This experiment is poorly executed:

1) Don't commit the llm code change and then nitpick it. It's always super fucked up. Do the back and forth offline and only commit working code that has passed the bar of an experienced human reviewer.

2) Don't make such large changes. The llm can't even fit the whole diff in its context window. You're asking for trouble. Keep each change small and focused.

3) Just close the PR when the whole internet shows up in the comment section. Probably the only thing worse than letting the AI do your critical thinking is letting reddit do your critical thinking.