Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

The rationale here doesn't seem very compelling. In exchange for being so inefficient, evolutionary methods are good at some things, like non-differentiable losses (eg. novelty search where 'new' would be hard to define in any kind of differentiable way) or highly rugged deceptive fitness landscapes etc. But these are just LLMs on ordinary benchmarks?


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: