zlacker

I don't really get the sentiment in this thread. There is an enormous chip shortage and we are just about getting started with useful AI. The chips are not optional and his or rather OpenAIs current mission hinges on their availability.

If Sama feels the current players are not doing enough, it seems entirely in line with his publicly stated and often repeated goal of advancing AI, that he would try and move things along. That's just extremely consequential execution.

>>jstumm+(OP)
Using OpenAI’s name to get funding for a startup that would profit by selling to OpenAI is not fair game

>>jstumm+(OP)
The supply of chips is limited by foundry capacity, starting a new fabless chip designer would do nothing to fix that bottleneck. A new chip designer would also not be releasing a competitive product for at least 5 years, even if done by the most talented and experienced engineering team of the generation (rather than by a career bullshitter like Altman).

This would most likely have been a pure grift play, finding some stupid money and using his position at OpenAI to convince them to invest in a doomed proposition.

>>jsnell+e9
> A new chip designer would also not be releasing a competitive product for at least 5 years

Best time to start was 10 years ago…

>>foogaz+zi
Sure, but there must be dozens of credible AI chip startups in various parts of their lifecycle started in the last decade. If the goal really was, as originally proposed, to solve the chip bottleneck for OpenAI you'd be better of backing one of those. And if literally all of these startups are on the wrong path and OpenAI beliefs a different kind of design philosophy would work better, I'm sure most if not all of those startups would be delighted to know about it, and quickly change their plans.

Starting from scratch would make sense if Altman personally were some kind of chip design savant with unique insight to the problem domain, and he definitely isn't that. Alternatively it makes sense if had nothing to do with saving OpenAI from the compute bottleneck, and was another one of Altman's Worldcoin-like grifter schemes.

>>jstumm+(OP)
Anyone with a passing interest in designing an AI Chip should listen to the last Latent Space pod with Dylan Patel of SemiAnalysis.

You either assume transformers are here to stay and somehow get an edge over Nvidia and others trying to do the same or you make a huge variety of bets on whats next.

https://www.latent.space/p/semianalysis#details