Re: About SWH, let avoid the wrong discussion

[Andreas Enge <andreas@enge.fr>]

Am Fri, Jun 21, 2024 at 12:12:13PM +0300 schrieb MSavoritias:
> and as I mention in my first email I want to apply social pressure and make it clear to package authors what is happening so we can move to an opt-in model.

Well, the opt-in model is in place: As soon as I put my code under a free
license on the Internet, I opt in for it to be harvested by SWH (and anybody
else, including non-friendly companies and state actors).

Now the code may not be found by SWH, and the moment someone makes a Guix
package out of it and adds it to the Guix main channel, SWH will find and
archive it; but the opt-in has happened before at the moment I put the code
online with its license.

Maybe I misunderstood to what you want to apply the term "opt-in" (after
reading your other message in which you use the term, this seems to be
the case). If it is to source code of packages being used for AI training,
there is actually no need to have a separate opt-in. Either it is legal
under your license (and then you have effectively opted in), or it is
illegal (in which case explicit opt-in already is a requirement).

Am Fri, Jun 21, 2024 at 11:14:18AM +0300 schrieb MSavoritias:
> Aside from that even Guix uploading all code from the packages to
> SWH that basically feeds it to a LLM model is indeed not honoring consent of the author of the package.

Guix does not upload code to SWH. It gives them a pointer to a public git
repository that SWH then harvests or not according to their rules (see my
reply to Dale yesterday). These are not the same things at all.

Whether or not one agrees with the SWH policy on LLM training (and I have
not looked at it well enough to form my opinion), I do not think there
is anything we should change at the level of the Guix project. Maybe SWH
should put into place an opt-in procedure for feeding LLM; but I do not
think we in Guix should put into place an opt-in procedure for informing
SWH of the source code we package. (Which would be completely ineffective
anyway: One single person in the world would be enough to run the code in
"guix lint -c archival" on all Guix packages in all channels they have
access to. For instance, SWH themselves.)

Andreas