Le Tue, Jul 28, 2026 at 08:00:52AM -0700, Russ Allbery a écrit :
I think this proposal is in effect a stealth "do nothing" option with regardes to the issues the other options are trying to address. That seems like a failure to be explicit about its impact.
[…]
This seems fixable, though, by adding an explicit statement to this option
Hi Russ, thank you for your comments. Yes, it is exactly my intent to provide a "do nothing" option, which I explained in my previous emails, so you are totally right about the "failure to be explicit" given the feedback that I had. I originally thouhgt it is better to have a ballot as lean as possible, but you are correct. Here is a revised version. ---- Draft ballot: Call the Free Software community to write Free AI ---- We are concerned that using the General Resolution process to establish rules and red lines on the use of generative AI in Debian will only result in increasing the total number of demotivated or quitting developers. We explicitely decide to not enforce new hard rules on that topic, and instead issue the following non-binding statement, that reflects our opinion that we can make a better future if we put enough energy in the right direction. We the Debian project, are a community that produces a Free operating system by integrating Free software core components as well as thousands of applications and accessories. Respect and diversity are among our core values. We affirm that: * AI models should not be used against the authors of their training data. * The output of AI models trained on software that requires that derivatives must be distributed under the same terms should follow this requirement. * The training data of AI models should be broadly accessible for study, audit, reproduction and modification. It has been said by some that this is impossible; we believe it to be wrong. We call to the Free Software community to write such AI models that follow our core values and can be integrated in our operating system to empower our users to uses of this technology while respecting Free Software, privacy, diversity and environment. ---- End draft ballot ---- To Adrian, I added the word "non-binding" to clarify that no decision is made. If you think that the wording is defective and dangerous, but like the idea of deciding not to decide, please propose an improvement. Otherwise, you can of course rank this option below NOTA, or invest further energy campaigning against. After all, this is exactly what I am doing about the other ballots. Regarning the non-DFSG freeness of software that does not allow AI training, it is not a problem for us because it will not be in Debian. On my side, I would welcome an AI-aware copyleft license that states that the output of a model trained with this input should be under the same license, and hope that we would accept it as DFSG-Free. Regarding right to fork, people are free to fork without AI or with an AI that is not trained with the software to be forked (one could imaging training two large models with two training datasets of similar but non-overlapping content). Also, traditional forking does not erase the recognition and rights of the original authors, while using a Gen AI trained on the work to rewrite the work does so, which I find quite disgusting. This is more than just the right to fork, this is a tool to erase people. Ian, I hope the above also answers your questions: why point 1 is not just "against the wishes" but is "against the authors", that point 2 is about rejoicing to the future possibility to have DFSG-free LLMs that produce output under a DFSG-free license, instead of the current "bad LLMs" that produces output that the user can copyright and license under the terms they want. I do not want of a GR that invites the vast majority of the developpers who did not participate to this discussion (have you noted that we are always the same small group of people talking, regardles the topic?), that invites them to decide which of the left or right side of the project is going to be pushed overboard. People talk about quitting in this thread; I think that less people will quit if we do not vote, vote NOTA, or decide not to decide. But of course I can be wrong. Please let me conclude by saying that in the past ~20 years I have participated to many heated discussion, and I eventually noticed that given enough time, there will always be the case that I happen to side with someone I opposed in another topic, and vice-versa. I see this as the sign of a healthy, diverse and resilient community. This is what I would like to preserve. Have a nice day, Charles -- Charles Plessy Nagahama, Yomitan, Okinawa, Japan Debian Med packaging team http://www.debian.org/devel/debian-med Tooting from work, https://fediscience.org/@charles_plessy Tooting from home, https://framapiaf.org/@charles_plessy
signature.asc
Description: PGP signature
