GameStop + eBay, Neural Computers, Elon vs OpenAI, Export Controls, Vibe Coded Projects

I see a large IPO on the horizon. You're surrounded by journalists. Hold your position. Let's misinformation. >> Courtney declaring order inbound. >> Let's just roll. We are surrounded by general. Hold your position. >> Cop. it up. Trust the experts here. Five to founder. I see multiple journalists on the horizon. Stand by. >> UAV online. >> Blaze. >> Double blaze. >> Triple. >> Another one. >> Double kill. by roll team deathmatch. match. We are experts. Triple blades. Let's just roll. Right. Mark clearing order inbound. Come on, get up. >> We are surrounded by journalists. Hold your position. >> Strike one. >> Strike two. Activate golden retriever mode. >> Market clearing order inbound. >> Five good. founder. >> You're watching TVBN. Today is Monday, May 4th. >> Would you look at that? >> 2026. >> Would you look at that? >> Would you look at what, Jordy? Everything looks fine to me. >> We're back. >> We're back. >> That's That's what I'm looking at. Uh we're we're getting ready at the last second. We may have installed uh Modern Warfare uh 2 on this large screen and gotten a little bit behind schedule with some of the interns playing. >> Yeah. Ben and Tyler were >> it was a knockout dragout fight. You you won by a lot, right? >> Yeah. >> Well, we'll have throughout the show. It was pretty embarrassing for Ben considering that he is still in that um not not not uh not chief chief producer Ben uh but other Ben uh considering he's still in like the main kind of like pro video game. >> But at the same time, I mean, you were probably two years old when Modern Warfare 2 came out. So, you got to sort of relearn the the old the old tricks the old tricks on Rust. >> Anyway, well, big show. >> Hope you guys had a great weekend. >> Hope you had a great weekend. uh big week. We'll be uh we are going to a conference on Wednesday and Thursday, so we might uh be off both those days, but we have some great shows for you planned Monday, Tuesday. >> Conference. >> It's a conference. >> Oh, I thought you said concert. >> No, although I think there might be a concert. >> Extremely that would be extremely out of character. >> I think there might be a concert or or at least some sort of musical performance at this particular conference. But uh >> who do we have coming on the show today? >> We have a bunch of folks. Nat uh Michael York from Casa uh Anne from AMP. >> We got G from Panthala. >> A little more context on Nat joining. He he joined Alpha School. >> Yeah, that's right. >> Uh he's the head of founder development. They're launching a founder track >> at Alpha School, which I'm very curious to hear about considering that >> uh a lot of the parents in my uh in my area are very Alpha School curious. Yeah. >> So excited to hear about that. >> Um well, uh I mean there's a bunch of news we're going to go through, but uh first I was just uh sort of reflecting on uh there was a great interview with Andre Karpathy at Sequoia AI Ascent. I think last week it went up on YouTube and uh it was he was sort of reflecting on how his workflow is changing around vibe coding and I was sort of reflecting on how my knowledge workflows are changing particularly around image generation now that uh image generation is really good at infographics and effectively designing slides or output. Uh and so we're we're starting to see the rumblings of this idea of like the neural computer. There was this uh people have been talking about this for years since the AI boom began, but the basic idea is like you would have a computer that basically has no software whatsoever on it. It would just have an LLM or just an AI model, just inference capability or potentially connection to cloud inference. Uh that would generate whatever you want, whatever you need on demand on the fly. And so I think Elon talked about this with macro hard a little bit. That was p a piece of the vision. This is uh always been uh theorized but it's becoming more and more real. And so uh Cararpathy describes this idea of like a neural computer this way. And I think it's an interesting framing. Obviously it'll have implications for uh SAS products that might be used in a headless under the hood scene uh uh way or uh might be competed with against these neural computers. Carpathy describes it as uh he says imagine a device that takes raw video or audio into basically what's a neural net and uses diffusion to render a UI that's unique for that moment. And so this sort of like on the-ly uh instantiation of the exact UI that you need for that particular question or whatever problem you're trying to solve, whatever you're trying to do, uh is an interesting paradigm shift that it feels like we're starting to see glimpses of. So, um, I most recently felt it when I was trying to understand Ryan Cohen's proposal for GameStop to take over eBay. This is a big story. We'll go through it today. Uh, but I haven't tracked each either either company closely. We've had Ryan on the show. Uh, and, uh, we've talked about eBay and GameStop intermittently, but I couldn't tell you off the top of my head what's the revenue for each company, what's the profit like, what what are the different multiples? Uh, and so in a pre-CAT GPT world, I would have gone to Google Finance or Yahoo Finance and pulled some data, maybe had two tabs up, maybe used one of their comparison tools. If I wanted to be really advanced, I would have copy and pasted the stats into uh a spreadsheet. If you're really working on Wall Street, you might have like Cap IQ or Bloomberg plugging into a sheet uh an Excel sheet that then can uh build you a comparison table and do like comps. Um and then once we got into the chat world, you might do a deep research report, pull all that data, put it into a table which is effectively markdown and and sometimes the tables renders a little weird and like you can kind of bounce around. Uh but now I the whole process from start to finish is just a single prompt and it outputs an image. So you can pull up the image that I generated. So this was this was one prompt. I said do a bunch of research on GameStop and eBay's valuation and key financial metrics. things like growth rate, topline, earnings, revenue, valuation, how the multiples fit together. Build a nicely designed side-by-side car comparison of the two companies. Uh, and you wind up getting something that is very digestible. Like just looking at this, I mean, it's obviously a little zoomed out, but you can zoom in and see, okay, eBay has about uh three times the revenue, 50 billion versus 15 for GameStop. Uh, and uh, revenue growth. eBay is growing. While GameStop uh shrunk by 5%, eBay grew 8%. Uh operating income, eBay has 10 times the operating income at 2. 2.28 billion versus 22 232 million for GameStop. And so you just get this like very easy. Okay, what's the operating margin? eBay's up at 20%, GameStop's down at 6.4%. And so you can start to see on a price to sales ratio, GameStop's at 4x, eBay's at four and a halfx. But on a on a market cap to net income, GameStop's higher has a higher value, 34x uh versus net income versus eBay is 25x. So you can just sort of see this table and this is something that um usually would have been like three or four steps to get here and instead it's just this single prompt. And so uh I think that there's uh like this is not a perfect result. Like in in that image, you can see that like it it chose red as the color for all of GameStop's financials, which is not what you'd normally do because red is usually for negative numbers, but those revenue figures are positive. Like, it could be better. I could probably go further and prompt it a couple more times to get exactly what I wanted, but it solved my problem of having like here's the summary of the question that you were actually asking, which is like how do these companies stack up to each other? What's the relative size of the business? what are the strengths and weaknesses of each of them? And then boom, you have a square image that you can easily text to someone and and it's ultimately sharable. And more importantly, I don't care what it used under the hood. It could have puppeteered a a spreadsheet and put it all in commaepparated values and make a CSV and it could have transformed it with Excel or Google Sheets. Under the hood, it could have written Python. It could have used pandas or scikitlearn. It could have done anything it wanted to, but it's all abstracted to me and I don't even think about it. And this is different from the previous era of like, okay, well, if I wanted to do some sort of stock comparison tool, I could vibe code a stock comparison tool with API integrations, make sure you have the the the data connections, but it's just kind of less necessary as the models get fatter and they sort of eat more and more of the process. And so, Carpathy describes this concept as software 3.0. And we should pull up his example because it's very similar. Of course, it happened like months ago because he's ahead of me on everything obviously. Um, but he gave a good example of shifting from like you have a problem that there's no solution for, so you're going to vibe code an app to just a few weeks or months later like the AI tools can just do it and you don't need any code. You don't need any system to build even though it's fun to build a system and it's interesting and it allows for more maybe more speed, more reliability, like more and more things are like oneshotable by the model. So let's pull up Andre Karpathy uh talking at Sequoia AI Ascent about his experience with software 3.0. >> I think one more maybe uh example that comes to mind that is even more extreme than that is when I was building um menu gen. So menu genen is this idea where you um you come to a restaurant, they give you a menu, there's no pictures usually. So I don't know what any of these things are. Uh usually like 30% of the things I have no idea what they are 50%. So, I wanted to take a photo of the restaurant menu and to get pictures of what those things might look like in a generic sense. And so, I built I've vcoded this app that basically lets you upload a photo and it does all this stuff and it runs on Versel and uh it basically rerenders the menu and it gives you like all the items and it gives you a picture that it uses an image um you know generator uh for to basically OCR all the different titles uh use the image generator to get pictures of them and then shows it to you. And then I saw the software 3.0 version of this which is which blew my mind which is literally just take your photo give it to Gemini and say use nanobanana to overlay the the things onto the menu. Uh and nanobanana basically returned an image that is exactly the picture of the menu that I took but it actually put into the pixels it rendered the different things in the menu. And this blew my mind because actually all of my menu gen is spirious. it's working in the old paradigm. That app shouldn't exist. Uh and uh yeah, the software 3.0 paradigm is a lot more kind of raw. It just um your neural network is doing more and more of the work and your prompt or context is just the image and the output is an image and there's no need to have any of the app in between. >> Um so I think that people have to kind of like >> apps. Um, yeah, it it it I mean it's it's real and uh and it's and there were I I had some takeaways from this like what are the implications for this and I think there's a few things. Um the the the first thing that was on my mind was that al although we have gone through this crazy vibe coding boom where everyone is vibe coding apps um it feels like a very temporary aberration and also I know that that like even though there are millions and millions of people that have used codeex and claude code and open claw like the numbers are big but it's not at 20% of the US population like it's just not at that level of adoption. as opposed to chat apps which are at like 70%, right? >> Yeah. The other the other thing that's been interesting is uh non like people outside of tech that have gotten into vibe coding that have been pitching me their ideas here and there >> almost every time they're pitching me the idea. It is something that the that cla code and codeex can do themselves pretty well today like just in one chat thread >> or the app like the apps can do them. And that's what I'm like what's blowing my mind now is that is that in many ways >> that's what I'm saying. So like they're using they're they're using vibe coding tools to vibe code something that doesn't necessarily need to exist because you could just use the app itself to do the thing >> and they're already widely available. So it's it's been >> it's been interesting. >> Yeah. So I think there's I think there's two things like one is that if you uh if you've been like hesitant to jump into vibe coding uh because like it's just it's a little bit too much a hassle. Like Andre Carpath is like obviously very very comfortable being like oh yeah let me deploy to Versell and do all this like you can figure all that out but that leads to this world where it's like oh I was staying up all night. I was I was really really like burning the midnight oil to get this app deployed and like do all this stuff. Well, like a lot of that's going to go away and like you're not going to need to do that. uh and frontier models are but then there's also the the this question what you had which is like there needs to be this higher order loop of thinking around okay you have a problem should you actually vibe code an app or should you just try and oneshot it with the current model capabilities because for a lot of things like >> and within an yeah within chat within Gemini within claude the actual apps like you can like take a picture of your food and say hey start tracking my calories. Like there's a lot of things that the that the apps can just do in one chat thread that people are doing, but I think that uh there's like this tension between when when you actually need to for sure go and vibe code something versus when you can just do it in like a one-shotted LLM context. Um and so like frontier models are already able to in basically 90% of situations I feel like instantiate exactly whatever is required to solve the actual problem uh under the hood. uh entirely abstracting away like code and tools like you will just not be aware of what's happening and it doesn't matter. Uh yeah and then uh the second >> and and I yeah I would I would add to to my previous statement by by saying that doesn't mean that there's not necessarily a business there because sometimes taking a raw capability and presenting it to people in a way that's very easy for them to digest. You can still deliver value and you can get customers and people will pay you money. >> Yep. But it is it has been fascinating to see like does this actually need to be an app? >> Yeah. Yeah. I mean there's there there's a ton of apps and software that will still be valuable whether it has a liquidity pool or some sort of unique uh source of strength or some differentiation point that that the existing chat apps can't hack at all. Um, and then there's also just like like marketing arbs effectively where it's like, okay, yes, the any any frontier model in a chat app could do this, but you weren't aware of it and this company was really good at running ads to actually get awareness going and then drive downloads of this specific thing. And so we see those in the uh uh in the app store all the time. Uh so the other thing that I was uh uh reminded of was did you ever read uh Union Square Ventures 2016 blog post fat protocols? Are you familiar with this? So fat protocols uh was this concept around how in in the web like web I guess 1.0 2.0 know there were protocol layers which are like TCPIP, HTTP, SMTP like file transfer protocols, HTTP and there and a lot for like a couple years the crypto community was like the the group that developed and maintained like HTTP. They basically created the standard that the web ran on and yet very little value acrewed to the creators and the maintainers of that protocol. and crypto would be different because the Bitcoin protocol would had the value capture component like baked into it. And so there was this idea of like the application layer in blockchain would acrue very little value and the protocol layer would capture the vast majority of value. So this is on the web the applications on top of HTTP. You can think of like Facebook as a beneficiary of the protocol of HTTP because that's how what how how the actual information and the photos and the text gets transferred to you. But the HTTP standard does not acrue the value the value acrru to the to the application on top. And if you scroll down you'll see the blockchain example which was sort of borne out uh that the application layer was pretty thin on top and most of the value went to like the tokens and the protocol below. Yeah, exactly. Ethereum is a good example. Salana is a good example. Of course, there's there's value in the application layer and there's some companies that are being built. But this was basically this thesis that uh the the uh the he says we see that very early. We see this very clearly in two dominant blockchain networks, Bitcoin and Ethereum. The Bitcoin network has a 10 billion market cap. >> Wow. I think it's like a trillion now, right? Isn't it 700 billion? Um yet the largest companies built on top are worth a few hundred million at best. Now we have Coinbase which is in the tens of billions. So both sides of the protocol application layer did very well but uh the point is still true. Uh similarly Ethereum has a $1 billion market cap even before the emergence of real breakout applications on top and only a year after its public release. And so that uh that was sort of the core thesis on this fat protocols thing. And and I think that there's something similar happening in the AI value chain. Of course, there's like a bunch of other dynamics going on in in the AI value chain, and there's a lot of capture and and complicated market dynamics, but the models feel like they're getting fatter every month, and they're sort of eating away at the edges of what you can do with them. And uh and and so increasingly, you can just get more and more out of the core model, which is just an interesting dynamic. Uh and then third, there's still like this huge question of like walled garden jumping. We've talked about this before, but uh it's almost we need like a different term for like the dead internet theory. It's like the it's like the walled garden internet theory. Like the internet's not dead. Like there's great there's great information in Substack on certain legacy legacy media websites and on Facebook and on X and on YouTube. But like all of those companies don't want to interact with each other. And so that's where you get something like, oh well, like if you write code, you do get access to it loosely. Or if you're puppeteering a browser on a Mac Mini, you get access to that. Or if you're digging through iMessage locally, like that can require a different workflow. Um, but that's more of like a legal and business discussion than a technical one. Like there's no reason technically that a single LLM wouldn't be able to just query every single web resource except for the fact that the big the various tech companies don't want each other to talk to each other. Uh and so um the models I think will continue to find a way under the walled garden over the walled garden through the walls like they'll they'll seep everywhere. uh and it's more of a question of just uh inference cost how how long it takes to actually grind through the wall but uh they they're already figuring out a way around and then and open claw is a good example of that where uh a lot of the walled gardens were sort of brought down by running >> yeah except I think SAP came out and said no a no unauthorized agents here they're trying to they're trying to put up the walls they're trying to build a moat they're trying to get some alligators to scare off the the agents >> yes But I would be very I would I would be very surprised if they're able to stop me from if I have SAP and I'm running it locally for me to take a screenshot of my computer and then tell the mouse to go where it wants. Like it's very hard to fight back against these like the computer. I mean I I've seen this guy on uh uh on YouTube who uh uses like ever more contrived aim bots. like he has one that's uh it's like a robotic mouse. So the it's using the actual mouse and keyboard, but it's robotic fingers on it and then it's controlling. I think he plays Rainbow Six Siege or Counter-Strike. Uh and he's cheating, but it's look it's a camera looking at the computer. So until you get to like, you know, worldcoin eyeball scanning, making sure that you're not using this. He's also done one where he I think he put like electro stimulation on his arm that would do the aimbot for him. And so it was his arm physically moving, but it was puppeteered by an AI essentially. And you know, it's very very demo phase at this point. But uh imagine that everything is so locked down that no AI agent can interact with SAP, but then I have the electro stimulation and I can just type it super fast because and it just does whatever it needs and I'm the I'm the the the human instantiation actually pushing the keys. Run had a good tweet where he was like, you know, people are now just like vessels for the AI where they just like the AI tells them what to do and then they just act exactly like what the the model said. >> Wasn't uh John about this >> John Collison was saying uh like the humans get the thing off the high shelf. It's just like every time I have to like go and like export a PDF and upload it to Chad GBD because I can't get it in there by default. Even though I could just give them a web URL, I have to export it, print or whatever. Uh the this is the >> let's talk about I think we should talk about GameStop. >> Let's do it. What's up with GameStop? >> What's going on with GameStop? Yesterday uh it came out through the Wall Street Journal. GameStop was preparing to make an offer for eBay as part of Ryan CEO Ryan Cohen's plan to turn the retailer into a hundred billion dollar plus juggernaut. >> GameStop has been quietly building a stake in eBay shares ahead of a potential offer and could submit an offer as soon as later this month, which they did uh uh this morning. if eBay isn't receptive, Cohen could decide to take the offer directly to eBay's shareholders. >> And uh they released a letter >> uh yesterday >> uh to Paul Presler, who's the chairman of the board over at eBay. Happens to be a friend of mine. Uh yeah. >> Wait, really? >> Yeah. Um uh I I I um um our his daughter and my wife are good friends. Um so I end up we end up hanging out a decent amount. Um and we're neighbors. >> Does he do a lot of does he do a lot of uh like podcasts or press appearances? >> It can it can we can it can be discussed. >> I I'm just saying like determinism of pressler would be like somebody who dominates the press. >> The press circuit. >> I think you press circuit all the time. Um, anyways, Paul Paul is fantastic. >> Uh, and Ryan wrote this letter to him yesterday saying, "Gametop is proposing to acquire all common stock of eBay at $125 per share. We have accumulated a 5% economic stake in eBay through derivatives and beneficial ownership of common stock and are file filing a schedule 13D and HSR notification tomorrow. Our offer is $125 per share comprising 50% cash and 50% GameStop common stock, which we will get to in just a little bit. Um because Ryan Cohen discussed this on CNBC this morning. >> Yep. >> Uh that represents a 46% premium to eBay's unffective closing price on February 4th, 2026, the day GameStop started accumulating its position in eBay. Uh and blah blah blah blah blah blah blah blah blah blah blah blah blah. But let's go straight to the CNBC. Quickly, there was a question. So, uh, Bizlet said, "Can someone please tell me how GameStop has 56 billion? It's not a 56 billion company." There were questions. And Ryan Cohen went uh in the ring with Aaron Sorcin over at CNBC or Andrew on, sorry, uh, Andrew Ross Sorcin on CNBC, Squawkbox to interview about the GameStop eBay acquisition. We could play this. walk us through how how you could get to that price and how it would work. >> It's on our website. It's half uh cash, half stock. Uh but but the details are are on our website. >> C can you help I I I've read them, but can you help our audience understand them? >> Yeah. What what which part exactly? Well, I think we can start with the idea that the market cap of of GameStop is call it 11 billion. Uh you have $9 billion uh on your balance sheet. Arguably, if you're if you're providing uh effectively all of your stock and then and then the cash that gets you to 20, you have this letter from TD, that's another 20. Uh we're now at 40. uh but we're still off uh by call it uh 16 and and the 20 as far as I understand while it's considered a highly confident letter meaning TD saying they're highly confident uh that they would provide the financing it's not locked financing >> yeah we'll see what happens >> um >> founder no >> doubt I I hear you >> I understand that. I'm I'm just trying to understand where the the rest of the money would come from. >> It's half cash, half stock. >> I I I'm I hear you. I'm just saying that that math doesn't get you to the to the price that you're offering. So, >> that's a pretty straightforward question. I don't get it. Thank you chiming in. >> Where's the rest of the money coming from? Andrew laid it out pretty clearly. >> I I don't understand your question. We're offering half cash, half stock, and we have the ability to >> issue stock in order to get the deal done. >> But the full details of of the offer on our are on our website, >> but you're on our air. We we thought we'd get >> So, but I don't understand your question. Where's the money coming from? That's the question. You're wondering record scratch freeze frame. You're wondering how I tried to buy a $55 billion company with $40 billion earmarked. >> Yeah. So I don't what what do you think he was expecting this to go out on Sunday, Monday, GameStop stock to pop like crazy >> and it's actually down today? >> I guess that's possible. I I I don't know. I I also don't understand why he can't just say like, "Hey, we're in the process." Like, "We got a highly likely letter from a bank for 20. Yeah, we need 16 more, but we're going to go get more letters from other banks. We're going to go get other equity investors. Like, this is a whole process. We're excited to announce this and this is like our first close. Like, we're not like fully ready." But maybe >> so the tough thing is he took this offer to Paul, chairman of the board. >> Yeah. Now Paul has to look at this and be like, "Okay, is this a real offer?" Yeah. >> And I imagine Paul will watch CNBC and it's not. >> That's not going to give anyone a lot of confidence. >> Yeah, it's Yeah. I mean, you could imagine it in the in the context of like buying a house. You know, you show up and someone says like, "I have 80% of the money and like my bank will underwrite me for 50% and I have 30% in cash and you're just like look, I need the full amount. I need the full amount for someone. I'm I'm wondering I'm wondering what uh Ryan I I'm he you know eBay is an incredible business. It's been remarkably durable. They've faced an onslaught of competition for every single category from sneakers to watches to art to name any cars, right? Any category there is like a vertical competitor to eBay. >> Yeah. >> And yet the business has done has been remarkably remarkably strong. It's up pretty meaningfully this year, right? management is executing and uh >> and and again it's unfortunate for Ryan and his bid that the most viral sort of video clip out of all of this is just him failing to answer like you know a pretty straightforward question and not having an opportunity to talk about okay why do you you know if presumably if you're buying this company you think it it can and should be worth a lot more why what what's your what's your plan what's your plan for the business why Why are you better suited to run it than uh than Jamie, who's been in in the seat since 2020, worked at eBay from 2001 to 2009. >> So, he's a veteran, knows the business very well. >> And uh you're coming in $16 billion short. At least that's what it looks like. >> And um anyways, uh this bid could be could be over before it really started. >> Yeah. Or I mean it could attract a bunch of investors who want to line up and fall in line and wind up producing the the 55 or so required. But uh it does feel like it's uh it's a ways away. Uh anyway, we have our first guest of the show in the waiting room. We have Matt from Alpha School and here he is in the TVP Ultradom. Matt, how you doing? >> I'm doing great. How are you guys? >> We're doing fantastic, >> dude. Incredible to finally have you on the show. We enjoy, we love when your name pops up in the chat and uh we've both loved >> following your career the last appreciate that as long as I've been on the internet. >> It's awesome to be here. I mean, I've been following you guys since >> it feels like must have been the first week or two. Loved the show and loved seeing you guys crush it. It's brought a lot of joy to my my weekly commutes. So, thank you guys for doing this. >> Amazing. >> Well, and yeah, we're we're super excited. I personally am excited that that you're at Alpha School now because um yeah, Alpha School is moving into my town. It looks like it's early early stages and I've just been curious to learn more about it and um yeah, especially like what you're working on specifically. >> Here to answer any of it back open up all over. So >> yeah, why don't you give give us an overview of like everything you've worked on the last few years because like every every I I think you've you have done a better >> job than almost anyone at identifying these sort of like big macro technology changes and cycles and then like just like experimenting and innovating in in a bunch of cool ways. So um so yeah, take it take us back a little bit. >> Yeah. So I I got the entrepreneurship bug in high school. didn't have many ways to get into it then and then in college I you know I went to business school and figured business school teach me how to be an entrepreneur realized really quickly that that wasn't the case and ended up just trying to go figure it out on my own so got super deep into SEO ran a really successful uh SEO focused marketing agency for a number of years until uh 2021 basically handed that off but eventually got acquired but that was kind of my first taste of seeing this like cool thing in technology trying to figure it out, trying to build a business around it, and just feeling incredibly fulfilled by by getting to do that. And so when I stepped out of SEO, I got pretty heavy into the personal knowledge management space. So was doing all of the like Rome research and and notion stuff, got deep into crypto during that 2021 2022 era, uh, and was like writing smart contracts and working with a couple teams through that. And then when like building with AI really became a thing a few years ago, I was using cursor like right at the beginning well over a year before we had these terms like vibe coding and whatnot and was just completely obsessed with it and was trying to figure out what I wanted to do with it for a while. I had a course that was really early on how other people could start learning how to vibe code. And at the end of last year, I said, "Okay, like I actually want to I want to stop doing the solar printer thing. I want to kind of like go after something much bigger. And while I was in the process of figuring that out and tinkering with a few bigger things I wanted to work on, Alpha School reached out to me and they told me that they were working on this wild idea for a new high school and they were looking for a kind of scrappy AI native entrepreneur to come help them build the curriculum and get the whole thing launched and built. And it was kind of one of those like, do you know anybody else who would be good at these kinds of things like you? And I'd never considered taking a job, certainly never considered working at a high school, but when they told me they wanted to build this and that they were looking for somebody like that, I I was just immediate like I I had to come do this. >> Okay. Before we get into alpha school, why you you're you're witnessing the explosion of of LLMs. Why didn't you get into uh like answer engine optimization or AI observability? I'm sure you thought about it. Uh and given your background uh building the the SEO agency to to exit, I'm sure you would have been well positioned to to do well there. Did you did you consider that at any point? >> I thought about it and to be honest, I was just kind of tired of the SEO game. I'd been so deep in that for so long that I just kind of didn't want to go back to that. And I just loved how fun building stuff with these AI tools was like on the software side. And that was kind of the route that I was starting to go down. And you know, thankfully like this still this still kind of heavily scratches that itch because like even though this is a school and we're we're learning how to, you know, we're teaching and trying to build all of that out, so much of the internal systems and how we figure out how to teach this requires getting really deep with AI and building a lot of our own internal custom software. And it's really cool to like be a part of a school that's also investing so heavily in that. Uh, and so it's been fun getting to bring those skills and that energy here. >> Awesome. Uh, introduce reintroduce Alphas School. We've talked about it on the show, but for anyone in the audience that that isn't familiar, um, I think it would be helpful. And then I want to talk about the program that you're building out >> for sure. And just to double check because I I did get a message that my mic wasn't sounding right. Is it >> is it not plugged in? I'm looking at it. It looks like there's no cable. >> Oh my god. just like looking at. Wow. >> Okay. And then I think you probably need to select it and change the thing over. But I was looking at it. I was like, wait, that's like not plugged in at all. It's hilarious. There we go. Works. >> Yeah. >> New studio, you know. >> Yeah. This is >> Oh, this is nice. >> Much nicer. Much >> silky smooth now. All right. >> So good. >> Yeah. >> Sounds awesome. >> Okay. >> Okay. So, yeah. reintroduce uh Alpha School for those who who don't know the model because uh I I I think uh it was many people learned about it through invest like the best I guess. I mean that's how I learned about it. >> Well, there were there was like there was like you know rumor mill around it right there was like there was a lot of there was a lot of lore. A lot of people were curious and then invest like the best I felt like was like this first mainstream kind of exploration of of the the vision and and what had been done to date and and how how you guys were approaching it. But obviously that was before you joined. >> Yeah. So, Alpha School actually started 12 years ago with Mackenzie Price here in Austin. And it was a smaller smaller micro school, but still built on that fundamental idea that you could compress the entire academic day to about 2 hours if you leaned on what we know about learning science, which, you know, the the the teacher in front of a room of 25 students trying to teach everyone at the same time is just not the best way to learn. Uh, and if you were able to kind of customize the education to each student based on where they were, you could get that day a lot shorter and then give them a lot more time for learning life skills, doing these interactive workshops, and making them just love going to school, right? Because no student feels like they're being held back or not getting a chance to get caught up. And so when generative AI came on the scene in 2022 or so, uh Joe Lamont, who he and Mackenzie had known each other forever and his daughters were already going to alpha school, he kind of went to her and said, "Hey, with AI, we can now take this model that you have figured out and we can start to we can bring this to a billion students. We can make this the way that like every student gets to learn." And that's when alpha that he brought um he brought big investment into alpha school started like building it up expanding it and that's why a lot of people feel like it just started in the last few years because that's when it really started to grow and take off and Joe had been kind of uh you know he wasn't very public facing figure for a long time. He wasn't doing podcasts, wasn't doing interviews, and then a year ago >> he would have loved the podcast circuit back in like the the '9s. He would have he would have crushed it. >> Yeah. Yeah. But then a year ago, he did that invest like the best interview. He did the Colossus piece that I know a lot of people have read and that's when he started really sharing more about what they've been up to at Alpha School because they'd had a few years of working on the new model, building out the software and now it's just I mean it's it's growing like crazy. We were at an info session in Boca Raton two weeks ago and they thought maybe 20 30 families were going to come and there were over a hundred and the the school that they're planning on starting there got filled up just almost immediately. Right. It's been very cool to see how much excitement there is, especially in the crowd of younger, more tech forward parents who are who have been thinking for a long time that like, hey, there's got to be kind of a better way to do this and it feels like we really have the technology now to make it happen. Is the is the model like start with kindergarten and then just grow with the class or or is there actually like let's get a bunch of juniors in high school to jump in so that there's like continuity from day one and it feels like a full K through 12 experience or is it like grow iteratively? >> We do both. So a lot of students do come in in we go down to prek4 right now. So, preK4, kindergarten, first grade, a lot of students who come in at that age range. Again, just because there are more younger parents who are excited about the way the school day works, but you do have a lot of kids who come in starting in high school. We've we have people I'm at the high school in Austin right now, and we have students who transferred in as 10th graders, 11th graders. It's kind of if they if if they're really interested in something or if they feel like their current school isn't working for them and they want to try this model instead will take them basically any time of year. And the benefit of starting earlier for a lot of students is that they don't have as much catching up to do. A lot of students transfer in from even really good private schools and it turns out that they're a year or two behind in math or reading or one of these subjects and so they do go through that catch-up period. But again, it goes back to just like the nice thing about having software that can personalize the learning experience to each student. If you're a seventh grader who's actually in fourth grade math, there's no seventh grade math class that will help you get caught up. But with with time back with our with our learning software, it can identify, oh, okay, you actually need to go back and really memorize your times tables because that's becoming a big bottleneck for you in trying to learn algebra. So, we're going to go back and do that first. Ryan, >> Ryan Cohen, if you guys could uh Ryan was being asked some math questions about the um eBay takeover this >> Oh, yeah. that the eBay program >> and there was some like adding and subtracting that they were doing that that um >> wasn't wasn't adding up. >> It wasn't things weren't adding up. >> Sign them up. Yeah. Extension program. >> Um talk about the founder school initiative specifically. How does that fit in? It feels like uh like a different track or how how separate is this from sort of the the typical alpha school experience? >> Yeah. So basically, you know, we we have our high school in Austin and we've been looking to expand the high school offering for Alpha. And so Joe went to all the high school students last fall and he said, "Hey, you know, what would be the just best most incredible version of high school that you can imagine and we have a really good mix of what the high schoolers are working on in their afternoon time after they finish their academics. We have a girl who's putting on a Broadway play. We have a few students writing books. One student opened the biggest bike park in Texas, but we have a lot of students who are really interested in entrepreneurship. And so they said, you know, it would be incredible if we had a highly dedicated entrepreneurship program so that we could just go after this as much as possible because we have all this time in the afternoons and we, you know, we just want more mentorship and more guidance and more support around trying to like build big businesses. And that got the idea going in his head to say, "Hey, what if we opened a high school where we were essentially saying that we are going to get the best mentor network possible. We're going to get the uh the best like resources possible for whatever bottlenecks, challenges you're hitting. We're going to build the best curriculum possible so that we can take what might be years and years of kind of fumbling around in the dark like a lot of people do when they first start that entrepreneurial journey and we try to teach all those core skills as quickly as possible. Give you the ability to develop expertise in whatever areas you're most interested in and then build a big business by the time you graduate high school. because nobody really doubts that a sophomore at Stanford can drop out and build a million or billion dollar business. And so like why can't a 17 or 18year-old and kind of like working back from that said, okay, well you get this incredible network at a place like Stanford of other students who can be your co-founders who you can work with and be motivated by. you get these incredible mentors who come in, talk to you, you know, give you some advice or just inspire you around what's possible. And then there is some stuff around the academics in the classes. But there's functionally no reason we can't pull that back down to the high school level, especially considering that these kids get like 5 hours in the afternoon every day after they finish their academics. So that's 4,000 plus hours over four years. And then if you just cut out all of the things that people usually end up wasting time on or don't have the resources to figure out, it is eminently doable to get somebody to a million dollars by the time they graduate high school. >> Yeah. at the at the very least like teaching teaching some of these fundamentals. I think this has been the thing that YC has always done so well which I think is amazing because it's all open source which is like okay like what are the what are the you know they can probably teach like the foundation of like a Y what what what the YC method to making a startup is in about an hour and it's like simple rules like talk to your customers that often times people don't do early in their founder journey where they're just like oh I have to spend like six months like making everything perfect before I can talk to a customer. And it's like, no, you actually can just go have the conversation first before you even start any of that. And so there's basic things where if you can get somebody to like bas start the clock earlier, start taking shots on goal, start going through the motions, but doing it with like some general guidance that you that that again like the you can get on the internet, but you have to kind of piece it together. And there's something about having having that in-person uh experience that I think is very important too. >> Yeah. Having somebody in the room with you to say, "Hey, you are wasting time on this thing or you need to go try to sell this to someone." That five minute conversation might save you months of your life. But a lot of people when they're starting out don't have someone to do that. And it's like just by providing more of those touch points and having that expertise in the room with you, I think you can really you can accelerate it quite a bit. And YC is a great example where they do an incredible amount of work with people in 3 or 4 months and we get four years. So I'm I think we're going to be the other ton of progress. >> Very cool. right now. John John and I uh were at dinner last night and one of John's friends uh is will likely get to like a million of ARR building a business that historically he would have always needed to raise venture capital for but given all the AI tools like he just doesn't need to raise any money and so he has all this one he has optionality he's not like raising money locking himself into this thing for a really long period of time but it's very cool because historically capital would have been a huge constraint for like a high school student where um you know even if they can go to to a private school they don't necessarily have like another you know a huge amount of money to invest into into some idea and now yeah >> they basically don't need any capital for a huge number of different types of businesses. >> No, it's totally true. It's when the like AI is the big thing that removes the bottleneck around like why can't a 17-year-old build a million-dollar business where you don't need to spend as long developing deep expertise in programming or software development because you know you can certainly for a V1 you can prompt a lot of it. you don't need to raise a ton of money for uh hiring or or design or buildout or any of these things because again you can get like an initial version going with AI and also if you're a 15year-old trying to get started and you're trying to hire a marketing person or whatnot like a content marketer might not want to go work for a 15-year-old like that'd be kind of an awkward thing right and so if you can just use AI to get started and get your business going yeah you take out a lot of the capital requirements take out a lot of the expertise build requirements and it it again it just shortens that timeline and makes a lot more things possible for them. >> How are you thinking about the or how are you wrestling with the idea that like so many great founders did not study entrepreneurship or business? They usually like when when when you give the other examples of like the person that's putting on a Broadway play. Like I I don't think that that's not an entrepreneurship path. I I I I I see that person and I'm like, "Oh, they could wind up owning like a media company or they could wind up buying and, you know, consolidating the Broadway industry." I don't know. There's like a bunch of different ways. And I and I when I look at entrepreneurship in America, I think it's one thing to look at like Silicon Valley tech startups and there's another to think about like the millions of small businesses that are like a dentist office or a law firm or some sort of business that's built around a unique expertise with a completely different path. And then entrepreneurship is just like the cherry on top at the last second where that individual realizes that they just want to build their own company around their their expertise or their discipline. And it feels like there's always this wrestling with like if you only learn the entrepreneurship piece like what industry are you actually an expert in to go and you know offer an improvement in. Uh people always go back to like Mark Zuckerberg I think he took some computer science class but he studied psychology and that makes a lot of sense. >> I just looked it up. So um Sundar has an MBA. Satya has an MBA. >> Those aren't the entrepreneurs though. >> No no I know I know but this is notable right like the professional CEOs. So Sundar, Satya, Andy and Tim Cook, they all have MBAs. Jensen, Elon, and Zuck do not have. Um yeah, that that being said though, I think I think um yeah, it's a good it's a good point which like clearly clearly you if you have a a vision and you work very very hard and you get some luck along the way, you don't need to be that good at business and you can like you just end up being successful. Yeah, you learn it on the fly. But at the same time, I think a lot of entrepreneurs, like people that want to be entrepreneurs will spend five years wasting a bunch of time kind of like learning these foundational like truths about entrepreneurship. And so I mean, when I think back, I would have loved to learn all the kind of like the YC way in high school, right? And just take some shots on goal, even if it didn't even if I ended up going to college. >> Yeah. >> Getting into to something totally different and then coming back to entrepreneurship later. >> Yeah. And and to be clear, you know, the the students aren't going to be like studying entrepreneurship and there's going to be this very heavy focus on expertise building in a domain that you are passionate about because that is going to lead to a lot of these opportunities and that's something that Alpha School already does really well. The high school students are heavily encouraged to or actually they're required to pick a topic that they really want to develop deep expertise in and then we're teaching them how to use AI, how to use writing, how to use producing, uh just like any kind of content media to develop that expertise and that's going to be a core part of this program as well. We also have this this whole additional aspect of it that's really important to us that we're calling kind of the philosopher builder Canon drawing on Cosmos Institute and working with them a bit where we don't just want people who are like slinging mobile apps to try to hit a revenue goal and then driving around in Ferraris, right? That's absolutely not the kind of like ethos that we're trying to inspire here. >> Yeah. Boo. I want to see the I want to see the alpha school parking lot in Austin just urus wallto-w wall. >> Yeah. Well, every Yeah. By the end they they all start an ecom business and then six months in there's they're ripping a course and they're like here's how I got a urus in at 16. >> This is not going to be a school on how to sling peptides or anything. >> Hustlers University. That's Andrew Jade school. Hustlers. Oh, I mean, we actually have a really rigorous freshman year reading list that's basically grounded in philosophy, uh, and kind of like trying to develop a broader sense of the world. And a lot of it goes back to one of the big challenges with being an entrepreneur, kind of doing anything in high school is getting out of the local high school mindset. Everybody wants to build an app or something for their peers, right? They're not thinking about the bigger problems in the world they could be going after. And so by spending a good amount of our time encouraging them to pick areas they want to become experts in and really deeply understand and encouraging them to think and read and write about like the bigger challenges uh in the world that they might want to attack then we can kind of encourage more of that thinking where it's not just again we're like we're not trying to build hustler school here. >> Uh and so like I'd completely agree that you couldn't like just teach entrepreneurship and expect that to lead to something interesting. It needs to be balanced for for it to work. >> In terms of the best way to learn, do you do you prefer the the the Stanford method where it's just two people, some microphones having a conversation, or do you like the Dwaresh Patel podcast way where there's an expert at a chalkboard sort of breaking things down line step by step because it seems like there's two sort of ways to learn these days. Yeah, >> John is doing a bit because people have been doing teaching classes at Stanford that are effectively podcasting Stanford. >> It's just it's just a hilarious inversion. >> But but I am but I am interested in like in like you know what is the world for just like hearing from an expert in an unstructured way more conversational versus like you know having information delivered in more structured pattern. Is it like, you know, right tool for the job, right person for the right flow, or is there is there one way that you think is like actually better? >> I think I think it really depends on what you're trying to provide. So if we were doing a workshop or a session on you know earned media or on Tik Tok or something I think that having very short informational sessions followed by uh implementation practice you know finding issues and then getting very structured feedback to you know the challenges that you're running into. That's going to help accelerate your learning the most on that like very tactical skill. But where the conversational longer form stuff becomes super valuable is hearing stories of how other people got started or what other people did where it's not so much like here's exactly how to do this thing because it'd be really hard to hold an hourong lecture on how to do Facebook ads in your head but hearing how somebody thought about developing expertise finding business opportunities getting started especially hearing those starting stories from people who are now really successful. Like that's something that I think those Stanford classes do so well is they make these incredibly successful people feel very human, very approachable where you realize like, oh, they were just like me at one point, just getting started. They didn't have all these resources and and those kinds of uh sessions are extremely valuable uh as well. So, I think it's just kind of like what problem you're trying to solve at that point in time. >> Jordan, anything else? Uh what any any predictions on end of this year? >> Was Joe inspired by Pama for die for the structure? Because it's a hundred Isn't it 150k a year? But if you don't hit a million dollars in profit, you get your money back. Isn't that it? >> That's correct. That is >> That's PMF or die right there. We got to lock these kids in a room. Stream it. This is the way. >> Yeah. >> No, I was gonna say Yeah. Yeah. predictions on on um highest run rate. >> Oh yeah. >> 12 months from now from a student, how how big do you think they can go? >> Because in today's world, like if they're not getting to nine figures in 12 months, like they should probably just jump down. >> Yeah. I mean I mean, you know, obviously you can't hold me to this. I But I would not be shocked if by the end of year one, we have at least stu at least one student who's, you know, well past the 100k mark. I I wouldn't even be horribly surprised if you've got a student who's like halfway to the million or even there, right? Just you find that right thing, you lock in and we're going to have some students coming in who are already making some progress on on their own things. >> Whenever we talked to YC founders, steel fellows are always like they were always making money in late middle school, high school, and it was they were picking ideas that didn't that had kind of a cap to it, right? or they were competitive market >> or super competitive, but if they just applied themselves to a to a category that was maybe a bit more boring than Minecraft stuff or >> Minecraft hacks and >> like there's there's no limit. So, >> but but I'll tell you what, guys, you know what would be a lot of fun is I think these students are going to crush it. And so, if we want to do like a TVPN founder school demo day next June, >> I think that'd be pretty sick. I I'm so excited to see what these kids do. >> Let's do it. Uh yeah, we we have to set the record for the youngest TVPN guest. We had that kid who uh applied to YC, right? Remember from Australia? He came in for a couple minutes. How old was he? >> Was he 16? >> I think he was like 13. >> I thought he was 13. Yeah, he's really >> Yeah. He came in with his dad. >> Yeah. >> Uh and then if any of them build products that we can use, let us know. >> We'll happy to be beta customers. >> Yeah. Yeah. I mean, I'm going to be shouting from the rooftops everything that they're working on. So, I'm I'm I'm very excited about that. >> Amazing. >> Well, congrats. >> I personally I don't I I don't think of myself as like a beta customer. I'm more of a Sigma customer, but you can do whatever. >> Okay. >> I thought you were going to >> If I'm going to buy a product from Alpha School, I want to I want to be the Sigma customer. Anyway, thank you so much for taking the time to come chat with us, you guys. We'll talk to you soon. >> Great to catch up. Congrat. We'll talk to you soon. >> Um there's some back and forth on the timeline. XAI's GPU fleet is running at about 11% utilization exposing how hard it is for AI labs to fully use expensive Nvidia hardware. Uh this is uh May 2nd from the information. Trace Cohen says, "How is this possible? There's infinite demand." Uh we're going to talk to Ay uh in a little bit about the infinite demand because it is real. But uh there's some extra context here. It says 11% almost certainly refers to flops utilization, not 89% of GPUs sitting idle. If if it were the latter, they would just throw more GPUs at training. It's notoriously hard to maximally utilize GPUs when even when saturating inference capacity due to things like uneven mixture of expert demand and and memory stalls. The 11% figure means their inference stack sucks with a large contribution from lackluster architecture. Contrast deepseek who do an amazing job maximizing utilization at every level including innovative load balancing. And I mean even if the utilization is low, it's like well they the the team seems ahead of the curve solving the problem by acquiring cursor or partnering with cursor because the the whole thesis of the cursor deal was uh XAI has compute capacity. But Tyler, what do you think about this? >> 11% does not mean like only 11% of the GPUs are like on. It's just that like >> you know the maximum like flops that you could get, say it's like a thousand, they only have like 110 like flops basically actually being used just because of like what their like inference stack is. Yeah. So, so the GPUs are are still on. It's not like they're just like sitting idle, >> but you want it at 100%, right? Or close to >> Yes. But that's like extremely hard because you have to like >> you have to be using it all the time because, you know, you're constantly loading, you know, memory on and off or whatever. >> And and and you might just see uneven demand. I mean, that happens all the time with these AI products where there's a boom. Like I bet you that uh utilization spikes during like big moments on X because a lot of people are tagging like at Grock, is this real? And a lot of people are are using Grock in X when there's a particular, you know, there's there's those moments where it's like an election or something and everyone's on X talking about the thing uh the rate on Osama bin Laden or something like that and everyone's like quering at the same time to get more information that's probably lighting the GPUs on fire during around around big model releases and whatnot. Uh anyway, Jordy, where do you want to go next? >> Our next guest in the in the waiting room. We can bring in Michael from CASA. >> Michael is in the waiting room. is the co-founder and CEO of CASA and he's here to talk about subscriptionbased home ownership platforms. How are you doing, Michael? Good to meet you. >> Good. Good to be with you guys. Thanks for having me. >> Great to have you on. Uh you announced a new round last week. Couldn't get you on on Friday, but you're here today. Hit us with the news. >> Uh we raised $20 million series A led by a bunch of great folks around the table as well. Um including a recent guest, Mr. Travis Kalanick himself. WA I didn't realize >> how did that happening? >> Uh well uh I give you a little bit of backstory. I uh I left school when I was 18 to join Uber. Uh Uber was uh 50 people at the time and um started out as an unpaid intern working on the supply side getting LA up and running. Um got a job offer later and decided to drop out and uh a year later moved up to SF to work on product and edge which was really my my passion. Did you did you know did you in your in all of your 18-year-old wisdom know that Uber was a special company prior to joining or were you just happy to have an internship? >> Uh maybe a little bit of both. I mean it was obviously something that spoke to me. Um but I think for me it was a way to actually just get immersed in the culture I really loved. But from afar I mean I know you guys are in LA. I'm from LA and uh certainly it wasn't what it even is now u with you know exposure to startups and tech and all that. So >> totally. >> Um, yeah, it was very fortunate. It actually happened on the backs of a tweet, which is a fun story for for maybe later, but >> the whole was a whole team built off of tweets. >> Basically, yeah. Uh, basically there was a Techrunch article that they were coming to LA. I followed it. A few months later, someone tweets from the account, says, "We have a press launch. Uh, email us here if we missed an invite." And I just hit up this email as a 18-year-old kid who was not in the press, but just wanted to meet these guys. and uh they sent me an invite and was like come eat our CEO Travis Kalanick and it was this restaurant in West Hollywood >> um where I was like half the age and height of anyone there and um smooed my way into an internship with uh the GM that they just hired for LA and that was kind of the start of it. >> That's amazing. Uh how long how long were you there? >> Was there for almost 6 years. I left the day that TK left. Um Wow. >> I knew that, you know, I didn't I didn't want to be there if he wasn't there. um I knew I was only going to work for him or myself after that. And actually I ended up working for him again a few months, you know, after he he left and all that madness. Uh he uh he was calling for for round two with Cloud Kitchens. And so I joined him there. We were about 30 at the time uh and was working on product and design for software side of our business until uh till 24. >> That's incredible. Quitting in solidarity with your with your boss. >> Underrated. >> Is underrated. >> Underrated. >> Underrated. Um, all right. Talk talk about talk about Kasa. >> Yeah. So, uh, not sure what you guys already saw, but Kasa is a personal property manager for your own single family home. Um, you know, I guess the origin story is coming off of Uber experience myself. I was really fortunate enough to consider buying my own first home. Uh, and I think, you know, like any first-time home buyer, you're super excited about the emotional aspects of having a home and making it your own and sharing with your friends and family. Um, it's also this huge financial endeavor, probably the most expensive thing you ever buy. And pretty much everyone realizes once they have it for the first time that, you know, at least a part of the ownership experience actually sucks. It's basically another part-time job. Um, and you know, the home is basically a thousand products in one. There's always stuff that's going wrong, always stuff you want to make better. And uh, you're really left to deal with it on your own. And right, most people don't have the time, the expertise, negotiating skills, sometimes even interest to deal with all this stuff. And they'd rather be spending that time with their loved ones and going on vacation and doing fun things. Um, >> uh, yeah. Yeah. make makes total sense. I've I've uh I there's so there's so many things around my house where I'm like months go by where I'm like I I should I should do that or you know whatever. I just I don't get around to it and then and then stuff comes up which is like oh do you have like where's the floor plan and I'm like I don't know. It's like in a image email. Yeah. What's paint color? When when did we redo the when did we change the HVAC filter? like all this stuff and it just gets lost in emails and text. So I I I totally get the pain point. U has there been a lot of shots on goal on this concept in the way that you're positioning it? Like it it's seems like one of those ideas that I'm surprised that I can't think of like a name brand company that does this. >> Yeah. >> So I'm sure you've looked for it. Um what do you >> Yeah, it's a good question. I mean, if you look at the companies, you know, at the intersection of home services and tech right now and the few that remain, it's like the Yelps, the thumbtacks of the world. And um, you know, they might market themselves as homeowner products. I don't actually believe them to be. And you don't have to look farther than their P&L to understand that, right? 100% of the money they make comes from the vendor side of the business. So, these guys are happy to send you shitty plumbers or super expensive plumbers um, if those guys are paying the platforms 20 bucks a click. And to be honest, the vendors hate them too for the same reason. So, I think this is prime for disruption. Um, you know, to answer your question, I would say I don't really think it's possible to build this in what I would say is the right way to build it. Um, until, you know, the last couple years with the multimodal models. Basically, how CASA works under the hood is you sign up and the first thing we do is we come to your house with a bunch of specialized hardware and software and in the span of a couple hours for a standardiz home, we will develop an extremely intricate understanding about everything about the home that matters from a servicing perspective. So it could be a full 3D model LAR scan of your house, uh, every appliance, every electronic, every exact paint color, the light bulbs inside of all your light fixtures. So to do that in the span of a couple hours with one or two humans was impossible a couple years ago and is now possible. And that's really what powers everything that we do. So we're actually really fortunate. We got to take this like very first principles look and approach to all these homeowner problems. that, you know, it starts with the physical day-to-day pain points, the things that a handyman might help with or, you know, another type of physical vendor, but there are also these other kind of meta concepts around the home. Anything from your property taxes to your utilities um to all these things that we uh that we now kind of monitor end to end and um we'll do a more interesting job of helping you out with over time. And I'm assuming you have the uh this is an account that theoretically should stay with the home because I remember when I bought my house, I felt terrible cuz I would hit up the seller like the previous owner all the time like, "Hey, do you know where this thing?" I just was like I it was fair to ask all these questions and they were happy to answer, but it was still like I felt like it was wildly annoying. And if there was just basically like a database that all this stuff had been like dumped in and and a single source of truth, they could have passed that to me, I would have been happy to to pay to have access to it. So theoretically, I'm sure you're building out your model for the business, but you're hoping that, you know, a house could change hands multiple times and the CASA would would be kind of like um continuous through the whole experience. >> Yep. Yeah, absolutely. I mean not to bring in a legacy name but you can kind of think about it like a Carfax for your home too or certainly that's the direction we would move in right so you have this neutral entity obviously we have no particular affiliation to any of the owners or whatever and we're able to just provide this very unbiased view at how has the house been maintained what are the general costs to maintain the home um and that might even become interesting in other contexts right like to lower your home insurance costs if we can provide the insurance entity with like a pardoned understanding of these are the proactive maintenance plans that have been happening for a home. Um there's no reason why you shouldn't be able to save money there. And so these are really all the opportunities when you're taking this homeowner approach um homeowner focused approach that you can kind of pursue that don't make sense for anyone else in the picture right now. >> Very cool. And have you been in stealth the last couple years or like Yeah. What's the history of the company? >> Yeah. Yeah. So we've been in a private beta since Q3 2024. Uh we basically put out this very very simple app. It was basically a two-way messaging client uh for a few of our friends or family in the bay. We basically said, "Hey, whatever problem you have with your home, send it to us uh and we'll try to figure it out." And for us, you know, we weren't charging anything. We just wanted to understand what were the bounds of requests that we were going to get. Where did people want to um have a concierge for their own home help with? And uh on the other end of that, that was basically my co-founder Michael and I just responding to these issues, you know, calling up vendors, figuring out handymen uh on the fly. And you know, I'd say in the span of a few months, we understood where the sweet spots were, what people were looking for. And the beta itself took on a life of its own. So what started as a handful of folks ended up with hundreds of homes um all in the bay and eventually in LA um on the uh on the backs actually most notably on the backs of a a tweet from one of our beta members Lenny Richitzky I know you guys know and uh yeah he he shared it uh you know he was like hey do you know want a few extra homes and we were like sure and then he tweeted about it. It was a very sweet and generous tweet and uh the weight list just went through the roof and honestly we've been working through that one for for many months. So, that was the first time a lot of folks had heard about us. Um, and uh, yeah, that's that's where we're at now. So, >> how what does it take to launch a new region? >> You probably know what >> you you've launched a few regions, I'm sure, back in your Uber days. >> Yes. Actually, the first thing I was doing at Uber was uh, standing around at LAX knocking on black car town cars, uh, recruiting drivers. And by the way, at the time it was like primarily 50 to 60 year old men who had not seen an iPhone before and were like, "Hey, do you want to, you know, change your whole life and use this phone and uh drive people around?" And uh yeah, I think for for us it's very similar in that, you know, I see this as a parallelizable playbook. So we uh you know, at Uber, we spent the first couple years really perfecting what was the playbook for landing in a city and going from zero to 100. Um and once that really was understood and stabilized, you saw us launching, you know, 50 100 150 cities a year. So, uh, you know, we'll do the same thing here, which is we're understanding what that looks like. LA is the second test of that. Um, we got a handful more markets coming this year. Um, but, uh, you know, I I think you'll see us kind of do that exponential growth curve on the number of markets we're supporting. Um, you know, all to the benefit of others. >> How do you think about uh disintermediation? Obviously, Uber did fantastically. even though you meet a great driver, no one really jumps from, you know, the driver that took you to the airport is now a full-time employee. Uh, very, very rare. Uh, at the same time, like dog walking had a much harder go where people found great dog walkers and said, "Hey, how about I pay you cash?" And we cut out the middleman and we disintermediate. uh it feels like there's some sort of interesting unlock here with the data around the house that is valuable and provides enough value that both parties would want to stay on the platform but walk me through the thesis of you know avoiding disintermediation at scale. Yeah, it's a great question, John. Um, actually, I think those are two perfect analogies, right? So, for Uber, it didn't make sense because a huge part of the value for Uber was that you can get in a car in two minutes. Even if you get a driver's number, forget about all the insurance and the other things you get, but it just doesn't make sense to call a guy and wait an hour for him to come get you, it doesn't even make sense for the driver. >> Um, the dog walker one's a bit different, right? Because the dog walker presumably lives in your city and you know, once you have that connection, you're good to go. >> Yeah. And it's like everyone the same the exact same pattern. >> Exactly. Exactly. exact same service and all of that. Um, we're very much in the Uber camp and I I tell you that honestly just from experience, which is that we've never had a single one of our vendors or handymen poached by a homeowner. >> And the reason for that is obvious to me from the inside and I think will become obvious to folks as as a service gets bigger and more prevalent. Um, which is that all the things we do to surround the visit, right? So in the Kasa app, you can book a handyman visit and you can say, "Hey, I want these three things done for my home." What actually happens at that point is a combination of a bunch of fun AI stuff and our own kind of concierge human team will look at those tasks. We'll look at the history of your home and all the information we have about your home to do that thing really well for you. Right? So you might say, "Hey, I want to hang up a bunch of uh family artwork in the hallway." And we'll come back to you and we'll say, "Hey, do you already have the frames? Do you want us to find them for you? If you do, we'll come back to you with a bunch of options based on your past preferences or we'll start to understand your preferences and we'll actually have those frames ready for you or in the hands of the handyman before they even show up. >> And all of that context about your home and the preparation for this work um is what really makes it valuable. And also getting the right person to your home, right? So, one handyman might be really great at carpentry and another might be great at a different skill. >> Sure. >> Sending the right person at the right time is super important. And also the fact that you can just get someone at your house usually within 24 to 48 hours at any time and you can't do that with one person. So >> So handyman exhibit handymen exhibit spiky intelligence and you solve that. That's what's going on. >> Yes. >> Yes. >> I love it. Uh well congratulations. I mean this seems >> where you you guys are based in SF. >> We're based in SF. Uh we launched with SF in LA. Uh so it's SF Bay Area plus LA. Bunch of other stuff to come and yeah. >> Oh and and 4Runner led the round. Fantastic. Uh well thank you so much for your time. I'm going to sign up. >> I'm definitely going to sign up this week. I'm going to sign up. Love it. >> Uh have a great rest of your day. We'll talk to you soon. >> Great stuff, dude. >> Thanks, John. >> Goodbye. >> Up next, we have Matty Hall from Living Carbon, uh with a absolutely massive plan to work on United States reforestation. Maddie is in the waiting room, but we'll bring her in to the TV show. Mattie, how are you doing? >> What's going on? >> Well, thank you all for having me. >> Thanks for hopping on on such a momentous occasion. Uh take us back though, introduce yourself and the company and then I want to get to this uh half a billion dollar deal. Absolutely insane scale. Congratulations already. >> I know it's very exciting. I think largest amount of money that's been raised for aforestation of degraded land in the US. >> Absolutely crazy. >> Uh prior to starting the company though, I was actually working at OpenAI. >> Oh no. Cool. >> Yeah. >> And explain OpenAI for for the audience. No, I'm kidding. Um very very cool. You may or may not have heard of it. >> Yes. Yes. >> Yeah. But I think it was pretty clear that the biggest blocker between where we are today and super intelligence is energy, right? Um and the math just continues to get to get more compelling. Like the hypers scale emissions have been up more than 50% since 2020. Um and at the same time like >> a lot of data center projects are being blocked like 64 billion something crazy. So, Living Carbon, our focus is on transforming uh old mine land and abandoned farmland into forests that either can produce carbon credits or sustainable forest products. And then we sell to the world's biggest companies, Microsoft, Google, Meta, Mckenzie to mitigate their increased emissions from AI and data centers. >> And then how vertically integrated is the is the business? because I can imagine this being something where like you're working with a whole bunch of contractors and different forestry organizations and it's a lot of financial work to you know bridge the gap between the hyperscalers which you have connections to and the different forestry organizations um but are you planning on like building machinery hiring people planting trees planting seeds like how how in the weeds will you get? Yeah. So, I think of living carpet as like the general contractor of the project, >> overseeing all of the different pieces. Um, that's largely what this octopus deal enables us to do. Um, it is them covering the cost of all of the uh site prep, the planting, the land lease, all of the above. Um, and then Living Carbon acts as the overseer and the manager of those projects, but then also shares in the revenue with octopus from their investment. So, we don't own nurseries. Uh we don't own land, but we have sort of programmatic tools to identify the sites that are best suited for our projects. Um and we do all of the endto-end negotiation, financing, uh offtake agreements, getting all of those pieces of the puzzle lined up. >> So, you're >> are there labor shortages and reforestation? Like, if somebody's wants to get in on the the AI boom, should they be like learning how to plant a lot of trees quickly? I mean, I think it's I think it's possible. I think, you know, what you have when you're doing these land-based projects is value that is going to persist, right, regardless of uh you know, how much of the software space the foundational model companies end up end up eating. Um, so, you know, it's a great long-term cash flowing business. >> Yeah. >> Uh, the goal is a quarter million acres in the short term. You've already done 25,000 acres. walk me through lessons learned from that, what that actually looked like, how that how that initial first project came together. >> Yeah, so our flagship project right now um is about 25,000 acres that we're in the process of planting um with our customers uh Microsoft, Google, all of the above. Um and that's taking degraded land in Appalachia, so uh abandoned mine land and reforcing it. And what's interesting about this project is it actually has the potential to offset uh the entirety of all of the emissions of San Francisco on an annual basis. So our plan with Octopus is to hopefully 10x that which would remove all of the emissions of New York City uh on an basis. >> Got it. Uh walk me through why mines cause so much deforestation. I think of a mine as like a hole in the ground and you know it it digs under but it's not exactly clearcutting like miles and miles or is it like what actually happens in the process of mining that affects the forest? >> Yeah. So a lot of these sites that that we're working on they actually have been like stagnant since the '9s and just really sitting for decades. It's not just the mining pit but all of the land um in close proximity to it. When coal went bankrupt in the 90s in the region, there was very little uh effort put into a lot of the remediation and restoration that uh would actually allow for that land to be put back into production. So, these sites have been really liabilities on the balance sheet of a lot of the mining companies and private land owners for 20 plus years. >> Yeah. Uh can you get me up to speed on some of the other like reforestation tradeoffs that are happening right now around carbon credits? Like I've heard one thing about like uh some carbon credits that are basically just designed where it's like okay we were going to cut these trees down but we didn't and so we want the credit and that feels like little bit edgy. And then also I'm wondering about international because I imagine just on a cost basis you could probably plant way more trees internationally because the land's cheaper but does that offset carbon emissions like globally potentially but do the hyperscalers get credit for it if it's happening halfway across the world? So how have you unpacked some of like the hot topics in reforestation? Yeah. So, I think our focus has really been on like an active intervention that wouldn't have otherwise happened without uh funding from carbon credits and the complex that we have there. It's hard to say with certainty whether or not a tree would have would or would not have been cut down, but we can say with certainty that these sites would not have been restored without our work. >> Um, and so you're looking at these these areas and the natural rate of regeneration is very low. um it is cheaper to do it in other parts of of the world and I think there are some amazing projects that are being developed um in Latin America and and globally. I think for us being in close proximity to the areas where data centers are being developed and having that regional focus such that our projects uh you know will offset carbon within the same region where uh large scale renewable projects are being built out and data centers being built out and that's been desirable. >> Yeah. So it's not just abstract to the local voter who's making a decision around environmental impact. Uh since >> Yeah. Is one of the solutions to how ugly data centers are to some people to just put a like forest fully surrounding it? Is that uh could we see that in the future? >> Oh, I I sure hope so. >> That Yeah. Yeah. >> A magical data center in the center of a forest. I would love that. >> I would love that, too. >> I mean, that's what we're doing. So, >> very cool. >> That's great. Awesome. Uh well uh tell us about the the the the new deal. I want to hit the gong. How big is the new deal? >> Up to 500 million. >> Congratulations and thank you so much for taking the time to come chat with us and we will talk to you soon. >> Very happy you're doing what you're doing. >> Have a good rest of your day. >> Sounds good. You guys, too. Great. >> Goodbye. >> Up next we have Azny. Uh is this the first time on the show? No, he's been on the show before. second time on the show. Well, let's bring him in from the waiting room >> as a free man. >> First time with his new fund AMP PBC. He's here in the waiting room. He's here with us in the TV panel. >> And how's Generational Run? >> Yes. Generational getting started, guys. No, generational run is used for people who are retiring. >> Oh, yeah. >> That's true. That's true. I think it's I think it's appropriate to re I think it's important to recognize when you're on a generational run so that you realize you have to actually level up even further if you want to stay on the same trajectory, right? >> Yeah. Yeah. >> Uh cuz if you just get complacent then >> Yeah. It's important to like count the chickens before they hatch. Like that's what you're saying? >> No, no, no. The o the opposite. Anyways, great great to have you back on the show. Um >> uh it's been fun to um but yeah, catch us up to speed on >> Yeah. Yeah. When did the fund launch? uh what you know what's the strategy uh and uh yeah like like like walk us through like the current thesis for how you want to actually develop the firm. >> No, I think we should talk about something more interesting. Let's talk about eBay. Let's talk about eBay. >> Okay. >> Yeah. What you got? >> So, so here's what I'm think and I want you guys to kind of spar with me, right? I looked at like Ryan did this CNBC interview and everybody's like pinging me and saying, "Oh my god, this makes no sense." Blah blah. And and so here's my my take on it, okay? which is that like if you read the GameStock deck like carefully for eBay, >> most of what's been said about the deal in the last 48 days is basically totally wrong. >> I I read I was just before jumping on I was reading Michael Bur's piece on it which you guys should check out. Yeah. Um and he's right that the leverage is pretty tight but I think he's answering the wrong question and so is Ryan on CNBC where he's you know they keep asking him like where's the cap? How are you going to fund it? 50% cash, 50% stock, 50%. >> I I don't think the question the question isn't can GameStop afford eBay. >> The question is whether the underlying business actually works. And I think it does, but not for the reason I was expecting Ryan to pitch. >> Okay. >> So, if you if you pull up eBay's 10K from February, >> Yeah. >> fiscal 25, and I did not understand this until I read it, eBay spent $2.4 billion on marketing. >> Mhm. How many new users did they get for that? I mean, you guys are marketers, so you understand. 1 million. >> Whoa. That's >> crazy. The user base went from 134 million users to 135 million users after spending 2.4 billion on marketing. >> So, and that's basically they're basic you you would have to imagine they're just having to reacquire all their old users, people that have been on the platform before or maybe even lost their account and they're they're they're coming back. I don't know. But it's Look, that's $2,400 of marketing per new user on a site that every American already knows exists. >> Sure. >> Yeah. >> So, where's all that money going? >> Right. Rough. >> So, I don't think Coen is is I I don't think he's buying eBay. >> Like, just watching Ryan, I don't think he's buying eBay because he thinks he's smarter than eBay's product team. I think he's buying eBay because he can see $2 billion of fat that Wall Street has been pricing as fixed cost. >> And so, he goes, "Okay, let me cut that." and the interest on the debt just pays for itself. >> Interesting. But he doesn't necessarily want to say that because he could kind of >> give that idea to the bank. >> He still has to put the money together. Right. But is your thesis that like the deal is coming together. He has investors that he's talking to, but it's too early to say, "Oh, yeah. I actually do have a fund that's going to give me another five over here. I got seven over here, and it will math out, but just give me a week." Or is there something else going on? >> It depends on which investors he's talking to. If he was pitching me, here's what I would under, right? I'd say, okay, that's the floor. The floor is Ryan's going to cut $2 billion from this thing of that, >> put that into treasuries, and we're going to make more money than it's currently yielding. Okay, so that's the floor. Now, the ceiling, because I'm a I'm a technology investor, right? >> Yeah. The opportunity bull case. So, Amazon's used and collect collectibles business has been flat >> for six years. >> They tried renewed, they tried collectibles, they tried trade in, none of it failed. Mhm. >> You cannot put a 1962 Mickey Mantel card through the same warehouse as a phone charger. >> That category, that category, right, collectibles is structurally defensible against Amazon. Okay. >> Amazon is for phone chargers. Mickey Mantle for eBay, right? >> Yeah. >> eBay has the marketplace. GameStock has GameStop has like 1,600 stores that could physically verify >> sure >> the goods. That's a real mo. >> Yeah. Yeah. >> And it's worth more in the AI era than the human era, right? Why? Because my agent >> I I collect I love collect like rare pens. Okay, this is a >> M blah. >> Sure. >> As I get older, I love it. >> I'm a pen guy and and I love old vintage glasses. I don't know if you can see my Jack Marimage box in the back, but >> Oh, cool. Nice. >> But when I'm out of time, >> what do I do? I have Claude go look for rare pens and glasses for me online. The biggest problem with used rare asset purchases is fraud. Yeah. So I often will tell so claw will be like an I found this amazing pen this mlab pen and I say okay can you triple check that it's it's it's real it's not fake >> and that's where things go up real because there's no way for him to verify that without messaging the agent and so on. >> Yeah. Yeah. >> But if you have 1,600 stores where people who have Mumblan pens can go and physically verify those assets at GameStop. >> Sure. Now Claude just says, "Yeah, I checked it has the physical verification stamp from somebody brought it in." >> Exactly. So you can't you need physical verification built into the system for Agentic Commerce. >> And look, the reason I know this is >> real is because a few years ago, I think you you guys we talked about this last time, but I sold my last startup to a company called Discord in 2020, peak pandemic. >> So I come on as the head of platform. My my my job is supposed to be, you know, an go build SDKs, APIs, and so on for gaming. uh we helped this company called midjourney get going AI you know generate >> but 12 months later suddenly I find myself running without realizing an e-commerce business because it was the summer of NFTTS >> and the the the board aes are blowing up and suddenly we have more than 10 billion dollars of GMV flowing through Discord buy channels >> and Jason and Stan are like hey brother your job is to like capture a piece of that pie >> I go okay homework ac accepted so we start doing a deep dive and we realize ultimately the what what these users need and pay for like so you have liquidity right ultimately that's what a marketplace like eBay and Discord um provide in sort of community commerce is liquidity but you cannot provide liquidity if you don't have physical verification and for Discord that was just out of scope you know use sneakers >> well and it worked you're saying it worked for NFTs because you had the onchain you don't need any physical there's nothing physical to verify owned it and you can transact yep >> exactly and So, we had we were bootstrapping the e-commerce platform at Discord with NFTs, but of course, everyone on the board is like, "Well, how long are NFTs going to last? It's a fad." So, an what else is coming? We go look at rare sneakers, rare keyboards, like pens, all this stuff that nerds like me love. But for those, you need physical verification. And once we realize physical verification was out of scope, we nix the problem. >> Yeah. You're not going to go have the 1400 like retail locations where people can drop things off. Yeah. That's that's that's quite that's quite interesting. So that's when I realized, okay, eBay is this undervalued asset, and I hope that Ryan has figured it out as well, because if he hasn't, he's giving me ideas. >> Yeah. Yeah. Have you have you tried to walk through what what uh you know, given that you're probably more AGI pill than than I would say 90 90% of of VCs, have you tried to play out what is it? What how would you how would you build eBay from the ground up today with an agent first approach? Is that even the right question to ask? >> Look, I I have not guys because right now I'm in a I'm a comput infrastructure guy, right? We started AMP as this public benefit corporation whose job is to be an independent system operator of the compute grid. We think about we think we're roughly in like 1885 industrial England where the steam engine's been invented. Everybody knows that you can make cool new products like, you know, steel and and notebooks and pens and cars. And there's this very scarce input called coal that everybody is hoarding. In this case, it's compute. >> And when I if you fly over industrial era England, you'll see all these factories getting set up and everyone's running a generator in their backyard at half capacity. I'm going this makes no sense. If I'm looking at all my portfolio companies, you know, these these clusters are running at like half utilization. In fact, Elon's like got 500,000 GBP300s in Memphis at running at 11% MFU and less than 60% node allocation. I mean, this is 12 billion dollars of comput being wasted. So, I set up AMP as an AI infrastructure organization where we buy a bunch of compute, we do long-term leases, we pull that all those clusters on the grid. We coordinate capacity, drive up utilization, and by the end of this year, I think we'll have, you know, several billion dollars of compute coming online. But that's what I've been focused on night and day since like I left Andre Horowitz in January. And so, no, I have not had time to look at how to redo eBay, but if Ryan called, I'd probably help him out. But right now, it's wartime on compute, guys. >> I I want to talk about uh AMP, but I also want to talk about I just last question on the the the the combination of eBay and GameStop. Like, I get the thesis, the bullcase. Uh GameStop is $10 billion. eBay is like 48 billion right now. You put them together, maybe you get to 100 billion. I'm in for the bullcase. The question is like how what's going on with like the plan because it feels like Ryan just doesn't have the capital but then he announced it like what like what what what do you think is happening behind the scenes because there's one thing where you could throw it out as like oh like these two companies should work together here's a bullc case let me know if you want to be on the team that does this and then there's the other one which is like make the offer before the capital's lined up but I just haven't been through enough of these stories to actually understand like why it's playing out this entire way. >> I think to be honest, I think he's sitting there with I think he has $9 billion of cash. He's in a $10 billion company. I think when he announced this, I think he expected the stock to pop like crazy and he'd be going on CNBC being like this merger >> could make sense. >> Sure. >> And I think that >> we'll issue another 20 billion of equity and then and then we'll merge or something. And I think if you looked at I think if you looked at like how kind of frothy some things in the market could be, you could have imagined that playing out. I mean the allirds thing was I'm sure you appreciated appreciated that from a meme >> direct competitor to you. >> Yeah. Direct competitor. >> You got to be careful. >> AMP AMP versus Allirds will be the new horse race. Um, anyways, that that was my that was that was my read because because they're basic GameStop is basically valued at like >> the brand is and the and all the retail locations and everything is valued at like a billion dollars, right? He's getting no credit for all the cash. >> Yeah. >> So, to be clear, AMP is not a cloud business. We are So, I started AMP as a holdings business. Yeah. And I've got an infrastructure business and a capital business. And the infrastructure business secures compute and passes on at cost to our portfolio companies. We have more than $1.3 billion in commits for our first fund. I've been at it 8 weeks. Very interesting. >> And so we do do venture capital investments. We put $300 million into entropic. >> Yeah. >> Oh, okay. Cool. >> But we we we need to raise another roughly, you know, $6 half billion this year. And a lot and more is getting committed by the day. Yeah. >> Um but we give away the compute at cost to the independent ecosystem because my belief is that you know that the the like sort of the the optimal unit of research today is a like a focused talent scheme outside of the hyperscalers. You know anthropic and coding which was I was the first one of the first I'm certainly I think I'm the first angel investor if not the first uh investor in the round. >> They're saying you're the Jason Calacanis of of Anthropic. Unfortunately, JCAL I could never top JCAL, but like if JAL's intern or something, fine, I'll take the win, you know. But uh but I I think more importantly like I think compute is the strategic asset which I've been yelling about for four years and it's a primary bottleneck on these teams and if you're not at the hyperscalers, you just can't get access. So we buy up that compute, we give it at cost to the portfolio teams and then we reinvest the profits of carry and fees to buy more compute and so on and so forth. And so I'll take as much capacity as I can get from Alberts. I love it when new people go into the business because that gives us more supply. So, if you're the Albert CEO listening to this, please send us your compute. We'll take it all. >> That makes that makes a lot of sense. Uh, what are you excited to invest in? You're investing in teams that need a lot of compute. You're trying to find things that aren't going to get steamrolled by >> Anthropic, who's another big portfolio company. Uh, there's there's, you know, >> there's stuff other different things that need compute. But what what do you think? >> Because I feel like I feel like a lot of a lot of VCs are would never say this out loud, but a lot of I get the sense from a lot of VCs that they're kind of like paralyzed where they're like they really don't they don't have a clear sort of understanding of where things will be in 5 years and they feel like they need to be active. Um, and so it's a mix of like doing new neolab uh doing some neo neolabs, maybe doing some application layer stuff and just kind of praying. But I would hope that you have a given given your background and how you're approaching this, you have like >> a stronger thesis on on where the opportunities to invest at the early stages. >> Look, in in some sense, it's it's back to the future. I started my career at Kleiner Perkins when I was 19. My first board seat was as an observer with John Door when I was 20. I wasn't old enough to drive technic oh drink. Sorry. I didn't have a I didn't have a driver's license. I wasn't old enough to drink and and I got the chance to apprentice with like the greats like you know Brooke buyers and and look that's the that's the vintage of venture capitalist I grew up admiring like Arthur Rock and that's my you know our thesis at AMP on the on the venture capital side. Our business is called the AMP foundry where we helped create co-design you know new labs one at a time. My current one is called periodic labs and we just decided to lead their series. I led the seed round last year with Liam who was the co-creator of Chhat GPT and Doge who was the um who led some of the quantum physics teams at deep mind and we're trying to find new high temperature superconductors there with physical we have a 30,000 foot facility in Menllo Park I spend 3 days a week there we do a stand up every morning from 8 a.m. to 8:30 a.m. And then we make our priorities and then go execute. And you know basically we have AIs predict new materials. We then have robots synthesize the new materials. We then have an X-ray defraction machine that tests whether the material has the properties that robots the AI said. And then we pipe that verification loop back into the training run like as many times as we need for the agents to continue predicting new superconductors. And in the last 90 days we've had more material verifications than I think in the last decade in the field. And so I'm a huge believer in unblocking frontier progress in domains where it the verification uh sort of loop is clearly just like we know it's going to work but execution is the bottleneck. And then I like getting these the best teams, the best scientists, the best engineers, the capital, the compute, the commercial help they need. Now I I think that the beauty about having Enthropic around is that it's made this idea of the bidder lesson and and scaling legible to the capital market. So now instead of me having to call up 22 friends, I'm getting 21 nos, which was the case with the seed round of Enthropic. Now I I make two calls instead we get like you know like we get three times over subscribed so capital is no longer the bottleneck which is phenomenal you know again been at it eight weeks have more than a billion dollars committed for my our first fund I'm a solo Eman GP on the fund and there's lots of institutions pension funds sovereign funds who are like an how much more can you put to work especially in publiclix privates buyouts it's a bonanza for people who want to be true partners who want to be the arthur rock of this era I think if you believe in the bitter lesson it's not new it's been around for ages you and I the three of us talked about this like over a year ago at the last Horwood's AGM. >> Still bitter. >> I I I'm I'm more zen than I am bitter. >> Uh how are you thinking about building out the team on both sides? >> Uh trust is the mode. So there's five of us on the team. My full-time engineering co-founder was Sebastian Bobo. We were roommates 14 years ago at Stanford and then he went on to build a grid internally >> success. >> Here we go. >> What was that one? >> Overnight success. >> Overnight success. Thank you. 12 year overnight success in California. >> Exactly. Um so so Seb and Mihi built the Borg export GQM scheduler which kept the Google internal capacity pool at more than 95% utilization. There if it was at 94% utilization that was considered a major outage globally. >> Wow. >> Um Andrew Erskin is my uh partner on the operations side. He was a a partner at Oric which was the you know outside council for Anthropic. >> Um and he was my GC at Ubiquity 6 which was the company I sold to Discord. And then Rosie who's my chief of staff ran coms for me from Edelman when I was at Andre Horowitz you know when I was a GP there. Um, and so >> you got the band back together. >> It's it's it's it's the reunion basically. And you know, we put we called AMP uh not after my initials. Sometimes people think that it's not an partners anything like that. It's about energy. It's about a you know, a unit of of energy. And we we want to amp things up because we think we're entering like the great renaissance in technology. And you know if you can have a small team that trusts each other across context you know compute capital sports teams buyouts leverage technology all of this stuff it's these are all buckets and categories that we've all put you know traditional capital allocators are put around these asset classes that shouldn't exist and I think if you have the flexibility to go back to first principles with a small team that you trust you can execute you know with with orders of magnitude less less size of a team as a firm in this new era with the right tools. I don't know if if that makes sense. No, totally. You you mentioned uh taking positions in public companies. Uh the fund structured as a PBC. Are you also an RAIA? Like how are you thinking about navigating both of those asset classes since that's a little >> We are in process. Okay. >> Yes, we are in process of get becoming an RA because you know we founded the firm barely >> Yeah. >> Yeah. 90 days ago. But I'm used to that cadence because And recent Harwitz was a RAIA. I was a general partner in the AI infrastructure fund for several years as you guys know and we were an RA. I'm used to the compliance, the regulatory sort of guardrails we got to follow. And I think LPs trust us to have that cadence from day one. And so we're going to make sure that we, you know, Zach, if you remember this this like, you know, 12 years ago, Zach went on TV to say move fast and break things. And then you have to update the thing to be like move fast with stable infrastructure. And I think we move fast from st with stable infrastructure from day one essentially because we are an AI infrastructure team. >> Yeah. Talk about the PBC. Uh like if I'm playing back like what year were you referencing? 1850 or something like that? >> 1885. >> 1885. So if I go back to 1885 and I think about the financeers that uh you know created the industrial buildout uh they were not public benefit corporations. They were they were personal benefit. >> Yeah. Uh and I mean there was a lot of good that was created. We got railroads and trains and you know machinery and cars and all sorts of things out of the industrial revolution. Uh there were also things that were rough and there was unionization and battles and back and forth like what is the PBC in service of solving? What what why PBC? Yeah, great question. So there's the um I'll tell you the substantive answer and then the vibes answer, right? Sure. So um from a substantive perspective, we do two things, right? We have a venture capital business and we have an infrastructure business. Both things have this very unique property called positive externalities. When implemented correctly, venture capital can unlock massive positive externalities for the ecosystem and for the world because you end up funding innovation when done correctly. And then infrastructure the same, right? When you have compute that's used by small focused talent dense teams like Enthropic that's able to produce 10 times more soda capabilities than like deep mind which is you know 60,000 or 160,000 people then you're generating positive externalities for the world by being much more efficient per unit of input with the output they create and so I was like huh well what happens when as an economist you look at positive externalities usually you have market failure you have underconumption of that good well how do you correct the market failure usually you get the government to intervene but if you don't have the government intervening in time what do you do you become a private sector sector participant. And then if you look at the arc of 1885, you know, private sector businesses that ended up correcting market failure by um by producing public benefits, they ended up getting regulated as utilities. That's what AMP is. AMP is a self-regulated utility of that provides venture capital and infrastructure to the world's leading scientific teams. The next Dario, the next, you know, uh Giam at Mistral who created Llama, the next Robin who created stable diffusion. These are the my generation's smartest minds. I'm not smart enough to be, you know, them, but I can be their intern. And instead of waiting for the rest of the space to come up with standards and institutions to enforce this, we're like, dude, let's just do it ourselves and show the world you can have fun while doing it with a small team. You don't need to be, you know, some something called a um, you know, like these words like RAIA, multi-stage asset class firm. Doesn't matter. just let's skip ahead to the part the good part and and like you know use all the proceeds that we get from management fees and carry to keep the space like innovating at the pace that we were promised you know 12 13 years ago and instead we got tweets and not flying cars you know when I was at Stanford I got the chance as an undergrad I had the chance to take Peter Teal's class the 0 to1 and he was you know his whole moniker was we wanted flying cars and we got uh you know um uh uh >> 140 characters >> well 240 characters thank you and now I'm back at Stanford teaching CS150 53 which is the largest class on campus. It's called AI Coachella. We've got thousands of people following along >> Coachella and >> it's called Frontier Systems cuz it's all possible now. We're literally I in our lifetime we're going to have flying cars. We're going to have room temperature superconductors. We are going to solve cancer. We just want to do it in a way that's stable, predictable. We want to skip all the boom and bust cycles. And the way to do that is to lead by example and say, "Hey guys, the public benefit, you know, is to provide goods and services that are utilities and make sure that we we don't like let's be the adults in the room and not do the stuff where we tried to be robber bars and monopolists and got greedy along the way." And so that's the substantive answer. The vibes answer is look, I don't want to get sued by shareholders for whom it's not legible why I'm giving away billions of dollars of compute at cost to portfolio companies, right? Because that's what we're doing. And that's shareholder that's you could argue that's shareholder value that we're destroying. I would argue in the long term we're actually creating orders of magnitude more value. And if you look at Ben & Jerry's, you look at REI, they've become stable, endoring businesses in categories that are fairly crowded. And eventually, I do think technology and AI will get crowded. It will get commoditized because technology is never the mode. Trust is, community is a mode, culture is a mode, execution is a mode. And so we're trying to skip ahead to that part, but it takes time for people to get aligned. So until then, we >> will see will we see an AMP joint venture with private equity to uh help distribute diffusion >> if we can. Yeah, you know, if you go to our website, it's called amppub.com because I do think if you look back to the like vulture era of private equity, remember like RJR Nabiscoco and what Barbarians at the gate, we should just learn from from their mistakes and go, can we do private equity but done right in an aligned way? Let's not rush to like lay off, you know, hundreds of thousands of people and then not re-educate them and prepare them for their new opportunities. I I'm an optimist, as you guys know. I came from Andre and Horwitz. So I I I'm a sort of a rational optimist. I believe the transition can be done in a positive way. >> Um >> is that me getting kicked off? There's like a bell ringing, but that might be here. >> No, >> no, I don't know. >> Okay, cool. >> The only bell we have is this. >> But no, we're good. Continue. Continue. >> Everything we Everything we do is governed by a public benefit charter. So if we do private equity, it'll be governed in the public benefit. If we do education stuff, that'll be in the public benefit. >> Yeah. >> Look, I've made more money than I know what to do in life. I'm 34 and I'm just getting started. So, my goal is I'll be remembered for having been a net positive influence in the space. I just got tired of telling people I told you so. Because after a while, they start looking at my returns and going, "Why didn't you give me a call?" And I said, "I did. Look at your email. I introduced to you to Enthropic in the series A and the series D and the series C." And so, at some point, I was like, you know what? I'm just going to go direct, talk to the LPs, set up a platform, build infrastructure, >> and uh hopefully be known as a generally like sane, common sense, rational point of view on stuff that can often be is not legible to people from different parts of the stack. And that's what the class is about. So cs153.stanford.edu, I would recommend anybody watching go check it out. Um the the lectures are all online and the first one went on on Stanford's official page on Thursday. >> Was that with Scott Nolan? So Scott, no actually uh Scott is lecture eight. I put up lecture, we put up lecture one, which was mine as a kind of the opening act because you know Scott is one of the mainliners, the headliners of AI Coachella. >> Scott's will be up um soon as well and then Jensen was last week so I think he followed Scott. >> Well, last question for me. Do you think the world is prepared for it not to be a bubble? >> That's a good question. the world if the world prepared not for it to be a bubble. Oh yeah. Yeah. Inertia is guys, inertia is a powerful thing. Most of the world still has no idea what AI is. It is crazy. And I've been flying to places where I thought there would be diffusion of AI by now and they just are barely using quad chat GPT. I mean these things are still alien to most of the world. And so if if if we stopped capabilities today and half of us in the AI ecosystem vanished off the planet, nothing would change. It's still so early. >> Yeah. Yeah, >> it is really >> very cool. Well, great to catch up. Congratulations on a very impressive fund raise and a very unique approach and uh looking forward to the next conversation. >> Thanks guys on a generational run and I hope the acquisition does nothing but give you guys more steroids and more fuel for the fire. We need more of you every day. >> Fantastic. We'll talk soon. Congrats. Have a good one. >> Thanks guys. Bye. Cheers. >> Up next, we have Ben Lamb from Colossal Bios Sciences. I made a YouTube video about Colossal years ago. I think this is the first time I'm talking to Ben uh actually live, but uh fantastic company. Super interesting. Ben, great to meet you. How are you doing? >> Hey, great. Wait, what what YouTube video was it? >> Uh something about the Woolly Mammoth coming back right after your announcement. you went viral for the first time and I sort of like uh made a video essay talking about the company a little bit about your background and then also just some of George Church's work trying to give more context on the science like uh the more incremental steps like I think the media generally wanted to jump to conclusions about like bringing back >> year that first year was a lot of uh interesting fielded calls for sure. >> Yeah. Yeah. And I think people were sort of like, "Oh, the, you know, the Jurassic Park analogy is so fun. They want to jump straight to like the T-Rex running loose in Time Square or something." And I was like, "No, like there's a world where this business works. Even if it's just like some cool new animals for zoos, like you don't even have to get that crazy just looking at the price of like different zoo animals and stuff. I know that the the obviously the mission's a lot broader. We talked about some of the uh not deforestation, but uh dealing with the uh Siberia. I'm blanking on this but I mean you can tell me about all the carbon modeling for sure. Yeah, we we you know we had our original thesis was you know synthetic this is like the hardest synthetic biology challenge and you know kind of access to compute AI and synthetic biology all paired together will create kind of this unique opportunity to you know build a lot of different tools and technologies. And so our view is like if we start with the hardest problem and we couple it with kind of like an existential problem which is you know losing biodiversity uh you know I think it forces us to to uh you know build a platform that's pretty robust because working with DNA that's a million years old or 73,000 years old is a lot harder than just working with something like right out of a lab or you know bought off the shelf from some like XYZ DNA provider. >> Yeah. Yeah. Uh so where is the company today? like what what what are the I mean it was this was always like a walk crawl run project. There was going to be iterative development. It wasn't going to jump straight to bringing back the oldest creatures in history but uh where have the successes been? Where have the setbacks been? Where has the the strategy pivoted? Take me through some of the recent uh developments. >> Yeah, so the biggest developments recently are you know last year we had kind of a couple like watershed moments. We showed the world the objectively cute woolly mice y which at the time were the most genetically modified multisellular organisms uh out there right where we took uh the uh mouse equivalent of the mammoth genes that we're targeting in our Asian elephant cells made woolly mice they went crazy viral and I do remember sitting at Southby being very concerned because I was like wow people lose their mind over these mice wait till they wait till they see giant wolves in a month. Um, and then we showed the world that the direwolves where we took about 73,000 year old skull and a 12,000-y old tooth and made puppies. Uh, and so those were kind of two really interesting data points that shows kind of the end toend pipeline. Not only could we identify extinct variants and we could replicate them, do the ancestral state reconstruction, model them in and actually put them successfully into living cells and then go through the entire quality control process to deliver actual living animals to Earth, but do it in a way that's completely humane that passes all of our Aayakook and and Humane Global certifications and do it, you know, get exactly what was predicted. And and one of the biggest things that we've learned from that is all of the uh uh AI systems coming online are just accelerating. Like you're seeing a lot of different markets and industries be affected in a lot of different ways and you know a lot of doom and gloom is being sold around AI but synthetic biology and all of the modeling and taking the data sets that that all of our teams are generating uh and pushing them all forward together is something that you know we're really seeing a mass acceleration here. What's on your short lists for animals to work on? I feel like it's like, you know, there there's a big long list and I imagine the list in your head is a little bit longer than what's on the website. And so I want to I want to dig into some of the deep cuts, the bides. >> Yeah. So So we've definitely announced Woolly Mammoth, Tasmanian Tiger, Dodo, MOA, and then as of last week, Blue Buck, but even with that last week, you know, we we didn't say, you know, oh, we're starting the Blue Buck, we're going to be on this, you know, 10-year journey. We're pretty far. we'd already done all the uh uh ancient DNA work, all the comparative genomics work, all the stem cell work, all the animal repro work of cloning in analopes. And so now we're really just in the editing phase. And you know, two years ago, we were doing three to five edits at a time. We're now doing over 200 edits at a time. So that scale function uh has has gone really quickly for us, and we haven't seen an upper end of the of the delivery. So, I think that um I think it's highly likely you could see a couple more direwolf like moments of of species that haven't been announced but you know just you know show up and I think we'll show you some additional project uh progress on our some of our big projects >> and then uh what's the team like? I imagine that you have a lot of scientists on staff at this point. Uh what's the shape of the business? >> Yeah, so we've got 260 uh full-time scientists. We have uh 17 academic partners around the world. uh 80 posttos in academia, 75 global conservation partners and now we have five government partners around the world. So it it's you know we've closed a little under $650 million and the teams are going quite well uh labs in Boston, Dallas and >> Amazing. Uh and then and then in terms of like commercialization is is it still uh like you know just focus on the science like the business opportunities will come? Is it experimental little test projects? Like how or are you already starting to think about like you know the the SpaceX Starlink like there is a real commercial enterprise that will be self-perpetuating not science funding. I'm sure you're bringing in revenue from a bunch of different sources, but uh have you >> my head goes to like zoos and and effectively living living museums where I think about all the things that you listed off I read about with my four-year-old all the time because these, you know, children's books are not like, you know, they don't discriminate between living or or dead species, right? I've read I I know too many dinosaurs uh at this point. Um, but if you were telling me like, okay, we have a zoo and you can go see all these animals throughout history, that would be pretty compelling. And you would have effectively a monopoly over over said animals. >> Sounds sounds kind of bad to say monopoly over over a species, but >> does sound a little bad and evil, but the, you know, ultimately, you know, we love zoos. We work with a lot of zoos. We're not anti. A lot of times I get the anti-zoo stamp on me because I say, you know, zoos are a little bit transactional. If you do look at all the data and the studies that the scientific studies have shown up around kids seeing animals in zoos actually gives them a higher appreciation nature, a higher appreciation for biodiversity. So I'm not anti-zoo, but it does feel like a little antiquated. It does feel a little um transactional. Like I pay money, my kid and I have young kids take I've taken them to zoos, right? You pay money, you go see zoos. Feels a little transactional still. So I think we have grander ambitions on that and so while you know we have no intention to make zoos I do think there's highly likely that we will have partnerships with ecoourism with the animals with countries where you can see the animals back in their natural uh habitats. >> Yeah. More like natural like a national park where >> it's like Krueger National Park and all these different locations, right? And and I think that also making that accessible uh and putting science on display versus animals on display is something that we're really excited about. like we want we think it's as important that you understand the conservation impact, the ecological impact in the science of how you make a dodo as seeing a dodo. So I do think that there'll be educational and media experiences. That's a large portion of the company that focuses on that. But I still think that's going to be dwarfed by the government work and just the synthetic biology pipeline. We've already spun out four businesses from the company in four years. Uh two of which we've announced, two of which we haven't. uh one of them kind of got leaked which is Astromec which we raised last round was at $2 billion valuation nine months old. So we are we are building fundamental technologies that I think have broad applications to uh government you know uh conservation as well as that of human healthcare and disease modeling. So, so synthetic biology is kind of this end to-end pipeline. I think it's pretty interesting. But then separately, governments are now coming to us and we're helping them understand the assets in their biodiversity, how they can actually data mine that and protect their species and kind of help them think about biodiversity in a different way. So, not as far uh as the long-term applications to nature credits and biodiversity credits, but how can we help governments underwrite the protection of their biodiversity of looking at, you know, massive non-model species wins like the Hila monster and how the venom from that actually led to a trillion dollar GLP1 market, right? And so, helping them understand these assets that they have should be protected, right? If we can't get them to protect it because they should protect animals and they should pro protect their environment. It's like, okay, if you can't love the animals, can you love the environment? >> Figure out the money. Yeah. No, it's a good >> Yeah. So, so all these things that >> it's a pragmatic look at conservation, right? It's like it's like hopefully we can get them to care about the animals, if not the ecosystem, and if not just themselves, and if not the economics from treating human disease. So, so it is working and we are we now have five government partners online and giving them the tools to really understand what they have and why they need to protect it. So, I I I do think that we can really have this nextgen conservation narrative while also helping countries monetize it in a way that's good for them and their people. >> Well, uh congratulations on the progress. Stuart, anything else? >> No, this was cool. I've been I've been Yeah. hearing about the company forever. It's great to meet you and and uh understand the vision and I'm sure you'll be come back on when you have uh your next uh >> Yeah, we'll let you know as soon as we drop something else crazy. We'll let you know. >> Can't wait. We'll talk to you soon. >> Great to meet you. >> Have a good rest of your day. >> Cheers. >> Goodbye. >> Up next, we have Jake from Serville. We're running a couple minutes behind. We got to catch up. He's the founder and CEO of Serville, launching the future founders program to train nextg operators. How you doing? >> I'm doing great. How are you? >> We're doing fantastic. Uh welcome to the show. uh introduce yourself in the company and the project. >> Yeah, I'm Jake. I'm the co co-founder and CEO here at Serbal. Uh we're an AI platform for employee support. So you've heard of AI for customer support. We supported internal employees when they ask for new laptops and access applications and and all that stuff. And we've got this new program called serval start to help uh deploy servo in these large enterprises and give future founders the opportunity to get inside an enterprise and deploy AI which we saw with the news today is is becoming really the uh the gap from uh technology to actual implementation impact. >> Yeah. Um t talk about like the integration points with the systems that we know. Is the is the point that you can work across different point solutions be sort of a meta solution uh or or will you wind up uh building point solutions and sort of becoming your own compound startup at some point? >> I think the idea is actually the latter that you become the own compound startup where we're increasingly ripping out systems of record like service now and actually taking over that entire service area. Yeah. Uh who's uh where has adoption been the strongest in terms of like massive enterprise fortune 100 down to SMB down to startups like where are you seeing the best traction? >> Yeah, we started like a lot of companies in that kind of tech startup world with great companies like notion and perplexity as early adopters. Um, but we've worked with some of the largest companies in the world, Fortune 20 companies, Fox Corporation, and increasingly, I would say, more excitement and more interest from the largest companies in the world because that's where they have the most pain. They've got thousands and thousands of employees asking for password resets, asking for access applications, all these things that can be automated. >> Yeah. Uh, talk about the future founders program specifically and like how that's structured, why this particular model, how everything plugs together. >> Yeah. And what we noticed was that the biggest pain point is increasingly not the strength of the models. It was actually the implementation in a large enterprise. Understanding the change management and the uh stakeholders and approvals and all the things that need to happen. And you actually need really talented people to go in and run that process that have to be able to one uh build relationships, sell the product, but also build product because you're going to find out all these feature gaps. And when you think about who is really good at understanding customers and selling and also building product, it's all people that are either former founders or future founders. And we started hiring that profile for this role and realized, hey, this is also a great training ground for them because you get to go into these massive enterprises and really go through the motions of being a founder. The same things that me and my co-founder did when we started the company. And so why not really formalize this and focus on hiring the next generation of founders to give them a training opportunity. So we bring them in, we give them opportunity to build enterprise automations, deploy AI, large companies. We accelerate their vesting. So they vest after 6 months >> and we have the expectation that like >> do this for a little while and then go start your company and we'll connect you to our investors like Sequoia and First Round and Redpoint and General Catalyst and we'll set you on your way and give a great reference and give you a great experience. >> Yeah, that makes sense. How how big do you want the program to be? >> We're going to start with a class of about 12. Uh, I think we'll expand that over time. We'll do probably two classes this year and then we'll see where it goes. We might make it bigger or or smaller over time. I think there's a big question across the board on how big a lot of teams get. >> U, but we feel like given our growth, we we need as many people as possible in the shorter term. >> Yeah. Sorry, J. >> No, it's cool. It's cool. I mean, a lot of there's a lot of people out there that want to be founders, but in order but haven't necessarily been inside an organization to discover the right opportunity to go and build. And so I can see this being a win-winwin. >> Yeah. Uh what's the what's the like uh the winning background for someone in this program? Like in order to I I imagine that working at a big company can accelerate you becoming a forward deployed engineer at a different company because you can immediately plug back into your former organization as long as you left left on good terms. Is that the correct model or is it more like oh come out of college and jump into FDE work? Yeah, I think it's actually closer to the ladder. I think you could be a new grad. You could also be someone with a couple years of experience. Intelligence ends up being the bigger biggest predictor and a technical background. So, someone who's CS degree or has worked as a software engineer >> because you're going to be asked to write code and it's going to be production code. It's not going to be just some bespoke automation. It's actually going to be impacting our product. >> And then you've also got to be somebody who can sell a large enterprise. So, we are hiring ranges of experience. So, some are new grads and some are 20 years of experience and have more experience in the large enterprise. We think that there's a fit for everybody, but I think it's going to be folks that are incredibly technical and and that could be starting a company instead. You know, that would be their alternative path. >> Yeah, makes a lot of sense. Uh, well, thank you so much for coming on the show and breaking it down for us. Have a great rest of your day. >> Crush this. >> We'll talk to you soon. >> Let him know. >> Thank you. Appreciate it. >> Cheers. Up next we have G coming back on the show from Panthalasa. One of the coolest companies farming energy from the ocean with mega machines, mega projects. We got to pull up some video of Panthalosa and what G has built over there because uh it is uh a magnificent structure when you see it. Uh we will figure out the waiting room in just a minute. People are showing deepseek progress versus labs. >> China's actually falling falling even more behind. Noah Smith is saying export controls work. Export controls work. Export controls >> work. >> So the map uh the the the chart is that uh if you look across uh GPT40 o OpenAI 01 03 mini 03 then Opus 4 then GPT 5 5.2 2 Opus 46 to 5.4 and then OpenAI is GPT 5.5. Uh America and the United States AI labs seem to be on a steeper curve compared to DeepSeek, Alibaba, Quen, Deepseek R1, V3, Kimmy K2, K2.5, and then DeepSseek V4 Pro. Big jump. Yeah. I mean, if you go if you re if you rebuild this line just between Kimmy K2.5 and Deepseek V4 Pro, it is looking steeper, maybe they'll figure it out, but uh at least for the for the current moment, the AI gap is bigger than you think and potentially export controls work. We'll see where they go, whether they stick around. But we have our next guest in the waiting room, G from Panthalosa on the show last year, but welcome back. How are you doing? Hey guys, how you doing? Long time. >> Long too long. But you've had you've made massive progress in the interview. >> You brought us back a big number. >> Yeah, you brought us back a big number. How much did you raised? Tell us what you >> We did what we could. We did what we could. >> Uh oh. Are we going to do a gong right away? >> It was 140. Let's do it. >> Fantastic. >> Okay, talk about the progress. What specifically unlocked such a huge fund raise? Break it down. I I mean, let me tell you, congratulations to you guys, too. Like >> very been a good one. Yeah. And thank you for jumping on the show so early when we were tiny show with just, you know, making phone calls to random founders. That was awesome. >> Hop on. It was a lot of fun having you on. >> Yeah. >> So, so sorry. What was what was the question? Progress. >> Yeah. The question is like Yeah. Yeah. Why this round right now? I mean, I can imagine a bunch of reasons. I can imagine like you built the thing successfully and it's working or I can imagine just it's very it's way more clear what cheap energy means because with LLM training you load some GPUs on this thing you send it up to a satellite like a lot of the pieces of the puzzle to underwrite this deal feel like they've fallen into place but >> yeah there were a lot of businesses that were pitching at the point that we first talked that have made less and less sense right even things like application layer companies would be one and it feels like >> your business is more more sense as the sort of >> especially on the back of like data center bans and and and energy expenses and uh natural gas flaring and G Vernova's out of stock and even if you get it it's going to be you know natural gas powered like going to the ocean going to somewhere uh far away uh seems to make a lot of sense. >> Yeah. Well, I think there's two things happening at the same time. Like number one is this is a deep tech play >> that we started working on a long time ago and we had to develop the whole new technology to do it. Yeah. >> You know, no one has built an autonomous system to go to the middle of the ocean, capture energy, turn it into anything. Yeah. >> You know, computing fuels is one of the things we're going to be doing too. So, you know, that took time and now we're at the point where we can actually start scaling these systems. So, that's a big unlock. That means that we can start building our manufacturing plants. It means we can start getting all the bugs out of that, getting the bugs out of the first fleet that we're putting out starting later this year. So, so that's the sort of inflection point on the company. But then you've also got, I think, everyone figuring out that there aren't that many ways to get energy on the planet. You know, it's the really hard thing. It's like there's, you know, there's gas, which you can scale, there's solar, but it takes a lot of people and a lot of land. And so, what are your options? people are saying, well, you know, you guys were on to something when you said middle of the ocean is a pretty good play. So, it's been it's been a lot of fun to gather that coalition of investors who are like, wow, this is a totally orthogonal play >> and it makes a lot of sense. We can go and do this. >> How much energy does one what are we calling a device ship? >> Node. We call them a node. Okay. An OD. Yeah. >> How much energy? because we talk about an average meta campus might be half a gigawatt or something and you know we're moving towards a gigawatt a month and there's different clusters and different sizes of campuses uh and I imagine you can create a fabric of these nodes that work together but uh a single node how much energy are we generating how can we think about it when we compare it to uh just the broader like data center campus world >> yeah the typical node will be on the order of 500 kilowatt That's what we're thinking it'll be. It's 100 kow up to a megawatt. >> So, it's like one rack to multiple racks depending on density. >> And yeah, they can talk to each other sideways. So, we'll have radio, of course, they're talking to satellite. >> And so, we're building this grid, you know, and it's it's not for synchronous training, right? This is not fiber between everything. But if you want tons of embarrassingly par parallel inference, you know, running all the same models, running future models that get bigger or smaller, we think we're going to be perfect for lots of lots of that. So um the way that we see this going is like once the energy starts to crash through what's available in the grid, it all just becomes about who can actually build in an elastic fashion to make you know 10 gawatt, 100 gawatt, 100 gawatt per year is the kind of thing that we're getting asked to do. So that's the kind of framework that we want to build up just pure manufacturing to get energy capacity out there. >> Yeah. Yeah. So you build a thousand. So I imagine uh huge industrial project now to actually scale the manufacturing process. Uh what what are you is there anything unique that you need to do around battery storage to maintain uh generation capabilities or would you just like take a node offline if it's in still water like because I imagine that there's a sometimes it's generating more than others depending on what the conditions of the ocean are. >> Yeah. So the the reason we chose this resource and it's not just any ocean that we're going to. We're not going to, you know, 200 miles off the coast of San Francisco or something. We're going to the Southern Hemisphere oceans, >> which are really power dense. The wind is blowing all the time. The waves are on even more than the wind because it's this big battery for wind energy, big battery for solar energy. So when we're there, we're on almost all the time. There's like a couple times per year where we dip down a little and we do have some battery on the system for that. But it's not like solar where you're using the battery every day. It's just a couple times a year. We don't need as much battery. We can cycle that battery far less frequently. >> Sure. >> And so the level of uptime that we want, you know, we can go to 49s, we can go to 3 9s, we can go to like 98% uptime if that's the economic optimum. And so we do a lot of that optimization to figure out for the chips we're running, for the workloads we're running, what is the exact right amount of battery, what is the exact right size of the payload and and all of those things. >> Where where are you thinking about placing a network of nodes >> around Antarctica? >> Antarctica. >> Yeah. The I mean essentially, right, is the southern oceans. >> Yeah. It's it's like way way north of Antarctica. It's like, you know, southern hemisphere, south of Australia, south of New Zealand, all the way around the planet. But like yeah, we don't go, you know, we don't go down into the those parts of this southern ocean. >> Okay. >> Yeah, we we can also do some in the North Pacific and so forth. That's where we're doing our pilot fleet. >> But the real energy resource that we think is optimum for this is that Southern Hemisphere belt. >> So do you I mean if you're like I'm just thinking about if you if the if the if the road is like road to a gigawatt that's like 2,000 of these units that will happen over years, I'm sure. Um, do you have to build the factory nearby to cut costs or if you make them in America, can you ship them down there effectively? How does that work? >> The the best deployment story is one, yeah, where you have your factory or factories pretty close to the resource. >> Sure. >> And so we're working with folks in some of those countries that are nearby to, you know, identify the right sites, go build there. Um, right now we're building our first pilot line though near Portland where we are because we want to like dial in all the manufacturing, get it really good, then we can go and start carbon copying that unit to the right places. >> Yeah, that's fascinating. Uh, Jordan, anything else? >> Absolutely wild stuff. >> It's such a cool project. Such a cool project. Uh, >> I cannot wait for the for for a future video where you are jet skiing between the different nodes. It's going to be it's going to be incredible. >> It'll fit a lot of I don't know if the nodes would even see each other. I mean, you put a thousand down there. >> I know. They're they're going to be jets distance. >> You can spread them out pretty far. >> But anyway, we'll get you guys down there on a yacht. >> Can't wait. >> So you can uh so you can check out the fleet. Yeah. >> Can't wait. >> Incredible progress. >> Thank you so much for taking the time to come chat with us. We'll talk to you soon, G. Goodbye. >> Talk soon, guys. >> Cheers. >> Up next, we have Katie Han from Han Ventures uh announcing a new fund. This is very exciting. >> This the biggest number. No, >> beater, >> but >> you almost had the biggest number on the show today. >> Uh, we were catching up with an but uh anyway, thank you so much for taking the time. Welcome back to the show. How are you doing? >> Thanks. I'm doing well. How are you guys? Thanks for having me back. >> Of course. It's always good to talk to you. Uh I mean I there's there's a lot of different directions we could go but let's let's talk about uh the fund the thesis uh expansion interactions between different thesis like where how you're seeing the market how you're seeing opportunity and venture broadly right now. >> Yeah. So it's been four years since we launched Han Ventures and you know the world looks a lot different four years later and uh we're we're excited to have a billion dollars in fresh funds to deploy behind >> uh founder. Oh, thank you. Even though it wasn't the biggest of the day, I appreciate that. >> Wild times that we are in in the venture world. For sure. For sure. >> Um >> well um we're we're we're taking this fresh funds of billion dollars and backing um founders who are building what we're calling the new economy. And by the new economy, I I'm thinking about three structural shifts we're seeing right now. The first is um new financial rails and infrastructure. I mean, think companies like Arabore, right? gone our gone our banker's hours, gone our wire cut offs. I mean, you're talking about global from day one, 247. This is a bank that opened on a Sunday. Um, the second structural shift we're seeing are new assets and markets. It started as stable coins, but it's quickly kind of we're now talking about tokenizing the stock market and you see giants like Black Rockck and you see Coinbase and Robin Hood doing that, but it's also opened up new markets. I mean, think prediction markets, think perpetuals markets. You've had a lot of those guys on the show. And then the third structural shift we're putting this new set of funds behind. And this is the earliest of the category, but of course we have an early stage fund too is what we're calling the agentic future. And it's not just AI broadly. It's where AI and crypto intersect. And I think that's a lot more areas than people realize. I mean, we're talking about building for a world where the end user is not necessarily a human of a financial product or service, but is a computer or an agent. And so, how does that impact things like privacy? How does it impact things like provenence and trust and what form will they use? How will agents pay for things or subscribe to services? And blockchains aren't good for everything, but they are really good for some of those things I just mentioned as are other cryptographic tools. >> Yeah. Yeah. I mean, I remember that being like back in 2015, 2016, the original machine toachine payment uh thesis around Ethereum and even some Bitcoin folks were talking about it. Uh tons to dive into there. Um just in general uh it it we we were we were talking to the Collison brothers last week about how uh stable coins are maybe in some sort of a winter like there was a big surge of 30% growth in the market and then it's been a little bit flat like do you think that uh this is a regulatory story is this a technology story like what is the next uh breakout adoption yeah the the next adoption hinge point for uh for just crypto technology to diffuse further. >> Yeah, sure. Well, I mean, I think actually I take the other side of that. I think you have now um Mastercard, for example, just paid 1.8 billion for one of our stable coin investments, PBNK. And that was the third largest ever in history acquisition by Mastercard. That's crazy. >> And so I don't I don't think we're in a stable coin winter. Um I think actually you have doubledigit trillions of dollars in transaction volume now flowing through stable coins. And um this is a technology that didn't even exist, guys, 10 years ago. Yeah. >> And now it's about to surpass the combined transaction volume of Visa and Mastercard. And we think that'll continue. But the story doesn't stop at tokenizing dollars because now of course a lot of economies want stable coins for their currency. Yep. So I think that's a tipping point. Another one is tokenizing other financial products and services. So like the stocks I'm talking about. >> Yeah. Yeah. Is that is that a disruptive innovation or sustaining innovation? Because I can imagine that the Black Rockcks, the Coinbases, like the Robin Hoods of the world like do very well in that world. They're already set up for them. They have the audiences and the customers and they, you know, expand into that category. At the same time, you could imagine we've seen upstarts like Poly Market, Kali like you know you know slightly new twist on an old idea and it's just huge new unicorn or decacorn emerges uh and sometimes multiple in a single category. How are you viewing the idea of like onchain equities as a venture opportunity? >> Well, and it's not just onchain equities. I think it's starting there, right? It started with fiat and now starting with, for example, securities or stocks. But I think it won't end there. I mean, I think it will be all kinds of financial products and services. And I'm not one of these people who is going to tell you all assets will be on chain. Every single physical asset will be on chain. Um maybe eventually that's an end state, but I don't think it's an end state necessarily in our lifetimes. But I do I do believe that financial services and project products will end up on chain and we're just starting to see that. You mentioned prediction markets. It's a great example again. And by the way, you talk about the original prediction market was Augur, right? And that was just ahead of time years and years ago. I mean, when was that? 2016. Um but that was before you had stable coins. And it turns out you need some of the infrastructure to really have prediction markets find true product market fit at scale. And that's what they're that's what they're undergoing right now. And I think we're still early in prediction markets. It's like sports and politics. >> Yeah. Last time you were on last time you're on I I think you you made the call if I remember correctly that yeah it was still early in stable coins and we sorry not stables but uh prediction markets and you were you were correct. There was a boom. there's a bunch of new kind of like vertical approaches. How are you how are you processing it? Do you expect to make like a net new prediction markets bet out of this fund or have you made your bets at this point and and you're just going to watch it evolve? >> Well, in fund one we made a bet and Coinbase bought that bet and that was the clearing company. So, I certainly hope my prediction is that we have a prediction market bet in fund two. We don't have one now. uh it is some an area where and and but the prediction markets that we invest in might not look like what they look like today. Remember we're a venture fund. We're not a hedge fund. So we're looking at over the next 10 years. And I think an area I'm particularly interested in and we as a fund are interested in for prediction markets is some enterprise use cases. Think about insurance. Um think about litigation predictions. Think about drug trials. think about risk hedging and all kinds of other business uses aside from the fun of sports betting, aside from the fun, if you can call it fun of pol, you know, betting on politics outcomes. Um, there are a lot of business use cases and institutional use cases and that's why you see folks like the New York Stock Exchange and ICE getting involved. Um, but I think we're really early scratching the surface. So the question is is it going to is value going to acrew there to one of the established entrance already or will some new upstarts come about and so we're keeping an eye on both and as I said there's huge opportunity but with huge opportunity is going to come a lot of legislation we think and regulation um and >> yeah how do you think the whole how do you think the whole insider you know given that you don't have an active bet it sounds like how do you think the whole insider trading >> oh yeah >> confirming we do Not we do not have an active bet. >> Yeah. Yeah. Yeah. That's that's what I'm that's what I'm saying. Um so I just feel like you can um >> but I I have a huge bet on her taking a bet in the next couple years cuz she just said that. So I'm going super long on that. But but my question is like >> in the category and we have maybe I should rephr I just wanted to ask around like how how you see the insider trading debate evolving because clearly if if you're somebody that is looking to prediction markets for alpha for data theoretically like like >> prediction markets >> you mean specifically like right now it's handled on terms of service level and it might eventually be handled at like the government regulatory level. >> Yeah. >> Right. I think now you have a little bit of what we'll call self-p policing and you know industry best practices but the industry is still early and what are best practices today might not be I mean and I think a lot of these platform look a lot of responsibility is going to fall to these platforms. Um this is novel issue with any nent technology you're going to have novel issues of first impression from a statutory point of view from a constitutional point of view. you're talking about insider trading. That's a criminal point of view. And as you guys know, I spent over a decade as a federal prosecutor at the Justice Department. And I've seen a lot of um a lot of situations emerge that frankly companies weren't really contemplating and that just come about and I've seen a lot of that in my career as a prosecutor. And I think that these platforms are really going to have to be thinking very deeply about these issues. And I think they know that. I mean, um but they're they're only going to be heightened because of the opportunity. And I think that's really exciting, but it's the flip side of the coin, right? And it's not just insider trading. Right now, these platforms are asking for federal preeemption, which I think is really interesting because you might want federal preeemption today where you have a a CFTC chair who is not hostile to prediction markets, but don't forget that Gary Gensler himself was once the CFTC chair. Yeah. >> Um and and so I think it's a really interesting question of do you go state by state? Uh do you ask for federal preeemption? And so there are a lot of questions beyond just insider trading. >> So so un unpacking that risk, there's a world where you get federal preeemption, but then as different rules and laws are written, if there's a different less friendly CFTC chair, uh the downstream like the implementation of that oversight could be disadvantageous to the industry. Is that basically the risk? >> You're you're exactly right. You could win the battle but lose the war. >> Yeah, that's interesting. Yeah, but but but that's I I do see why a lot of those platforms are looking for federal preeemption right now. But again, you got to look at this as a multi-t trillion dollar asset class and you've got to look at it over the next decades. Yeah. And I think, you know, social media was a huge category, a huge market opportunity for investors. And yet also these platforms had to get really sophisticated over the last decade with how they police content for example and how they work with regulators. And it's not just in the US. As I said, these this this structural shift of new assets and new markets means it's global from day one. So you're not just thinking of the US. You have to think about globally. Um what do regulators across the world think about this? And they don't always just follow what the US and the CFTC think. Um, on the intersection of AI and crypto, are you equally excited about bringing crypto to existing AI agents? Someone has an open claw and they wanted to buy something and so stable coins speed that up, machine to machine payments, micro payments, all of that. Or is there actually more of an opportunity or maybe an equally or maybe less discussed opportunity around uh bringing AI to crypto? thinking about like like a a cursor, but it's really good with writing smart contracts or something like that where you're still primarily selling to the crypto community, but you're bringing AI tools to bear for that world in the way that there were several big crypto winners that were sort of web 2.0 SAS products, but they applied their strengths to the crypto world very successfully and and built huge businesses. Are both of these equally exciting? Is one of them more hyped than the other? How are you processing those two opportunities? >> I think they're both really early and I think they will naturally they're >> we're interested in both. Um I think there's a case to be made for both and we can talk a little bit about that but I do think that it's more of an opportunity for early stage fund right now because we are very early at this. Um I think to the first point of you know you mentioned micro payments. This is something that those of us in the crypto community over 10 years ago were talking about uh with companies like Chainchip and other things and it turns out that fast forward now there is a a use case and I think we've heard John Collison talk about um how he's tickled about this um and I think that's the word he used tickled about the case for micro payment. Yeah. >> And you know we don't think agents who work 247 around the globe again they don't work bankers hours. we don't think they'll necessarily just be using credit cards for payments. I mean, think about the need for instant settlement um and if you think about the need for finality um and these transactions that are again around the globe but also microp payments and we think crypto rails are actually perfectly suited for that. So that's the first thing but in terms of bringing blockchain based solutions or cryptographic tools we also think there's a lot of exciting areas there where there's an intersection and again it's very early um but you can imagine provenence you know the blockchains are really blockchain based systems are really great at proving provenence and I think in a world of AI where you're wondering the proven you're going to be wondering ever more about the provenence of things we've already been wondering about the provenence of things where most things were created by humans Uh, and now that's starting to shift and more and more is being created by computers and we think more and more will be created by agents. And so what does that really mean for Providence? What does it mean for privacy by the way when all of a sudden all of your data we have things you know GDPR and the California equivalents that are out there? But in a world where you have different AI agents that can really quickly much more than humans can kind of undo that privacy, do you look for zero knowledge base based proof? um solutions for that and that's again a cryptographic tool. So I would say reputation systems, provenence, privacy, um agentic finance, they're all fair game and right within the wheelhouse of what we've been doing guys for the last four years and we're excited for the next four years. And I think we are probably it's it's for a billion dollar fund. We have the trust of 35 or so global institutional investors from around the from around the globe um who very much believe in this future of the intersection of these technologies. >> Yeah. No, it makes a ton of sense. Uh yeah, talking to the Collison brothers really uh I don't know just re very much like reset me on the uh the actual need for micro payments, the value that comes from that. talking about uh token theft and basically setting up an account getting on 30-day billing not being good for it at the end of the 30 days like that's solved if you're paying in a income streaming payments actually makes a lot of sense there beyond >> and agent and if agents have something to go look at that's immutable this layer that's immutable and so you start to see and you know it's not just the Collison brothers and stripe that are doing this obviously Coinbase and companies like Robin Hood are really leaning heavily into these areas And then you have again the giants like Mastercard or Visa who I mean I think Visa's chief product officer um was saying this next period over the next decade for Aentic Finance is one of the most interesting. So they're keeping their eye on it closely too and and so are we >> amazing. Uh Jordy anything else? >> Tremendous progress. >> Congratulations. Thank you so much. >> Let's not let it go another another year for your next appearance. But um >> absolutely not. >> Congrats. We'll talk to you soon. Have a good one. >> Thank you. >> Goodbye. >> Thank you all. Bye. Bye. >> Up next, we have Nick from Rivet. He's the co-founder and CEO. He's been on the show. Has he not been on the show before? >> He's been on the show. >> He's been on the show before, right? Nick, welcome to the show. You've been on the show before, right? >> Yes. Quickly, first heat round. It was a great time. >> Welcome back to the show. Uh we need to update our our CRM, our database. Uh cuz we have you down here's first appearance. I know that's fake news. Anyway, uh, a lot of people have been worried about artificial intelligence. They say, "Is it worth the water? Is it worth the energy?" But I think after they see these results from Taxbench, they're going to change their tune. Tell tell us about it. >> Yeah. Uh, so Rivet is >> Well, before Yeah. Yeah. Introduce the company first. >> Yeah. Um, Rivet is an AI enabled accounting firm. We do tax returns and tax advisory work for thousands of companies. Um, >> when we implement these models Oh, there you go. Yeah. uh when we implement these models, right, we need a way to test them. So, we've had an internal benchmark that we've used for a long time to see whether these models are actually doing our job as accountants effectively. >> Uh it was time to publish that benchmark and that came out today. >> Uh the results are pretty striking. Um the top models don't get most questions right if you ask it the same question five times in a row. >> Interesting. >> Uh a lot of them do incredibly well when you ask it once and cross your fingers that you're going to get the right question or the right answer. >> Yeah. Yeah. Well, you asked the same question five times in a row. Are these models actually reliable? Turns out they actually really aren't, especially at scale. >> Interesting. Uh, and >> and why are they not like doing enough RL like you would think that like these these these verifiable there's verifiable kind of >> like um results, right? Like this should be something that the models could do. >> We need lean, but maybe it hasn't been >> prioritized enough. So most of the answers require five, six, seven, eight steps and you have to get the right answer every time. If you get one step wrong, the ultimate result is wrong. >> And what we've found when you ask it these types of questions is that it'll get a lot of it right, but that single big skip, you know, hey, my client lives in New York City. Uh they're about to sell their company. uh does their uh you know does their stock qualify for QPS? >> It'll check if they held the stock for 5 years. It'll check if they uh were under the gross asset limit when they acquired the stock. >> But then it'll pull a brand new article saying, "Hey, New York's considering getting rid of QPS." >> Oh, >> oops. It must not recognize QPS. No, they don't qualify. And and ultimately the client is is is left out, you know, in the dust. >> Yeah. >> Pricing too newssheavy, too like web heavy. not like law. >> Yeah. And it's an interest it's a very interesting domain because these are these are this is not like doing like research for like a project where like you know the the the consequences of getting like a fact wrong or like somewhat defined like maybe you just like came to the wrong conclusion. But this is like >> you if you if you don't work with a human at all and you just work with a model and then you get something wrong, it ends up being like maybe it costs you $100,000, maybe it costs you a million dollars, maybe maybe more, right? maybe maybe it's even five grand and then still and so I think like this I I expect a long period of of human in the loop on and potentially forever on this domain specifically because for for a number of other reasons. Um anyways full employment for Nick >> the final the final boss. Um, have you have have any of the labs like reached reached out and and like are they are they actively working on this or is it not not a priority because they know people like you will will be putting them the models through the paces and and kind of like verifying the work. >> So, we'd love to try all the new, you know, secret stealth models that no one's, you know, that are not available online quite yet. Uh, we do have an email entry on the website if you want to try your own model against our benchmark. We'd love to run it against it. >> Cool. Uh, nobody's reached out quite yet, right? And and I'm not super surprised. Researchers have very different priorities than real businesses running these workflows. >> I don't think in philanthropic cares about my benchmark, which is fine, right? That's not their business. We're going to pick the best model for our for our purpose and move on. >> Yeah. >> Um, funny enough, even the new models don't do as well as some of the old ones. Uh, we've actually found regressions as they launch new models. They perform worse and worse. >> Uh, 47 performs worse than 46 from Enthropic. Um 5.4 Pro performs better than 5.5. Uh the 4.1 model from Grock beat their 4.2 model. Uh so it's not always newest is best, flashiest is best. Uh you really have to test the specific model that's come out to make sure that when you're implementing it, it actually does improve your results as as a business and you know the person calling the model. >> Interesting. Yeah. >> Really fascinating. how how uh you started this business not I I my understanding is you didn't start this business because you saw like models getting much better but really that you just thought you could build a better service around tax for startups and you now I guess get the benefit of of uh the intelligence explosion. Where are you like even with the flaws? Are you still getting a lot of leverage out of them? Uh or is it not really because of the lack of reliability? Is it is it not giving you the the quite the leverage that you'd like yet? >> No, it's a totally fair question. Uh we get a ton of leverage as long as they are paired with a human who really knows what they're doing. Um you know, you talked about the stakes for some of these questions. If you're just playing with cludge to write a poem or make a website, you as the end user know what you like and don't like and you can steer it towards what you want to see and don't want to see. >> If you aren't an accountant, you have no idea what to steer it towards. You have no idea to prompted, hey, please double check that New York recognizes QSBS. Hey, please double check that you pulled the, you know, new tax rules under OOTB. >> Uh, you need to have somebody skilled who knows how to talk to it and engage with it and structure the the prompt correctly. uh and make sure that when it tells you something it actually smells and you know sounds right. Um so we do get a ton of leverage out of them. Uh we have quite a few workflows that are powered by them internally. Um but it requires a ton of work on top of that call to make sure that the answer that's ultimately shown to the client is correct. Uh and and not just you know hallucinated figures or hallucinated rules that that'll ultimately uh you know nobody's going to jail but they're going to get a lot of letters in the mail and and owe quite a bit of money. >> That's not good. Yeah. Uh and uh Andre Karpathy was talking about how uh there was a huge jump in chess capability between GPT3.5 and GPT4. And what he attributes it to is just like the researchers were interested in chess during that training run. So they fed it a bunch of chess data and it got better at that. And it feels like there might be a moment in the future where like oh like this lab on this model date got really interested in chess and went and bought like all the or in tax data and bought all the relevant training data and now it's now it's uh jumping up on tax bench. Um if you were to I mean I'm looking at some of the stats here and it shows like uh Grock 4.1 fast reasoning at 4.2% 2% uh you know some others are in like the 10% 12% 31% range. Um uh is this is this uh like uh ARGI where where where we're just at very very low uh passing rates and and you're sort of waiting for them to start climbing up exponentially but we haven't seen that takeoff yet. So to the models's credit, the scores that you're referencing are mostly from data retrieval, which is the most operationally difficult category of questions that we ask the models. >> Uh they're given access to a client's entire data packets. This could be hundreds of pages of PDFs and Slack messages and emails. Okay. >> And asked to answer a question. A question could be something like, yeah, >> find their 2023 tax return. Uh find the carry forward capital losses that go under a 24 return. >> Mhm. If you ask a junior CPA straight out of college to do this, look at it, write 100 times out of 100. It's not difficult. It's in the exact same spot in the exact same form every time. >> Yep. >> These models really struggle. >> Uh they really struggle. They um we'll just make a number up. >> Yeah. >> They will get frustrated when they can't find the 1040 off the bat. Uh maybe they have trouble OCRing. You know, the client took actual pictures of the the tax return versus, you know, a real raw PDF. Yeah, >> they really struggle with the combination of searching >> and then analysis on top of that. A lot of them skip uh they'll pull the wrong carry forward figure and ignore that there's a $3,000 allowance against current income for 23s. You have to subtract 3,000. >> It's it's small ticky tacky things, but if you try to deploy these models into a real production workflow, they're going to get it wrong. You have to have a human review it right. You have to have a neck to choke. Something goes wrong. Uh, and so we've we've hired a great team of necks to choke, so to speak, uh, to actually make sure that the work uh, that's delivered to the client actually, uh, does what it says it's going to do. >> Have you thought about building a harness? It feels like a lot of the the unlocking software was was, you know, on the back of codeex and cloud code like like something like the harness is is somewhat simple sometimes, but it clearly unblocks the thinking model in a very important way. And it feels like for some of those rules-based or or tool usages like just something a thin wrapper around these uh these models uh could potentially lead to much higher scores. Have you looked at that? >> Uh you are hired as our next product manager. Welcome to the team. Thank you. >> Uh you start on Monday. No. Um it's absolutely on our list. Okay. Uh, one of the most impressive tools in the market that we've seen today is Thompson Reuters's co-consel, which is >> the UI looks like a any other model, but it's built on top of their uh, tax library. >> Oh, interesting. >> They don't give anybody API access today. It's hundreds of dollars per seat per month for >> uh, and so I would love that as an API. Uh, we'll be building on it when they launch it, I think, coming later this year. That's what we were told. >> Yeah. Yeah. >> U, but in the meantime, we'll be building our own and, uh, jumping down the rabbit hole, so to speak. Good luck. >> Yeah. Anything else? >> Well, great to get the update and um yeah, we will uh every every researcher that comes on >> tax bench. This is the most important one. Forget Eric AGI. It's all about tax AGI now. I don't care about >> G. Well, I think I think as as the IPOs start happening, they'll start caring more about uh about performance on tax >> probably. Yeah. I mean, it's a huge huge market, huge uplift. You've seen what happened in the coding market. Uh you if you add accounting and finance to that, you you can see the next leg up on all the revenue charts. Uh well, thank you so much for coming on the show. >> Great to see you, Nick. >> Have a great day. >> We'll talk to you soon. Cheers. >> Goodbye. >> Bye. >> Uh we do breaking news for gamers out there. >> What's that? >> I know a lot of gamers out there, they have laptops. They're worried about the price of these laptops, the batteries. Oh, electricity is getting expensive. What are you going to do? How are you going to charge your laptop up? We got the solution for you. It's a gas It's a gasoline powered laptop. So, it's pretty simple. It's a one-of-a-kind gasoline powered laptop. It's offered for just $850. Looks like it's running Windows XP. >> You might not be able to play the latest and greatest games. Will it run Crisis? Maybe, maybe not. >> With a full tank, you can get an hour and a half of runtime out of this. >> It runs a two-stroke engine and it's perfect for off-grid computing that's hot right now. And so, uh, the laptop specs, let's take you through it. It's got an Intel Core 2 Duo, 2 gigs of RAM. RAM's going up in price. This is valuable. This is an appreciating asset. 120 gigs of hard drive space that's going to hold a lot of games. When you're off-rid, you only have a little bit of gasoline. If you want to game, this is the laptop for you. It's a good running Windows XP. >> And it says that it starts easy. The the the two-stroke engine on the gasoline powered laptop. Yes, it does start. >> Colin in the X chat says it gets 300 tweets to the gallon. >> 300 tweets. >> Oh no, I accidentally set GTA 5 to max settings. Well, at least it'll serve as a benchmark test. Uh how many Chrome tabs can it uh open before it crashes? People are having fun with this. But uh the gasoline powered laptop, this is true hacker mindset. Whoever built this is uh is an incredible engineer and did something. They did the impossible. They built a gasoline powered laptop. I've seen a couple other of these like uh gasoline powered uh projects, people making all sorts of different things. It's always a funny gag, obviously. >> The actual breaking news is the White House is considering vetting AI models before they are released. >> Trump admin, which took a non-interventionist approach to AI, is now discussing imposing oversight on AI models before they are made publicly available. Well, FDA for AI, we'll see. >> Uh, it would be potentially an executive order to create an AI working group that would bring together tech executives and government officials to examine potential oversight procedures. Okay, so this would be an executive order to create a working group that could potentially create an oversight uh body. >> Okay, so we're a couple steps away, but >> seems reasonable. I don't know. Depends on what the what the what the what the benchmarks are, but uh you certainly don't want Trump says, "We're going to make this industry absolutely the top because right now it's a beautiful baby that's born." >> Interesting way to put it. >> We have to grow that baby and let that baby thrive. >> It's a real quote. >> Are you messing with me? >> This is a real quote >> that Trump said >> about AI. He said, "We have to grow that baby and let that baby thrive. We can't stop it. We can't stop it with politics. We can't stop it with foolish rules and even stupid rules. >> It's not a baby. It's a 10 trillion dollar industry. It's like the the engine of the global economy. Uh anyway, >> Dean Ball's got a quote in here. Said the techn is moving extremely fast and there are few formal procedures, but they don't want to overregulate. Said it's a tricky balance. >> I say don't release it unless it's acing tax bench. It's got to be able to do the taxes before it gets out into the wild. No, obviously you want these models to be safe. You want them to be reliable. You want them to avoid negative externalities and uh anything that's gets us in that direction is probably good, but everything comes with trade-offs. So, >> final post of the day >> from Tommy. >> Hi, PhD in hammerology here. >> All right, so what we're looking at is a nail. >> That is the correct mindset. When all you have is a hammer, everything looks like a nail. Uh, also go check out Riley Walls's new project. He's shipping stuff every week. This one got a million views. You probably already saw it. 27,000 likes. 10% of AMC movie showings sell no tickets at all. So, if you want to go see a movie in a private theater with no one else, he made a site that finds empty theaters and tells you exactly when you should go and book. You can go see Project Hail Mary at 12:30 p.m. today in New York. Uh, if you don't have work, you can go see Project Hail Mary in your private theater. It's available at walzer.commptscreenings. walzr.com emptyscreenings. You can search by zip code. Let's see what's around us. Is there anything good? >> There's 10:45 p.m. Devil war product 2. Okay. Around us, zero seats. >> Enjoy it. Enjoy being a last. >> Got it. >> This is very funny. Yeah, there's he does he does uh surface some that have one seat or two seats. Interesting way to make a new friend, me and you. Cuz you think so. You think, "Oh, I I got the zero seat theater. You're you're in the one seat theater." And somebody's like, "I want to meet the psycho that went to the empty theater." And then they're talking your ear off. Who knows? Well, maybe maybe you can enjoy it. Well, thank you for tuning in. We'll see you tomorrow at 11:00 a.m. Sharp Pacific time. Leave us five stars on Apple Podcast and Spotify. Sign up for our newsletter at tvpn.com. And we will see you tomorrow. Goodbye. A >> little bit missed time there. See you tomorrow.