Re: [RFCv3] Counter-Proposal -- Interpretation of DFSG on Artificial Intelligence (AI) Models

Simon Josefsson <[email protected]> Mon, 12 May 2025 08:56:32 +0200

Newsgroups	gmane.linux.debian.devel.vote
Message-ID	<[email protected]>

Thorsten Glaser <[email protected]> writes:

> Counter-Proposal -- Interpretation of DFSG on (AI) Models (v3)
> =========================================================

I don't know if further seconds are needed for each new version of a
proposal, but for clarity I'll second this version too.

>>I realized that I have one additional generic concern: You claim that
>>models are a derivate work of their training input.
>
> Yes. This is easily shown, for example by looking at how they work,
> https://explainextended.com/2023/12/31/happy-new-year-15/ explained
> this well, and in papers like “Extracting Training Data from ChatGPT”.
> It is a sort of lossy compression that has shown to be sufficiently
> un-lossy enough (urgs, forgive my lack of English) that recognisable
> “training data” can be recalled, and the operators’ “fix” was to add
> filters to the prompts, not to make it impossible, because they cannot.

I don't think this question is legally established or socially agreed
on, and I think it will be an area of conflict for many years.  I also
don't mind this text in your proposal because I happen to agree with it.
But I think it would be possible to disagree on that (and many people
do) and still agree with the rest of your proposal about what Debian
should do in this situation.

/Simon

signature.asc (application/pgp-signature, 1.2 KB)