Post

As an open-source maintainer, I'm not very bothered (yet) with PRs containing LLM-generated code. It's the ones with LLM-generated *descriptions* that bother me.

0 reposts7 likes

The tell is a description listing five improvements for a three line diff. A template field asking what you tested by hand filters those fast, because a model has nothing true to put there.

The description is what a reviewer reads first and what git blame shows a year later. A confident summary that does not match the diff is worse than none, because it primes the reader to approve the wrong thing.