Hi. Thanks for engaging in this discussion. Your proposal is
interesting and I think we're less far away ideologically than it
might appear.
I have some reservations about your proposal within this context,
though. I'll try to explain.
Charles Plessy writes ("Ballot option: Call the Free Software community to
develop Free AI."):
> First of all, I think that we do not need rules on the use of LLMs,
Do you mean that the existing rules (for example, the DFSG and the
Code of Conduct) are sufficient?
The reason I think we need a GR on this specific topic is: so many
people have dxifferent interpretations of what these existing rules
mean.
For example:
> We affirm that:
>
> * AI models should not be trained on data from authors who clearly
> disagree with this use.
Obviously I agree with this proposition. And I'm sure you agree with
me that existing LLMs have very much been trained on data from
nonconsenting authors (and scraped from nonconsenting hosters).
Indeed, I think this proposition of yours is a consequence of the
DFSG. But that is very much disputed.[1] LLM boosters apparently
think the current situation is just fine.
> * The output of AI models trained on software that requires that derivatives
> must be distributed under the same terms should follow this requirement.
And, obviously I agree with this. Again I think this follows from
the DFSG.
But this gives us a severe practical problem, because the implicit
social contract in the floss world, that we wouldn't randomly violate
all preexistiong norms, has broken down. Just to take one example,
Linux is tainted with authorship-laundered LLM output, now.
I think there is room for an operating system project that takes a
more principled approach to this problem, and completely rejects all
LLM output including from upstreams. But I don't think Debian can be
that project. It would throw out too much of our existing investment.
Or to put it another way: what do you think is the status within
Debian of the output from "bad LLMs" (the ones that are unethical,
according to your principles) ?
Regards,
Ian.
[1] NB that I regard this as primarily a question of ethics, rather
than law. If certain legal systems declare the current LLMs fine,
that doesn't the DFSG has to see it them as fine.
--
Ian Jackson <[email protected]> These opinions are my own.
Pronouns: they/he. If I emailed you from @fyvzl.net or @evade.org.uk,
that is a private address which bypasses my fierce spamfilter.
