00:23:04 · Aditya Agarwal, Elad Gil
[00:23:04] S1: That's interesting. I mean, you could argue that whatever baseline speed you are starting off, a startup should be able to accelerate it at a far fast start exponent. So you should have technically an advantage in a world where you can crack out a lot of code. I I would argue it's probably still harder for Microsoft or Google to generate a bunch of... But you never know. I get it. They're fast moving. They're also a good company. [00:23:27] S0: Yeah. They're good companies, and they have that distribution. And, traditionally, they've been... You know, probably the biggest hectic for them is they end that five to seven years, and maybe they have two or three years. Right? But, there is time. [00:23:37] S1: There is time. Yeah. Yeah. What is the most common mistake that you see AI startups make in our legs? Or maybe even call it startups in... Like, is there a Blah. Is there a common failure mode that is emergent based on the environment that you're in right now?
00:23:52 · Elad Gil
[00:23:52] S0: You know, I think there's two types. There's three types of failure mode. I think one type of failure mode is your thing isn't working and you stick with it too long. And it's back to, like, you should eventually iterate into something really working. And if it's taking two years and nothing's working, you should you should rethink. In most cases, if not all. Again, there's counterexamples. The second thing is there's really bad advice given to things that are working that are working for people who've never had anything work. So what used to happen is you'd get advice from founders who'd seen product market fit to founders without it, and they say go hire a sales team and scale really fast and burn a lot of money into it, and that was terrible advice. But it goes the other way. If something's really working and people tell you to buy, stay as lean as possible. They'll hire executives. Don't scale. That is awful advice. And that's where a lot of companies break for a while and other people, competitors who are gonna scale come in. And it just... It's it's, a very common pattern for things that are working that founders don't actually build out their company as a product, and they don't really go for it in an aggressive enough manner. So I think that's a big failure mode. And another failure mode is spending money on stuff that just doesn't matter. You know, you start training a crazy model on something instead of just testing something sheer with an existing, like, soda lab and just seeing if people want it. You know?
00:25:09 · Aditya Agarwal, Elad Gil
[00:25:09] S1: Maybe for that last one, I'm curious. I mean, you're deploying large amounts of capital often to a pretty early stage companies. I guess, for a lot of companies that are raising 100, 200 pretty early on in their, I would say, life cycle, have you found that they actually know how to use it well? It's kind of hard. You know? Like, if you're like,. I mean, maybe you're training the models. Maybe it's all kind of, like, ultimately That's that's I [00:25:35] S0: actually have not funded many things like that. [00:25:37] S1: Okay. I mean, so I suspect that we have the same priors because I think it's quite hard to spend a 100... Like, you know, if you raise a $100,000,000 as a seed round, you can't really pay yourself a $100,000,000. You know? You know, as a founder, you have I mean, you could. Anybody who wants to [00:25:53] S0: see how he ran all his way to her face.
00:25:53 · Elad Gil, Aditya Agarwal
[00:25:53] S0: see how he ran all his way to her face. [00:25:56] S1: You gotta see it in the back. Don't see variation in the back. Some salary. Yeah. But it's it's quite hard because your range of what you can pay yourself is pretty bounded, I think, in terms of what is accepted in Silicon Valley. There's only a certain way in which you can hire and create people. It's not easy to hire well and quake. So often, if you raise, you know, a 100 at 500, I tell people that you can do it, but it requires everything to go right, like, for 12 to 18 months. And if you had any hiccups along the way, it's really hard to zig and zag the company, which is actually opposite of what you'd expect. As most people are like, boy, raised a bunch of money. Now I can, you know, afford to get it wrong. But I actually think you can... It's it's much harder to get things wrong. You might have the money, but you don't actually have room to face instead of my ding. Yeah.
00:26:43 · Elad Gil
[00:26:43] S0: Yeah. I think you're you're setting different expectations for yourself. I do think there are circumstances where you raise a 100 of my dollars and you bill, you know, like what Pi is doing, you know, on the robotics side or building the foundation models for robotics. Like, that needed a lot of money because it needed a lot of computing, did a lot of training to next. You know? Great. There's a rational... A rationale behind it. I I just think it's tough depending on what you're doing, and so it really has to be tailored against what are you actually trying to accomplish and what proportion of it is GPU versus headcount? And often you see people raises the grounds, and, you know, 80% of it is supposed to go to GPU or something, and the other 20% is headcount. So you you see these differential ratios versus I'm raising a 100 just to have people, which I think is a a tougher thing for a very early company.
00:27:28 · Aditya Agarwal
[00:27:28] S1: And, Elad, when you came here last, we're gonna switch topics a little bit. It's a lot of AI, so we're gonna switch topics a little bit. We'll come back to the AI. Don't worry. I think when you were here last, crypto and Web three was still pre the sadness, pre the winter. And I think it's actually kind of, you know, we're starting to see some interesting companies come about again. So I guess trying to... Where are you in terms of, would say, whether it be crypto rails, distributed fan of fi transactions, in terms of are you doing any investments there? And there's obviously a lot of talk around how agents might end up being one of the key catalysts to essentially allowing you know, kind of like the transaction rails that only... Utilizing the transaction rails that only exist kind of at Cripple and Web three. Are you a bull there? Like, do you agree?