Kyle Corbitt: evaluation

Transcript context

…That's, yeah, it's fascinating. This may be hard to summarize and I don't know if anybody, you know, has enough outside of the labs, I guess would have enough information to really characterize this, but like, Is this a good business to be in? I can see it kind of going either way. I would assume if you've got a good environment, all the labs want to buy it. But then at the same time, they're buying a ton of stuff. How much does your one random thing add to the whole mess of things they already have? And also, it's depreciating, as you said, for you, right? So you've got to strike a deal. before they already saturate your thing and then truly don't need it anymore. Would you say this is a hot, good place for up and comers to go, or would you steer people away from it? It's clearly a good business in the sense that these companies are scaling to tens or hundreds of millions of dollars in revenue in months. But your question is, would I steer someone into it, to founding one of these companies? I think it's working out quite well for them. I've been asked to invest as an angel in a number of these, which I have declined to do. I have a hard time seeing them as a durable, long-term, venture-shaped business. I think they're potentially really, really good businesses for the founders if they don't take capital and just kind of like take the profits while they're good. Yeah, like, I don't know, like, at the same time, I'm kind of like on the record is like being very skeptical of the human data labeling business, which is sort of like the prior thing. And we have multiple, you know, decacorn style exits, or at least valuations on human data labeling. So, you know, I may just be like miscalibrated on, you know, how durable the demand is for these things. But yeah, I guess my short answer is I have not invested in any of them. Yeah, interesting. That makes a lot of sense. I mean, a lot of things are like that. I feel like in AI, there's a lot of kind of fleeting, maybe great cash grabs while they exist. But every next generation of the model puts a lot of those things kind of not necessarily out of business, but like certainly makes them a lot less exciting than they used to be. On that data labeling point, how do you think about I was recently listening to Dylan from Semi Analysis talking to Dwarkesh, and there's kind of one world where compute is abundant and especially, you know, it's going exponential, but maybe that'll be abundant enough, maybe it won't. But if compute is there, then maybe we don't need much human data labeling anymore because we can just RL the hell out of everything. You know, who needs to pay humans hundreds of dollars an hour when, you know, you can get obviously millions of tokens for less. So that's one theory is that like we just won't need that much human data anymore. Then another story would be like, well, compute is so scarce and like, you know, I did check the prices of even A100s, you know, these days are like higher than they were last time I checked. So these things are not depreciating in the traditional sense. So maybe if supervised fine tuning or even, you know, abstracting away from technique, if like human data can save you compute and compute is the binding constraint and you have all the money in the world, then maybe the human data industry continues to go strong because even if it's like sort of an inferior good, there's just not enough. compute to drive what people would, you know, they can't spend as much on compute as they would like. How do you think about like where we are in that story and maybe where we will be as we go ahead?…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.