Comment by bla3

Mirrors my experience: LLMs are really good at optimizing, better than most humans. But also, they tend to not reach absolute peak performance where people made an effort to optimize something.

Since most problems see fairly little optimization, that's still a big win most of the time.