Altman says he was too cautious on compute investment

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

1. 48:04

   I mean, I definitely badly undershot on the compute investments. Could you have known going in with the right mental model? Yes, but I got like psyched out by the financial markets or something. I don't know. That was clearly a mistake. How do you correct that in the future? Well, I mean, hopefully I'll learn from it and I'll make some new mistakes, but I won't make that one again. on the compute side this is kind of like the biggest infrastructure project of maybe all time or is about to be yeah are you going to try and like vertically integrate under open ai's hood everything from like power generation to like token generation to you know serving some person in chat gpt or codex or something like that or are you going to try and have a bunch of partners we're not literally going to try to do it all inside of open ai obviously but we will try to

2. 48:50

   do a better job of a well functioning supply chain than we have done so far like you know we have brought chip design and model design together and that's been good i don't think we need to bring like electron production together in the same way because that's more of a commodity uh however i am when you really think about the whole supply chain that has to come together to produce intelligence i would say there is relatively too much focus on algorithms that create better algorithms and not enough focus on data centers that can create more data centers which in a world of robots and a truly automated supply chain you can totally imagine doing you can like spend a data center's thinking power to drive a fleet of robots to make more copies of the data center and that's probably a very wonderful thing to do uh relative to other uses

3. 49:53

   of that compute if we're right about what that additional compute will eventually unlock how much of your time is spent thinking about how to like successfully design that fully robot end-to-end future where intelligence almost no time spent thinking about almost all goes into like execution like the idea that's a very obvious idea it's very easy you can like kind of relatively quickly say here's the pieces that need to come together but then to like get that whole machinery and all this company is working together um that that you know none of the glory of the big idea and trying to like think the big thoughts but like a lot of grinding when you think of like execution what does that typically mean for you not there's no typical there like the