Hi Phong,
On Wed, 04 Mar 2026 at 16:25, Nguyễn Gia Phong via "Development of GNU Guix and
the GNU System distribution." <[email protected]> wrote:
> I tried to be really specific in phrasing when I started this thread
> to avoid this topic; I wanted a discussion about incorporating
> non-trivial snippets from LLMs into a codebase, not other usages.
> It's a very specific action, not a lifestyle or filosofi.
Sorry, I’m not able to ask something to upstream when I’m not able to
think how the same applies to us. And it appears to me more
approachable to first think about how do we deal with contributions to
Guix itself before shooting strong conclusions about upstream.
Especially when the main color is grey.
> In my mind Guix has two keys properties: a complete dedication
> to free software, and reproducibility. This dedication
> is why I contribute to the project, and I assume that reason is shared
> by many other contributors.
I share something similar. The point of my previous message is: these
two key properties are hard to encode at the level of PRs, IMHO.
Therefore, I’m proposing another level.
For the rest of your message, I agree, more or less. :-) Where I’m less
convinced is about where your proposal leads, but maybe, I overlook
something.
Cheers,
simon
PS: IIUC the reasoning, you wrote that “free software” is important and
because LLM would appear to us an issue about the license, thus
hurting the legal framework, then the project should refuse to
package upstream source code that relies on LLM.
Far from me the willing to be picky here, applying the same
reasoning: “reproducibility” is also important and please note that
(1) packages that are notoriously unreproducible are packaged and
(2) non-fully bootstrapped programming languages as Haskell or OCaml
are packaged although they weaken their reproducibility. Therefore,
does it mean we should also refuse to package all these
unreproducible and/or non-bootstrapped source code?
Again, maybe I understand or overlook something.
