>You could view LLMs as a super-charged mutation function inside a genetic programming harness.

There's probably something there, though im not sure what. I'm thinking about LLMs token-filling and the 4 amino bases