Ihor Radchenko <[email protected]> writes: > "Dr. Arne Babenhauserheide" <[email protected]> writes: >> Or if people start saying “let AI do a pre-review” -- that just means to >> force contributors to read AI output. If I as reviewer don’t want to >> read unchecked AI output, I shouldn’t force contributors to read such >> either. > > What about pre-review only for LLM-generated code? (That's what I did > for John Wiegley's patch).
I consider that the job of the contributor. If you LLM-review an LLM contribution, you are coaching the contributor on the use of LLMs. That’s OK, I think. Ideally the contributor then learns to use the LLM better -- and then takes the time to actually understand the code. Because it doesn’t remove the requirement from contributors to understand what they contribute. In Hyphanet we now have the rule: "* ensure that you understand the change you suggest. You must not suggest code that you know you don’t understand." > An alternative could be providing LLM usage guidelines, but that may be > too much. I think that boils down to the same -- but giving such guidelines as official project guidelines could imply that the project encourages contributing code that was only ever written and read by an LLM. So maybe that should stay outside the project space itself. Best wishes, Arne -- Unpolitisch sein heißt politisch sein, ohne es zu merken. https://www.draketo.de
signature.asc
Description: PGP signature
