Re: GR: Ban LLM contributions from Debian

Hi,

On Sat Jul 25, 2026 at 2:36 PM -03, Adrian Bunk wrote:
> The #1142332 example from Lucas is also quoting, so that's obviously 
> permitted?
>
> If not, why would it make a difference when quoting LLM output that
> had been pasted to an upstream location?

I think the difference is between quoting a message posted by another 
person, and text generated by a language model prompted by myself.

> I do also remember a certain "editorial amendments" GR where it was not
> clear to voters what was actually voted on until after the vote.

It's nice to have more clarity before voting, thanks for the questions :)

> "messages to humans must be drafted solely by humans" in the CoC means
> that using nmudiff is no longer permitted, and "without LLM assistance"
> in the CoC means that a spellchecker that uses AI must not be used.
> The former seems unintended, and the latter is on the thin line
> between unavoidable and disciplinary action.

As the context is rejecting LLMs, I guess the output of 
non-language-model programs, like nmudiff, is fine. The spellchecker 
concern is trickier, as it depends on what you consider a spellchecker.  
While I wouldn't consider problematic the usage of a program which 
corrects clear errors like "helol" -> "hello" and happens to use an LLM 
to do so, text editors which have "rewriting" features like [1] may be 
more problematic, yes.

[1]: https://support.apple.com/en-gb/guide/mac-help/mchldcd6c260/26/mac/26

> "blog posts on Planet Debian must be drafted without LLM assistance"
> doesn't leave much space to post about how to use LLMs.
> Does the GR intend to ban posts about LLM usage, or should this stay 
> permitted?

I would say they would be permitted, see below.

> Does "We request that everyone, even outside Debian, should refrain from 
> using this technology." also imply that LLM packages can no longer be 
> maintained in Debian?

The "REQUESTS" section are not requirements. Also, I think that 
development of language models, when done in ways which avoid the issues 
in the "BACKGROUND" section, is not bad per se. It is also acknowledged 
that today ethical use of this tech is hard, but that should not forbid 
working towards improving the situation.

(note: I'm not advocating for "ethical" chatbots, but I believe that 
research for narrower use of the underlying technology is not evil; but 
that's not what "big AI" companies are trying to do anyway)

> It is puzzling that such a strongly worded anti-LLM GR does not mention 
> LLM packages in Debian - this is an area that is under our control, and 
> it should be clear to every voter reading the GR text whether it makes 
> any decision on this.

I think that it clearly does not disallow that, even if the wording 
criticizes the technology.

Note that this is just my opinion and interpretation of the text. I did 
not write it and do not expect to know the answers to your questions.

Bye!