, actual non-profits, you make a choice when you create a company, whether you can be a for-profit or a non-profit. And non-profits disclaim from the beginning, they say, how I define fiduciary responsibility is fundamentally different than something that is for-profit. And I think that just, that is why universities, which are non-profits by definition or by requirement, work, function the way they do. And it is a critical thing that I don't, and by the way, I'm a capitalist. I build companies as well. And I'm, I believe in the power of that, but I do agree that you can't disentangle the incentives. And so the answer is, if we weren't to do something different, if we weren't to innovate, we end up with is essentially a few organizations, maybe two or three having all the, all the power and, you know, potentially kind of in this profound way, given the AI is shaping up to, to have all this power. I call that feudalism with better branding. Yeah. There's so much of it that just seems kind of like these days, like, well, if I don't do it, others will. Do you think there is an argument to make at that level or just trying to use the machinations? Of capitalism to try and better fund alternatives? I actually believe that it's not even just the mechanisms of capitalism to fund alternatives. There are other funding mechanisms. The U S government has deep pockets. It's one of the, it is, it has actually been the primary source of innovative innovation funding, NSF DARPA being the two biggest. And it's a no brainer for the, for both the U S government and democracies and governments across the world to spend more money than they have historically on any other type of initiative. Uh, you know, they should be 10, a hundred X-ing they're spending on how to make sure that AI frontier AI work is embedded in the government as a government concern, um, that does require innovation. Like I said, they need to operate with, uh, more experts. So there needs to be computer science researchers who have roles within and adjacent to government at, you know, far more depth. Yeah. And volume than we have today. We have some, but they're very rare in my community. Right. The parallel would be like right now they're being two or three companies building the internet. Right. Whereas like internet is the perfect parallel if that were to be happening right now, it's really good. Alex. I liked it. We'd be having the similar conversation about concentration of power. Yeah. And it was a problem in the early internet that a few scientists cared a lot about and raised their voices about. So vent surf who actually will, uh, participate. Um, a lot of it coming up as for this exact reason, uh, and invented the TCP IP protocol that was foundational to the internet at the lower layers. I played with some TCIP in my life. Yeah. Does that already say it? TCIP, TCP IP, TCP IP transmission control. You tweaked your MySpace a couple of times. Well, you've used it probably like 10,000 times today. At least that's so many packets have been flying off that iPad, uh, via TCP IP, um, that, that early chapter of the internet where certain scientists had to stand up and say this. It should be open. You know, people don't realize the internet was offered for sale to AT&T in the early days and AT&T said, no, it was like, I think in the seventies, uh, maybe 74 and the internet here's the internet at the time. $10. Uh, I don't remember how much it was for, but AT&T AT&T said no, because they had a competing protocol that they owned and they could have bought this other one, but they already had this big investment on that. They wanted to win. So why would you want the two? Cause they thought they could win. Um, and there were other competitors to the internet multi, you know, like a half a dozen that had viable shots. At, at making it a closed thing, that early stance of several researchers, including Tim Berners-Lee, who invented the worldwide web protocol that opened up the internet for people to be able to, you know, have chat rooms and have web pages in the first place. It blew the doors off the internet in the, in a way he was another staunch advocate of open. And that's why the protocols in the first place were published and, and the, you know, decided on by many key people across industry and academia and that foundation that they built and maintained in the open, not because. It was the right thing to do, but because, uh, not because everybody saw it being the right thing to do, but because a few scientists believed in this, uh, led to a subsequent paradigm, like a generation of Florida, like wild west innovation, tons of, and what we've ended up with today. Just like so many, everything is basically stands on the internet and stood on the shoulders of the internet, uh, since then cloud computing data. Um, I mean, just look at the way the LLMs are trained, the pre-training phase still. Most of them, the majority is scrapes of the internet. So like the internet itself is the, was this one of the primary secrets to the breakthrough for AI, um, that it's kind of, it's like immeasurable, the productivity output of keeping that early base open and that infrastructure. So what I'm championing, like, I'm not just hopeless that like, we don't just need to accept feudalism with good branding. We should do something about it. But the, uh, what I believe is you should provide a great alternative. So a heavyweight contender in the open that's in the ring with the big labs and brings the world's best minds to bear on the problem across institutions. It's, uh, we need an open AI infrastructure commons, and we already have academia, which is a technology commons where a lot of innovation happens, but it needs to be tighter than that. And it needs to be, uh, at the intersection of academia industry and the public good. So the us government and philanthropy, and it needs to have the best minds who doing open science, having enough resources. That's GPU billions of dollars and access to state-of-the-art models that includes proprietary models, if possible, uh, in order to make sure that you've got the right balance of open researchers and diversity of perspectives with control and influence over the protocols and experts, not only policy makers, but you want policy makers in there too. And, uh, and, and also, um, having access to actually continue to innovate. So like make new breakthroughs that they publish. So that's what, what I believe needs to happen. And that will address, would address much of the worry, my worry. Yeah. I mean, uh, fun facts, little known story that most people probably don't know about is that, uh, Sam and Greg, when they were initially doing funding for open AI, the fundraising, they approached the us government and asked for money and the government said no. And it's kind of wild that like what could have been, right. Uh, and, uh, yeah, I don't know if I have, I keep saying like cat out of the bag, but I, I don't know if, uh, all. Yeah. All this can be done at this point. I mean, if you go back to where the internet was in those early days, the amount of money, the amount of power and scale. was so much smaller compared to where ai is now right and so has has the train kind of already left the station i guess the seneca and me uh i could i could hear why you might think that i stand at a position where i know dozens and dozens of the smartest people in computer science research uh because that's what i've built laud to do is go meet them and give them the resources they need to be to be successful in the open and um i believe that if if we build it they will come uh that it's it is those things we talked about resources you don't need both you don't need to pay them better than the big labs irrational pay salaries and uh and give them access to the frontier you can pay them less and give them even they actually don't even need to be playing that the the race with scaling laws of the big labs is you need these billions and billions of dollars of compute these massive data centers you can thrive in scarcity i want an open contender in the ring with anthropic and open ai and and google uh but i don't think it needs to fight with the same way like where they fight in rows we can fight in ditches uh you know like scarcity can breed innovation so there should be some pre-training we don't we can't do it with a few million dollars we need billions of dollars and for for this open commons like the internet uh but it's still for sure in the race i think that window's closing though really yeah um every every month another phd student like realizes the situation we're in and makes the jump and professor and uh yeah i assume at this point you know the labs will just show up in a class and be like raise your hand like do you want this or not and like the whole class is gone and just like they used to do uh at the big business schools when uh price waterhouse coopers would come down and say who's getting on the bus yeah i mean do you see that these like raptures in these in these programs or no well the labs are also very selective uh but um it's a slow i would say it's a slow it's a slow shift well uh everything's accelerated so it's not all at once they don't come and get 10 people at a time or a whole classroom just disappearing uh but week over week and month over month it's happening and it's not for lack of funding like a few billion dollars in in the economy and for our u.s government or any you know modern government that's aligned with democracy the kind of western democracies uh they could afford it it's just the it's that absence of leadership again like getting the right folks to team up uh where they have to have the level of trust you do have to get the best researchers in open to team up and that's something that i personally care a lot about doing uh it takes a certain uh you don't do research yourself if you're meeting with researchers and convincing them to co-fundraise together or to kind of show up to an event and talk about their roadmaps and uh put their names on a website together to to like message and talk to journalists so uh but it's i think it's one of the most critical tasks of our time to to get this group to come together to represent the commons uh the way the internet and there couldn't be more than one internet one had to win and it could have been at&ts but i just hope that and and by the way the ai infrastructure of today there isn't an infrastructure that has won today there are models that have won and each lab has their proprietary infrastructure and if you look at let's say what satya just blogged about satya nadela he blogged about his belief that we will have an ecosystem a thriving ecosystem where enterprises will continue to exist not everything is going to collapse to one or a few ai companies that will eat the entire enterprises can continue to exist and i think that's what satya's point is that we don't want a few labs to eat the whole economy and uh yeah if you were to synthesize um all of the things you hear from the top researchers you talk to every day about this moment and how they feel about it what would that be what would be the the whole economy and what would be the whole economy and what would be the whole economy the high level summary of the pulse of the super elite group that most of the world never hears from i love that question there are very different points in the space uh some people who've left to join anthropic they believe anthropic has this aura around it very different than any of the top labs they're special uh in that they people leave anthropic a lot less often and the sort of messaging goes deep uh about the exponential about the safety concerns about the need to get there first for safety reasons what dario says gets repeated a lot so that's one perspective that's kind of pretty far out there and on one side and then certain researchers who are holding uh strong views on open you know like strong views on the need for open models and like they believe in its essence to democracy so if i had to pick one tldr that describes both one thing that is common across the whole spectrum i think is the the profound understanding and internal experience and now i think getting to emotional experience of how ai is going to change everything so there's this awe and a little bit of because most of these people are builders they're engineering types they like to work on hard problems there's a giddiness like an excitement that like wow like what a time to be alive and that there's an optimism to it and i think there's that too um and there's worry but so i would generalize it to that this like sense of profound power that something's going on uh that they're part of something big is a common thread yeah i mean if you handed someone a tool and said use this to build your replacement you know i don't know how many people in different fields of life and work would look at that and go yeah sign me up i'll do it and for some reason these researchers they are psyched about it and it's it's uh it's an incredible intellectual experiment um it's also kind of dystopian it's also like maybe um maybe it's like everything's going to be great maybe it just leads to more invention yeah but i just that idea to me is something i've been wrestling with lately the idea of you know willingly and excitedly designing your replacement which is what maybe replacement's not the right word but for for those that are less worried about that or for whom they're not losing a lot of sleep i would say that there's this experience that we've gotten from technology that every time that displacement and reskilling has had to happen along the way there were you know switchboard operators at some point and um there were many jobs like they were at some point in the united states a vast majority of people were farmers worked really adjacent to farming technology has reduced the number of people that do farming uh there are elastic and inelastic job types and that's what i've learned through the reskilling projects that were funding um so technologists tend to have this optimism where as technology displaces some types of skill sets that humans do we reskill and we move to a higher level of abstraction which gives us more leverage over the world around us and lets us i think this is where the different points of view emerge of some believe that lets us to can let us get to a spot where we just we live these prosperous lives of just doing whatever we want this renaissance life is it still as relevant though when that uh transformation is so overnight and so cross industry relatives in the past i think it breaks this it makes it so that you can't lift and reapply how this has worked smoothly in the past without having to solve new types of problems and that's why talking to the people who are doing research about reskilling is so interesting and economists working on it and thinking about it and like the experiments they're running in the beyond coding like people who use ai in call centers because that's another part of the economy that's been disrupted how like do ai help the best performing uh people who answer customer support calls or the worst performing or the median and are any of those people worried and um can we use it itself to teach people faster to be good at customer success or customer support so kind of hearing the some of the results from the people studying reskilling i do think it it is it will be possible to uh to achieve some like we should be working on it and i i'm an optimist when it comes to i believe we can solve problems i believe we may need to do things quite differently and be more innovative and uh hopefully and i and i also think that there's some big problems some big problems and some hairy ones just rather someone say the human uh desire is infinite and we'll always make things for that you know what i'm saying as opposed to kind of like some something a little bit fuzzier i don't know uh i sure hope it works out you know i do i i believe um that the the risk of downside like there are others who are worried about safety right so there are clear downsides uh biological threats job displacement is a big one um just economic instability from the rate at which jobs are are or job function this place one of them one of the most concerning for many people is the widening chasms between different pockets of perspectives on and on you know look at the left and the right or accelerations than Doomer's or these spectrums and We seem to be pooling at the ends and that's different than when I was a kid, like in the era of Walter Cronkite, you know, people kind of tended to talk to their community more and you disagreed more respectfully, more frequently than you do now. And I think that that's actually a set of social conditions. Centralization of power. That was a lucky centralization of power then. Uh, wait, which part of that was centralized? Oh yeah, sure. Uh, he, I mean, he wasn't the only one. No, there were like three. Yeah. Okay. So there, there, uh, that maybe, maybe that was true. I guess there was centralization across the broadcasting networks. That's not what you're pointing out. Um, but, and I think that that, that is something that polarization that I worry a lot about. And it actually was a trend that we were seeing before the AI paradigm from the social media paradigm. Maybe you and I talked a little bit about this last time we chatted. Uh, and I think AI can make it much worse. So, um. And the only thing in common through it all is meta Facebook. I hope that it comes for podcasting last, you know, I, I don't know when we're going to send a podcasting agent into an interview instead of us, but maybe a year, maybe two, who knows? Uh, maybe never, maybe, uh, you know, I, I don't know that I believe we get to decide, you know, um, what we're going to do. And I believe we have to. Uh, prerogative and it's, I guess maybe most importantly is I think a choice you two have been made to be journalists right now, independent journalism, covering AI and a choice that I think, uh, I advocate for computer science researchers to do, which is to say no to multi, uh, million dollar per year offers, uh, have that in common of expressing agency at a time where you, you, you, you two have voted with your feet that you think you can make a difference. You don't even want to know the offers. We've turned down Andy, uh, blow your mind. Um, well, maybe let's end here optimistically. So you're a year into law or almost a year. Um, you've done a lot in a couple years. If everything goes right, what has happened? How is the world different? Great question. At the heart of it all. As I said, we give research resources to researchers who can have impact on the, on the, the world. Uh, we, I want to help researchers. Build technology that moves humanity forward. So we happen to be at a time in history where things are moving forward faster than they ever have. Um, we have to deal with that lots of hard problems to solve, and I'm out there trying to talk to the, the absolute best candidates to solve that kind of problem all the time. So in, in an ideal world, we have this, uh, thriving set of, you know, ecosystem of researchers working on the world's hardest problems. Being super smart. Being super well-funded, uh, and spinning out in impact in the form of projects and products that, uh, kind of continued to make a better future for my kids rather than one that we could lay awake worrying about. Hmm. Well, this will leave it there. Yeah. Thank you. Yeah. Thank you. Thank you. Super fun. I appreciate it. Yeah. And that's it for this week's show. Stay tuned after the break for Alex and my reactions. We'll be right back. You think you know a browser, but Gemini and Chrome, that's new. All right, Ellis. I feel like we should have worn robes for that. I know. Being in this room. Well, so you called him the Gandalf of AI. Is that something you made up? It is. I hope he doesn't feel offended by that. I think he'll like that. It's meant to be a compliment. I mean, when you see the beard, how can you not, right? Oh, of course. I think all beards know that Gandalf is the goat. A wise sage. You know, I think he definitely fits that bill. As I'm trying to embody. Yeah. The Rick Rubin for storytelling vibes. I did begin growing a beard, which you can see in some previous episodes, but it just looked like absolute shit. And so. I like the look now. You like it now? Yeah. I just shaved last night. So I look freshly bar mitzvahed. It looks good. I don't know how long you'd have to grow a beard for it to be like Andy's. Yeah. I just don't have the density. I don't think. The density is impressive. Many secrets in that beard. I was going to say many secrets in the beard, many tickles for the children. Uh, that sounded terrible, but, uh, as you heard on the episode, hopefully you're talking about a better future for our kids. Yes. That was the context. That was the context of that comment. Yeah. I didn't expect this to get so, uh, you know, big picture, um, you know, problems in the world. Well, there's two kinds of big picture, right? There's like lame big picture that like maybe an executive will say that isn't actually saying anything, but, uh, the dude has clearly read his books. I mean, really nice to just like have a historian on this. I mean, I don't know, I don't know about you, but like so many of the folks that I talk to and I'm around or younger and often working on like first principles, we can deduce whatever we need to do here. Of course we can, you know, but it's like, there are actually some lessons to be learned from history and it's not that they're ignoring it. It's, it's just that, you know, it's like build it from scratch, do it yourself, figure it out along the way. And I'm not sure how often, uh, the history textbooks come up, wouldn't it be something if, uh, th this would actually be. Maybe a good prompt, maybe I could add this to, to my, uh, prompt for Claude or poker or whatever. Um, anytime you tell me some information, uh, grounded in a piece of history that is, that is relevant or instructive in some way. I mean, wouldn't that be something when you ask something of, of ChatGPT or Claude? You would be that guy at every dinner that's like, well, did you know that in the sixties, blah, blah, blah, like you'd just be doing that all the time. I mean, better than only knowing what you read on Twitter, I think. That's true. And you're saying this as Ben Franklin's book. Bust is literally behind you right now, which is very apt. Yeah. Uh, speaking of though, I mean, we talked a little bit about, um, uh, Andy's a fan of science fiction and so am I, and there just don't appear to be too many science fiction writers who are, I mean, there hasn't even been time I think, to like respond to the AI wave and then write an actual book about it. And I mean, can you imagine like if maybe there's a book coming out next year, that's talking about ChatGPT 3.5 or something like that, they would have had to cancel it or something. Yeah. But, uh, I don't even see too many people asking it. Yeah. I mean, I mean, I don't know. I don't know if there's like AI about what the utopia would be and, and run crunching some numbers on it. We should try that. It just seems like not, not very many people are working too hard to try and, I don't know, play it out. And so it was nice to hear, uh, from someone who really is thinking about it. Yeah, for sure. Uh, we didn't say this during the episode, but the last time I hung out with Andy was a dinner that he had Neil Stevenson come to, which was quite a flex and a bunch of top AI researchers were there and to see how they were all just so fixated on Neil and had, clearly read all this stuff and played a formative role in shaping who they were and what they were working on. I mean, sci-fi does become reality and it is like every day, especially right now. Yeah. I, uh, that surprised me because I didn't get the impression that Neil Stevenson and the other, you know, elder Statesman science fiction writers are fans of tech in any way, shape or form. And I think a lot of them are like very actively antagonistic and I do believe like, yes, while AI people read their work as a teenager or a young adult, in a lot of ways, those were like very different versions of the future, like very different versions of where technology was going to go, whether it was blade runner or, uh, you know, otherwise, whereas I think AI is kind of like this future that in a way no one really expected. I think all we really expected were like, oh, some days we're going to have robots who could talk like people, but I don't think anybody really in the science fiction world, uh, at least that I've read, and I've read a lot of it, expected that there was going to kind of be this like embodied AI character that could summon the power of 5,000 PhDs. I feel like one of the closest things I could think of is Dr. No in AI artificial intelligence, the underrated Steven Spielberg movie. And that was kind of like a chat GPT type situation, Steven Spielberg, dude, I just saw disclosure day. That shit was so trash. Oh my God. Should I? Oh man. I have tickets for, I have tickets for tomorrow night. Oh my God. I just, I feel like it's like a mandatory viewing. It's so bad. Oh no. Anyway. Um, yeah, no, I, I really appreciated, uh, everything Andy shared with us. It's nice to talk to someone who can be that, uh, LLM for the modern AI researchers brain because he talks to so many of them. Can you imagine the group chat? Oh, I'm sure it's crazy. What would you do to, to, to be a fly on the wall for that group chat? Well, I need to figure that out with Andy after this. Um, but yeah. Uh. great. It's really fun to do this in person too. And, uh, to do it. uh, in this really cool undisclosed location. Yeah. Well, one of the things we talked about too is just kind of like, you know, not having too many leaders to look up to. I mean, is there anybody that came to mind for you? I wonder that's kind of on the, on the, on the more optimistic side. And, and I think we'd all agree that, you know, having a bit of both is a very healthy, healthy tension and, and healthy skepticism. And I do think if you're in the, in the world of, of science and research, you have the skepticism built in, you know, the last tech co-founder leader that I felt like really spoke for the quote unquote common man that I remember being impactful in my life was actually like Steve Wozniak. Um, and they're really, there's no like Wozniak of AI. Yeah. So no, no role models to speak of. Yeah. Well, what do you make of Dario's latest writings though? Uh, to be honest, I haven't gotten through all of it. Um, yeah, I just don't think he's a sympathetic character, right? I mean, he's, uh, he's the guy out there saying all the jobs are going to go away and that he's inventing the thing, doing it. And then it's also like too powerful to be used, but then freaks out when the government says that and makes it so, so I don't know. I think anthropic, uh, it could be a really hairy, you know, fall for anthropic given just the trajectory of all this stuff and the way they've messaged it kind of coming to head with the tech and rolling it out and trying to get it out. And in the absence of, uh, I think he has written a bit about like more specific predictions for the future. And, you know, Andy was talking about curing cancer, which is, which is definitely a common one, but in the absence of a single North star, that's kind of like an outcome. Uh, it certainly seems to be moving researchers to join them just to kind of be, uh, to, to save the world from itself. That's certainly a motivating mission. Yeah. I mean, I think Andy's doing something that's like strategically very unique. And, uh, I do think a lot, I mean, I just know this, that a lot of top AI researchers are paying attention to it. And I think that's a really, really important thing. And so, um, yeah, I mean, I think the thing is like, they've all made so much money that now already that they, I think, want to be doing something that, uh, will be something that their kids look at and go, Oh, wow. Like, I'm glad you did that. I don't know. I think stuff like what Andy's doing represents that for a lot of researchers. It's like, Oh, I can like, I don't have to go just make chat GPT better. I can do something that's open and helps push forward academia and all of that. So I think what he's doing is really, really important. Yeah. I remember reading that bit, uh, even from, I think it might've been from, uh, from Dara's point of view on like how few folks are leaving because of that mission. And it is an attractive thing. And I think Andy's telling a different story in that regard as well. And, um, I mean, people are also not leaving Anthropic because that stock to do go up. Um, but, uh, yeah. What does an AI research buy with their, uh, FU money, Alex? I don't, I don't know. Um, they don't, you, you wouldn't really be able to tell. I mean, there's not a ton of social signaling in the AI research world. So I don't know, bunkers. Well, should we leave it there? Yeah, let's leave it there. That's it for this week's show. Thanks to Andy for coming on. If you like the show, don't forget to like, and subscribe everywhere you get podcasts. We are access.show online, and you can find us in video at access pod on YouTube. If you really liked this episode, we would love a five-star review or even share with a friend or agent. You can find my newsletter at sources.news. You could find me at hamburger on Twitter and at meaning.company for your storytelling needs. Access is part of the Vox Media Podcast Network, and the show is produced by Hooked Creators. Bye. To a state surrounded by nature. That's nice. Go on, book it. It's easy. Booking.com. Booking. Yeah. I didn't like what you said, Odin pauses, about the future. This is the love story of real hinge couple Odin and Edward. Written and read by me, Curtis Garner. Listen to the free audiobook now.