On 6/19/26 09:56, Simon Tournier wrote:
Hi Ludo,
On Thu, 18 Jun 2026 at 22:26, Ludovic Courtès <[email protected]> wrote:
IMHO, the rule must be explicit: “Namely, the use of genAI means …“ or
something along this. It appears to me a bad wording to rely on “does
not prevent for X”. Does it imply that it prevents for Y?
Yes, I agree, it is currently still too unclear.
Related, I mainly feel that even with these "not prevents", the GCD
shuts the door for many experiments where genAI might actually help our
humanist project.
The availability and power of Apertus, and that even RMS considers it
free enough, really changed my mind about being willing to accept the
pledge.
There will be future LLMs that are even more free than Apertus, and I
would like to see how we can use them. Maybe we can't, but it is too
early to tell and shut all of that down.
Somehow, “setting a standard for what we do collectively within the
project“ is mainly about the output of LLMs – it’s concrete and
actionable –, hence the word “pasting” and also what I also propose [1]
“never use genAI for directly or automatically interacting with the
project”.
That sentence was the most difficult for me to understand in your
reformulation. Because if someone would use genAI to as questions about
the code; I would see that as 'interacting'. As in 'having the LLM act
in between me and the project'. But you mean providing information to
other people (as in, code, pull requests, emails). Thank you for
clarifying. I agree with your sentiment.
Well, there is not much of a room for clarifying the standard
and what we do collectively about “use of GenAI” without, IMHO, slipping
toward some “moral superiority” or some “virtue signaling”.
Some of that 'higher ground' is even fine.
Like with being strict about free software. Guix enticed me to ensure I
can run a libre kernel. (Shoutout to Technoethical and Think Penguin.)
And I'm willing to pledge that.
But I don't feel the same about genAI; I believe using (libre) genAI can
be fine; that it is possible to account for the externalities and use
them ethically. (Unlike e.g. flying or having cattle, that is only
ethical in very extreme cases, in my opinion.)
It takes effort to use genAI ethically, and I'm willing to pledge making
that effort. Most of my earlier experiments would not reach the bar I
think we should set.
That is, I would pledge to "only use genAI under these strict
conditions", but feel that pledging "not using genAI except for these
narrow cases" would be dishonest (when coming from me).
Note that these conditions can also be more strict than the current
pledge: e.g. if we use genAI because the documentation is bad, we should
also improve the documentation. The GCD as-is wouldn't care.
And we have plenty of 'strict conditions' going around here; I think all
objections can be rephrased into a positive condition that I would
easily agree to.
Hugo
