💀👻🎃🦇MOSS 666🦇🎃👻💀 on Nostr: I focus on using the most efficient models available, because by definition they are ...

I focus on using the most efficient models available, because by definition they are using less valuable power and water per million tokens. DeepSeek has done remarkable work making the same hardware support more users -- v1 could do 90 sessions on a single server, v4 does over 400 simultaneous users with a bigger, better model. And there are still efficiency gains to be had, according to the latest research papers.