Hi, On 27/07/26 at 18:09 +0200, Johannes Schauer Marin Rodrigues wrote: > Yes. The difference is, that in this case, it would be something which Debian > itself and some of its contributors endorse which would make me spend my time > on other stuff and I think that is an important distinction. > > We are working together, sharing a hobby, collaborating in a common club. It > is > important that we more or less get along in this club and that we feel good > when we work on something together. We should not feel that our contributions > are meaningless because of how others in the same club choose to work. > > As I understand others in this thread, current LLM offerings are more than > capable and would have a very easy time automating a lot of actual packaging > activities. Others in this thread have expressed fear that Debian would start > dying a slow death if we forbid LLMs but would it not be the opposite? I'm not > here to crank out as many packages as possible but to enjoy the work itself > and > the interactions with others surrounding it.
I believe that there are many different motivations to contribute to Debian. Some (many!) do it because of the community. Some do it because they like the kind of activities they do in Debian (the joy of the work). Some do it because it's a way to contribute to something great. Some do it because it's a way to interact with cool new technology. You might not be here to crank out as many packages as possible, but some people might be interested in the efficiency gains that AI brings, because they think they could do more for Debian using it. And even if you think that those efficiency gains are just promises that will turn false, some people would be frustrated not to be able to experiment with AI in the context of Debian. I really don't think that Debian would be dying if we accepted AI-assisted contributions. On the contrary, I think that AI can help busy contributors focus on where they really add value, and semi-automate more tedious tasks. If Debian banned AI-assisted contributions, it would feel a bit like a retro-computing project, or a project around an esoteric programming language such as brainfuck: I would totally understand why some people find it fun to work on this, but that's not something I would personally like to spend much time on. > And for many things which are > tedious and repetitive we should feel encouraged to look at improving our > tooling instead of throwing the slop machine at the problem. (I think) What about using AI to improve our tooling? Two recent examples: 1/ I recently worked on debaudit.debian.net[1], a service to verify the reproducibility of source packages. Then I added a reproducibility dashboard to UDD[2]. Both of those developments were largely AI-assisted. I would not have attempted to create debaudit without AI, because the work needed largely exceeded the time I could have spent on it. 2/ Several AI-assisted MRs[3] were submitted to improve the performance of lintian on large packages. While they require more work, I think that they are useful examples of how using AI to analyze our tooling can lead to interesting findings. [1] https://debaudit.debian.net/ https://lists.debian.org/debian-devel-announce/2026/03/msg00004.html [2] https://udd.debian.org/reproducibility/ [3] https://salsa.debian.org/lintian/lintian/-/merge_requests/769 https://salsa.debian.org/lintian/lintian/-/merge_requests/770 https://salsa.debian.org/lintian/lintian/-/merge_requests/771 Lucas
signature.asc
Description: PGP signature
