On Sat, Feb 28, 2026 at 9:17 AM Ekaitz Zarraga <[email protected]> wrote: > > On 2026-02-28 12:44, [email protected] wrote: > > Ludovic Courtès <[email protected]> writes: > > > >> One thing I wonder is whether the project should have an “opinion” on > >> generative AI in general and its relation to computer programming and > >> human creativity. I do have an opinion :-) and the question is whether > >> taking a stance as a project is worth it and feasible. > > > > guix has already taken a stand on a number of issues, which, while > > easier to police, i find, personally, far less problematic than the use > > of llms. > > > > guix is a free software project, first and foremost, and that has been > > made clear on its hardline stance against firmware blobs. to my > > understanding, this is done for the explicit purpose of encouraging user > > freedom, even if it comes with some pain, because freedom is > > paramount. any criticism i have of certain policies is through that > > lens, and any criticism should first be made through that lens. > > > > "does this promote and extend freedom" should be the first test before > > adoption of anything. in most cases the answers are pretty clear. in a > > few they are not. > > > > but llms are not just not free, they are anti-free. they are in direct > > opposition with freedom. as such, guix should, in my opinion take a very > > aggressive stand against them. > > > > i will refrain from enumerating the incredible, manifold ways that llms > > take away freedom in this email, in order to not derail this into > > arguing back and forth on specific points, because, to me, the important > > question is only this: "will this make us more free?" > > > > llms are about as black-and-white an answer to this issue as i can > > imagine, and the answer is a resounding "no." so guix should reject them > > with as much force as it can. > > > > -bjc > Your fantastic argument, which in the root I agree with, has only one > flaw in my opinion. > > Guix can only decide on the things it does. The stance on binary blobs > is that Guix does not share them. It does not prevent the users from > using them (in fact, I do use binary blobs). That is also freedom, the > freedom for people to take their own decisions, regardless if they are > good or bad. > > As I said before, from the Guix side, I could have been using LLMs for a > while and Guix wouldn't notice (and I made a few commits!).
LLM software generates code, and the license of this generated code is not certain. Treating an LLM as a compiler, it receives two inputs: - A description written by the user - A truckload of data, usually licensed source code What is the license of the output? > > The LLM does not demand anything from Guix like the "binary blob" stance > you mention. They might make the review process harder or other things > we can discuss, but that's not Guix itself having to change to > accommodate or include them. It's not the actions of Guix we are > discussing, but the actions of individuals that we are discussing if we > limit or we don't. > > That is, really, the most "anti-free" action one could think of. > > In practical terms, I would just don't do anything and wait to see if > LLMs are actually a concern in the project. They might never be (mostly > because those who take part in GNU projects do care about software and > freedom), and we are here discussing for nothing. > > I know we all are super smart and our ethics are better than anybody > else's, but still. > > Maybe, if we want to actually do something about LLMs, maybe we should > take a look to our actions first and how we encourage training of LLMs > with our actions. I'd say that conversation is in scope. Whatever a > person does in the intimacy of their computer is none of my business. >
