>>>>> "Charles" == Charles Plessy <[email protected]> writes:
Charles> Hi all, First of all, I think that we do not need rules on
Charles> the use of LLMs, especially the first ballot option which
Charles> is an invitation to many people to leave Debian, but
Charles> alogether all the current options as they decide on new
Charles> rules that we can not enforce efficiently (and therefore
Charles> fairly). I will vote NOTA and hope that this will be the
Charles> outcome of this GR.
Charles> This said, if there must be an option to win, then I would
Charles> like to propose one that calls for the development of a
Charles> truely Free AI. The dogma is that it is impossible. I
Charles> think that it is propaganda, "naked king" attitude, and
Charles> fear of AI bubble bursting. I am sure that useful and
Charles> totally Free models are possible, it is just that I do not
Charles> have the competence to write them.
Charles> So I am looking for seconds for the following text. If you
Charles> would like to second it, please help me to fix it: I have
Charles> avoided the use of AI to write it, and as a result it is
Charles> clearly non-native English. Let's improve it together; I
Charles> will send a signed email later as I am away from my key
Charles> today).
Charles> ---- Draft ballot: Call the Free Software community to
Charles> write Free AI ----
Charles> We the Debian project, are a community that produces a Free
Charles> operating system by integrating Free software core
Charles> components as well as thousands of applications and
Charles> accessories. Respect and diversity are among our core
Charles> values.
Charles> We affirm that:
Charles> * AI models should not be trained on data from authors who
Charles> clearly disagree with this use.
Charles> * The output of AI models trained on software that
Charles> requires that derivatives must be distributed under the
Charles> same terms should follow this requirement.
Charles> * The training data of AI models should be broadly
Charles> accessible for study, audit, reproduction and modification.
Charles> It has been said by some that this is impossible; we
Charles> believe it to be wrong. We call to the Free Software
Charles> community to write such AI models that follow our core
Charles> values and can be integrated in our operating system to
Charles> empower our users to uses of this technology while
Charles> respecting Free Software, privacy, diversity and
Charles> environment.
Charles> ---- End draft ballot ----
Seconded.
I'm not really happy with
the first bullet point, but I suspect that I'm in the minority.
The free software movement grew out of the AI lab, and in particular
rms's frustration with authors stopping people from using the work they
created in the interests of proprietary use of that work. I think there
are a lot of ways in which authors restricting how their work is used
for training advances proprietary interests more than it advances
freedom. I absolutely agree that a closed model where work is used
against and author's interests advances proprietary interest more than
it advances freedom. I agree that we as a project should not go further
the eventual conclusions are on copyright, fair use, and training data.
That is, if infringement in a lot of jurisdictions we care about, we
should not support it as a project.
But personally, if the courts let us use AI to free work in open-weight
models, I'm all for that.
So, I'd also second (and prefer) an option without the first bullet, or
with a more permissive first bullet.
signature.asc
Description: PGP signature
