Yes, I've found some really interesting bugs using LLM feedback, but it's about a 40% accuracy rate, mostly when it's highlighting things that are noncritical (for example, we don't need to worry about portability in a single architecture app that runs on a specific OS)