1. This is a kind of fuzzer. In general it's just great to have many different fuzzers that work in different ways, to get more coverage.

2. I wouldn't say LLMs are "better" than other fuzzers. Someone would need to measure findings/cost for that. But many LLMs do work at a higher level than most fuzzers, as they can generate plausible-looking source code.