Hi Maxim, Maxim Cournoyer <[email protected]> skribis:
> I think the Motivations section probably needs to be toned down, as Right, please see what I pushed just yesterday. https://codeberg.org/guix/guix-consensus-documents/pulls/13/files > I've given it a new read. I still stand behind my earlier comments > about point 1 [0]. > > About the first (1) pledge point: I disagree about blanket pledging not > using any of the LLM tools. As others have found in our community, they > can have a use. I'd suggest dropping this point of the pledge. There’s “no blanket pledging not using any of the LLM tools”. On the contrary, we have change the wording not only to allow for some uses, but to explicitly list exploratory analysis and non-legally-significant contributions as accepted uses. > I'm not sure why we need both a set of commitments and a policy; it > seems the later could be written to cover the actionable commitments? > I'd prefer having just a policy, for clarity and simplicity. The commitments set a general direction for the project. The policy determines what contributions are acceptable. I wasn’t entirely sure whether that split was needed, but I felt like they are on different levels: the policy is directly applicable rules that one can look at when assessing contributions, and the commitments says where we’re going today as a project. > It seems the most disagreeable points are also in the commitments, such > as point 1, which appears to discourage any use of LLM for any Guix > contributors (only to mention later in the policy that perhaps that's OK > as an "exploration" or for submitting non-creative changes). Commitment #1 is about “the project”; the policy is about any contribution we receive. There’s overlap (people in “the project” are also contributors), but it’s not quite the same thing. I’m happy to consider rewrite that would raise any ambiguity that you see here. > --8<---------------cut here---------------start------------->8--- > 1. GenAI use disclosure. Contributors using genAI MUST disclose its use and > how they used it. > --8<---------------cut here---------------end--------------->8--- > > I think if we want to keep this point it should specify how (an > Assisted-by: git trailer? that'd be a kind of advertisement, Exactly, that’s why I explained earlier that I would not want such a tag. I suggested something like ‘LLM-Assistance: debugging’, but I think just a plain text description is fine. We can always improve on that later if we feel a need for it. Disclosure of genAI was added so point #2 of the policy can be exercised. > Point 4. of the policy is again trying to tell us what it's OK for us to > do with LLMs; similar to the first commitment; which I think is > unwelcome and unnecessary. The critical policy points to me appear to > be 2. and 3. If we could focus on these alone, I think we'd probably be > in a good position to find consensus. Earlier several people, such as Greg Hogan, suggested that exploratory analysis be explicitly listed as acceptable genAI use. I think there is value in being explicit, though I also like the simplicity of “what’s not forbidden is allowed”. Either way, it doesn’t change the substance of the policy IMO. > While I will be away myself too during that period, I doubt the reduced > time to discuss this difficult topic will help converging toward a > final, agreeable document. Perhaps, the GCD should be withdrawn and > resubmitted at a later date, e.g. in September when people are back from > vacation and re-energized? It will have been 7 weeks of intense discussion and work. I’m willing to go to the deliberation. Worst case, the GCD doesn’t pass and someone else picks up the torch later. But after all these efforts, I would hope that most major concerns have been addressed by then. I’m calling for everyone to be specific about what would prevent them from accepting the text in its current form so we can try and address those together by then. Thanks for your feedback! Ludo’.
