Re: GR: Ban LLM contributions from Debian

Andrey Rakhmatullin <[email protected]> Wed, 22 Jul 2026 23:40:43 +0500
Newsgroups	gmane.linux.debian.devel.vote
Message-ID	<[email protected]>
On Wed, Jul 22, 2026 at 05:35:37PM +0200, Matthias Geiger wrote:
>    2. Quality  LLM output has many well-known problems with 
>	accuracy.[3][4][5] A LLM can never "know" if its output is correct 
>	since it merely produces syntactically likely combinations of the 
>	training data.
>   In some environments this is good enough. In Debian, it is not. For 
>instance, in packaging, each Debian source package is unique. Since 
>packaging syntax and best practices have changed over time,    a 
>LLM-produced package will have a mixture of contents spanning the age 
>of the archive, with watch files that do not work, overrides out of 
>context, imaginary copyright, 	and will generally be unfit for upload.    
>A seasoned Debian contributor with packaging expertise may find some 
>limited usefulness here, but a new contributor cannot, and would not 
>know how to fix it.    These same quality and accuracy concerns apply 
>clearly to all of the areas listed in the scope of this proposal 
>above.

All of this applies to human-written contributions even more, and modern 
good models are evidently good enough to write acceptable packaging:

https://github.com/TheJonaz/moraine-backup/tree/main/debian

https://github.com/Derrity/TinyServe/tree/main/debian

https://github.com/istax/eos/tree/main/debian

These are all in the better 50% of new packager contributions, the second 
one even reported that it ran `lintian -EvIL +pedantic` while many new 
human packagers don't run lintian at all...

"a mixture of contents spanning the age of the archive, with watch files 
that do not work, overrides out of context, imaginary copyright" applies 
to many existing packages, including fresh uploads by DDs, and not all DDs 
follow "packaging syntax and best practices have changed over time". I 
think we should instead ban low quality submissions.

>Debian intentionally grows this 
>community through many means, and new contributors are always 
>encouraged to join. 

*cough*

> Allowing LLM contributions breaks this.

You want to forbid "assistance of large language models" while it may be 
Debian's only chance to get new contributors in the absence of good new 
contributor docs.

>New 
>contributors submitting LLM output for review places an unnecessary 
>strain on the reviewer

Agreed. Though just like the problem with spam OSS repo PRs always existed 
and LLMs mostly increased quantity of that as opposed to qualitative 
changes, Debian always had a problem with low quality new contributor 
submissions (see 3 new packages I linked above, none of them deserves to 
be in Debian, and they would be proposed just the same 5 years ago, just 
with worse debian/), and I think the solution is the same as with spam 
PRs: reject the contribution instead of reviewing it, reject the 
contributor if they persist or are clearly bad actors.

> which can lead to burnout. 

Burnout of RFS reviewers is not only older than LLMs but older than RFSes...

I'm mostly talking about RFSes because I assume we don't have a noticeable 
number of new contributors submitting patches to internal services or 
someone else's packages (except trivial ones or in bad faith).

>Debian is not here to generate as much code as possible requiring 
>manual review by a shrinking number of human volunteers, or to package 
>every piece of software, or to rush new features, but these are what 
>LLMs are used for.

Agreed, and we should actively reject more RFSes instead of silently 
letting them rot, but LLMs are also used for debugging, reviewing, 
drafting plans and drafting patches, which are things taking most of our
volunteers' time.


-- 
WBR, wRAR
signature.asc (application/pgp-signature, 894 B)