Post
No, I think that's still true, but lots of people are selling inference while not doing training, and that shouldn't be forgotten. And OAI/Anthropic's training budgets, tbh, don't really interest me now either -- we are at the stage where model dev could 100% stop and we would be Just Fine, so, eh.
0 reposts27 likes
This is quite long but addresses a lot of the cost part as well as anyone outside the labs Dylan Patel – Anthropic & OpenAI will have most of the world’s compute by 2028 — Dwarkesh Podcast hopper.webdirections.org/share/aaf663...
hopper.webdirections.orgDylan Patel – Anthropic & OpenAI will have most of the world’s compute by 2028Dwarkesh interviews Dylan Patel, founder of Semi Analysis, about the rapid growth and centralization of AI compute, projecting that Anthropic and OpenAI will control most of the world's usable compute...