On Mon, Jul 27, 2026 at 11:12:35AM -0500, Joerg Jaspert wrote:
>
> As one of DAMs I dare to say that forbidding this type of usage is simply
> not enforcable. How should we (or anyone) be able to know an LLM got used in
> such a way? It is entirely unrealistic. Sure, there is this argument about
> the social parts about rules and lying to your peers on what you did. But
> trying to enforce a "you used an LLM to help you write mail" does not work.
> (Examples of overly useless long mails with bullshit content exists, I know.
> Those aren't what this is about. Such an LLM vomit is easy to refuse now
> already even without the rules here.)
I think it is a very bad policy to define something which can not be
enforced as part of the CoC. If we want to have position statement
about how Debian feels about any use of an LLM to write e-mail, that's
one thing. But to say that someone should be potentially kicked out
of Debian or forbidden to participate in mailing lists because of a
suspicion that they might have used an LLM seems to be... unwide and
unfortunate.
Even in the latest amended proposal, where we ask the moderators to
enforce this at their discretion, seems to be unwise from a project
governance perspective.
I will note that some people have been able to write very long
messages without using AI, and it's been pointed out to those people
that writing human-generated "slop" can in the end be self-defeating.
- Ted
