glibc AI policy

[Florian Weimer]

* DJ Delorie:

> Florian Weimer <fweimer@redhat.com> writes:
>> For the record, I'm not happy with requiring Assisted-by:.
>
> I think we need some way of identifying content generated by AI, even if
> not legally significant, until such time as the copyright situation is
> much clearer.  Knowing which AI created the content may help if there
> are AI-specific copyright concerns (or, for example, governments banning
> certain AIs from certain channels).  If a court finds that one
> particular AI's training methods causes its output to be a copyight
> violation regardless of significance, we need to have a way to deal with
> that.

I don't think how this could be useful.  For example, one vendor seems
to claw back copyright assignment on generated content to the user after
they violate service terms, but only for free accounts.  How would we
model that in analysis tooling?

I also don't understand why we are suddnely worried so much about risks.
May be let the FSF do the worrying on our behalf?

We accept contributions from people without probing them to see if they
did anything risky when preparing the submission.  Before DCO, there
wasn't any training as part of the copyright assignment procedure,
either.  It's just a risk we accept to move the project forward and be a
generally welcoming project.  I don't see why this has to change, or the
risk profile changes.  As I suggested before, not being on Github may be
all that we need.

> However, I'm not opposed to folks just attributing the AI content to
> "AI", such as "Assisted-by: AI (Makefile changes)" during this "legally
> insignificant changes only" phase.

There is no such phase right now.

Thanks,
Florian