Neural Computers, GameStop’s $55B eBay Offer | Diet TBPN

0:00Would you look at that?

0:00John Coogan:Would you look at what, Jordi? Everything looks fine to me. We're back. We're back. That's what I'm looking at. We're getting ready to last second. We may have installed Modern Warfare 2 on this large screen and gotten a little bit behind schedule with some of the interns playing. Yeah, Ben and Tyler were, I was cooking him. It was a knockout, dragout fight. You won by a lot, right? Yeah. Well, it was pretty embarrassing for Ben. Ben, considering that he is still in that, not, not, not, not chief producer Ben, but other Ben, considering he's still in like the main kind of like demo video game.

0:41John Coogan:But at the same time, I mean, you were probably two years old when modern warfare two came out. So you got to sort of relearn the, the old, the old tricks, the old tricks on rust. Anyway, big show. I hope you guys had a great weekend. Hope you had a great weekend. Big week. We'll be, we are going to a conference on Wednesday and Thursday. So we might be off both those days, but we have some great shows for you planned Monday, Tuesday, Friday. Did you say conference? It's a conference. Oh, I thought you said concert. No. Although I think there might be a concert. That would be extremely out of character.

1:14John Coogan:Well, I mean, there's a bunch of news we're going to go through. But first, I was just sort of reflecting on there was a great interview with Andre Carpathy at Sequoia AIS Sent. I think last week it went up on YouTube. and 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 image generation is really good at infographics and effectively designing slides or output. And so we're starting to see the rumblings of this idea of the neural computer. 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 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.

2:15John Coogan:That was p a piece of the vision This has always been theorized, but it's becoming more and more real and so So Karpathy describes this idea of like a neural computer this way. And I think it's an interesting framing. Obviously, it'll have implications for SAS products that might be used in a headless, under the hood scene way or might be competed with against these neural computers. But Karpathy describes it as, 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 fly 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, is an interesting paradigm shift that it feels like we're starting to see glimpses of.

3:04John Coogan:So 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. But I haven't tracked either company closely. We've had Ryan on the show, and 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 are the different multiples? And so in a pre-ChatGPT 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.

3:39John Coogan:If I wanted to be really advanced, I would have copied and pasted the stats into a spreadsheet. If you're really working on Wall Street, you might have like CapIQ or Bloomberg plugging into a sheet, an Excel sheet that then can build you a comparison table and do like comps. and then once we got into the chat GPT world, you might do a deep research report, pull all that data, put it into a table, which is effectively markdown. And sometimes the tables renders a little weird and like you can kind of bounce around. But now the whole process from start to finish is just a single prompt and it outputs an image.

4:10John Coogan:So you can pull up the image that I generated. So 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, top line, earnings, revenue, valuation, how the multiples fit together, build a nicely designed side-by-side comparison of the two companies. 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 three times the revenue, 50 billion versus 15 for GameStop and revenue growth. eBay is growing while GameStop shrunk by 5%.

4:45John Coogan:eBay grew 8 % operating income. eBay has 10 times the operating income at$2.28 billion versus$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 4.5x. But on a market cap to net income, GameStop has a higher value, 34x versus net income versus eBay is 25x. So you can just sort of see this table. And this is something that usually would have been like three or four steps to get here. And instead, it's just this single prompt.

5:26John Coogan:And so I think that there's like, this is not a perfect result. Like in that image, you can see that like, 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 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?

5:56John Coogan: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 it's ultimately shareable. More importantly, I don't care what it used under the hood. It could have puppeteered a spreadsheet and put it all in comma separated 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 Scikit-Learn, 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 Vibecode a stock comparison tool with API integrations, make sure you have 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.

6:41John Coogan:And so Karpathy 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. 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 one-shot-able by the model.

7:16John Coogan:So let's pull up Andre Karpathy talking at Sequoia AI Ascent about his experience with software 3.0. I think one more maybe example that comes to mind that is even more extreme than that is when I was building Menugen. So Menugen is this idea where 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. 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 bytecoded this app that basically lets you upload a photo, and it does all this stuff, and it runs on Vercel, and it basically re-renders the menu, and it gives you like all the items, and it gives you a picture that it uses an image generator for to basically OCR all the different titles, use the image generator to get pictures of them, and then shows it to you.

8:08And then I saw the software 3.0 version of this, which blew my mind, which is literally just take your photo, give it to Gemini, and say use Nano Banana to overlay the things onto the menu. And Nano Banana 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 spurious. It's working in the old paradigm. That app shouldn't exist. And yeah, the software 3.0 paradigm is a lot more kind of raw. It just, neural network is doing more and more of the work.

8:44And 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.

8:52John Coogan:um so i think that people have to kind of like it's over perhaps um yeah it it i mean it's it's it's real and uh 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 first thing that was on my mind was that although we have gone through this crazy vibe coding boom where everyone is vibe coding apps it feels like a very temporary arbitrary aberration. And also I know that even though there are millions and millions of people that have used Codex and Claude Code and OpenClaude, like the numbers are big, but it's not at 20 % of the U.S.

9:30John Coogan:population. Like it's just not at that level of adoption as opposed to chat apps, which are at like 70, 80 % penetration, right? Yeah. The other thing that's been interesting is 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 CloudCode and Codex 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 in many ways. That's what I'm saying. So like 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.

10:15and they're already widely available. So it's been interesting.

10:19John Coogan:Yeah, so I think there's two things. One is that if you've been hesitant to jump into Vibe Coding because it's a little bit too much of a hassle, Andre Karpathy is obviously very comfortable being like, oh yeah, let me deploy to Versal and do all this. 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 really, really burning the midnight oil to get this app deployed and do all this stuff. A lot of that's going to go away And like, you're not going to need to do that. But then there's also this question, what you had, which is like, there needs to be this higher order loop of thinking around, okay, you have a problem.

10:54John Coogan:Should you actually vibe code an app or should you just try and one shot it with the current model capabilities? Because for a lot of things - You mean, and within - Yeah, within Shachipiti, within Gemini, within Claude, the actual apps, like you can like take a picture of your food and say, hey, start tracking my calories. There's a lot of things that the apps can just do in one chat thread that people are doing. But I think that there's this tension between when you actually need to for sure go and vibe code something versus when you can just do it in a one-shotted LLM context. Frontier models are already able to, in basically 90 % of situations, I feel like instantiate exactly whatever is required to solve the actual problem under the hood, entirely abstracting away code and tools.

11:42John Coogan:You will just not be aware of what's happening and it doesn't matter. And then the second - Yeah, I would add to my previous statement 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 and you can get customers and people will pay you money. But it has been fascinating to see, does this actually need to be an app? Yeah, yeah. I mean, there's a ton of apps and software that will still be valuable, whether it has a liquidity pool or some sort of unique source of strength or some differentiation point that the existing chat apps can't hack at all.

12:24John Coogan:And then there's also just 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 the specific thing and so we see those in the uh in the app store all the time the other thing that i was uh uh reminded of was did you ever read union square ventures 2016 blog post fat protocols Are you familiar with this? So fat protocols was this concept around how in the web, like web, I guess 1.0, 2.0, there were protocol layers, which are like TCP, IP, HTTP, SMTP, like file transfer protocols, HTTP.

13:12John Coogan:And for like a couple of years, the crypto community was like the group that developed and maintained HTTP. They basically created the standard that the web ran on. And yet very little value accrued to the creators and the maintainers of that protocol. And crypto would be different because the Bitcoin protocol had the value capture component like baked into it. And so there was this idea of like the application layer in blockchain would accrue 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 how the actual information, the photos, and the text gets transferred to you.

13:58John Coogan:But the HTTP standard does not accrue the value. The value accrues to the application on top. And if you scroll down, you'll see the blockchain example, which was sort of borne out that the application layer was pretty thin on top, and most of the value went to the tokens and the protocol below. Yeah, exactly. Ethereum is a good example. Solana is a good example. Of course, there's value in the application layer, and there's some companies that are being built. But this was basically this thesis that he says, we see this very clearly in two dominant blockchain networks, Bitcoin and Ethereum. The Bitcoin network has a 10 billion market cap.

14:34John Coogan:Wow. I think it's like a trillion now, right? Isn't it 700 billion? 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 the point is still true. 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 was sort of the core thesis on this fat protocols thing. And I think there's something similar happening in the AI value chain. Of course, there's like a bunch of other dynamics going on in the AI value chain.

15:14John Coogan:And there's a lot of capture 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 so increasingly you can just get more and more out of the core model, which is interesting dynamic. Third, there's still like this huge question of like walled garden jumping. We've talked about this before, but it's almost that 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. There's great information in Substack on certain 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.

15:56John Coogan: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, 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 and so the models I think will continue to find a way under the wall garden over the wall garden through the walls like they'll seep everywhere and it's more of a question of just inference cost, how long it takes to actually grind through the wall.

16:44John Coogan:But they're already figuring out a way around. And OpenClaw is a good example of that. A lot of the walled gardens were sort of brought down by running. Yeah, except I think SAP came out and said, no unauthorized agents here. 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 agents. Yes, but 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.

17:18Rune 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 yeah the model wasn't uh John Collison talking about this?

17:30John Coogan:John Collison was saying uh like the humans get the thing off the high shelf. Every time I have to go and export a PDF and upload it to ChatGBT, I can't get it in there by default, even though I could just give them a web URL, I have to export it or print or whatever. 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, it came out through the Wall Street Journal. GameStop was preparing to make an offer for eBay as part of CEO Ryan Cohen's plan to turn the retailer into a$100 billion 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 this morning.

18:10If eBay isn't receptive, Cohen could decide to take the offer directly to eBay's shareholders. And they released a letter yesterday to Paul Pressler, who's the chairman of the board over at eBay, happens to be a friend of mine. Really? Yeah. Wait, really? Yeah.

18:33his daughter and my wife are good friends. Oh, cool. So we end up hanging out a decent amount. And we're neighbors.

18:41John Coogan:Does he do a lot of podcasts or press appearances? It can be discussed. I'm just saying, like, dominant determinism of Pressler would be like somebody who dominates the press. The press circuit. I think you would be doing a press circuit all the time. Anyway, Paul is fantastic. Yeah. And Ryan wrote this letter to him yesterday saying, GameStop 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 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 because Ryan Cohen discussed this on CNBC this morning.

19:27Yep. That represents a 46 % premium to eBay's ineffective closing price on February 4th, 2026, the day GameStop started accumulating its position in eBay. And blah, blah, blah, blah, blah, blah, blah, blah, blah, blah, blah. But let's go straight to the CNBC.

19:42John Coogan:So quickly, there was a question. So Bislett 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 in the ring with Aaron Sorkin over at CNBC. Put the gloves on. Andrew Ross Sorkin on CNBC, Squawk Box, to interview about the GameStop eBay acquisition. We can play this. How you could get to that price and how it would work. It's on our website. It's half cash, half stock. But the details are on our website. can you help i i've read them but can you help our audience understand them yeah what which part exactly well i think we can start with the idea that the market cap of gamestop is call it 11 billion dollars uh you have nine billion dollars uh on your balance sheet arguably if you're providing effectively all of your stock, and then the cash that gets you to 20.

20:56You have this letter from TD. That's another 20. We're now at 40, but we're still off by, call it, 16. And the 20, as far as I understand, while it's considered a highly confident letter, meaning TD's saying they're highly confident that they would provide the financing. It's not locked financing. Yeah, we'll see what happens.

21:26Founder, no. Never doubt. I hear you. I understand that. I'm just trying to understand where the rest of the money would come from. It's half cash, half stock.

21:44I hear you. I'm just saying that that math doesn't get you to the price that you're offering.

21:55Ryan, that's a pretty straightforward question. I don't get it. Where's the rest of the money coming from? Andrew laid it out pretty clearly. 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 the offer are on our website.

22:18John Coogan:But you're on our air. We thought we'd get... 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. Do you think he was expecting this to go out on Sunday, Monday, GameStop stocks pop like crazy? Yeah. And it's actually down today. I guess that's possible. I don't know. 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.