> But what do we mark? Packages that use LLMs aggressively? Packages > that accepted any LLM contributions? Packages that have commits that > look suspicious, but aren't annotated as such?
As mentioned deciding what to mark is going to have to be subjective regardless... That being the case, does it make sense to leave packages/versions chosen as less reliable due to AI with the testing keywords (eg ~amd64)? It was my understanding that some human decision needs to be made to promote versions out of testing anyway, and users are already required to mark acceptance of testing packages via accept_keywords with the knowledge that they may be unstable. Naively that same system seems like it might be applicable here?
