Good LLM-generated code comes from good developers that can provide detailed requests, filter through wrong/misleading/cruft results, and test a final working copy. If your personal bar for PRs is high enough then it won't matter if you use AI or not because you won't contribute shitty results.

I think the problem has become people with no bar for quality submitting PRs that ruin a repo, and then getting mad when it's not merged because ChatGPT reviewed the patch already.