Perhaps LLM-guided Superoptimization: <https://en.wikipedia.org/wiki/Superoptimization>

I recall reading about a stochastic one years ago: <https://github.com/StanfordPL/stoke>