Re: Draft permissive ballot option (was: GR: Ban LLM contributions from Debian)

On 28/07/26 at 21:39 +0200, Marc Haber wrote:
> Hi,
> 
> thank you for your reasoning.
> 
> On Mon, Jul 27, 2026 at 08:14:53AM +0200, Lucas Nussbaum wrote:
> > 1/ I believe that it's important for Debian to state that
> > LLMs raise many concerns (and not just related to copyright). That's
> > what this paragraph is trying to achieve (and I would welcome
> > suggestions to extend it from people against LLMs):
> > 
> > > The Debian project recognizes that AI-assisted contributions raise
> > > many concerns, e.g. about the technical quality and maintainability of
> > > such contributions, and their legal status. AI itself also raises
> > > additional concerns, about its impact on society at large, on the IT
> > > industry and on Free Software; about its environmental impact; and the
> > > aggressive or non-compliant practices of AI scrapers.
> > 
> > 2/ I believe that a clear set of safeguards is useful to make LLM usage
> > by others less inacceptable for contributors who would prefer to reject
> > LLMs. Specifically, I think that disclosure is important because some
> > contributors might choose to apply different standards to AI-assisted
> > contributions (and that's fine).
> 
> I do not think that is fine, though. Why would I put a sound suggestion on
> the bottom of the pile or bin it just because it was written with the wrong
> editor or using a mail service I don't like?

With "different standards", I was thinking of "read it but take it with
a different grain of salt", not "ignore it".

Errors in AI-assisted/generated content tend to be of a different class,
so as a reader or reviewer, it's useful to know the content was
AI-assisted, to use a different mindset during the review.

Also, it allows the reviewer to adjust the way to communicate with the
submitter: if the submitter says the code was LLM-generated, then one
can point a logical flaw and blame the LLM rather than point it and be
super-careful not to harm the submitter's feelings.

Also, I think there's a matter of honesty. When someone submits a
high-quality AI-generated piece of code, or a contribution that would
otherwise require a lot of work, it feels dishonest to take credit as if
it was purely human-generated.

Two examples here again:
- Getting to the AI-generated bug report in #1142332 probably took me
  between 10 and 20 minutes. If I had not had AI-assistance, I would
  probably have skipped the detailed analysis and just reported the hang.
- Generating discussion summaries as in
  https://lists.debian.org/debian-vote/2026/07/msg00017.html would take
  hours of manual work, vs minutes with an LLM.

Lucas