Comment by nz

Not contradicting this (I am sure it's true), but why is using an LLM for this qualitatively better than using an actual fuzzer?