On +2026-05-18 18:24:43 +0200, Simon Tournier wrote:
> Re,
[...]
>
> If there is currently no output, mentioning FreeBSD here is misleading
> and thus it should be removed.
>
> However, it appears to me worth to mention NetBSD which makes it clear
> <https://www.netbsd.org/developers/commit-guidelines.html>.
>
>
> > * LLVM
> > https://github.com/llvm/llvm-project/blob/main/llvm/docs/AIToolPolicy.md
>
> To me, considering the Pledge, the item #2 is still unclear. I propose
> to clarify by reusing “Extractive Contributions” from LLVM policy. For
> instance,
>
> 2. We kindly ask contributors to respect this choice and not use LLMs
> for their contributions to Guix. Nevertheless, code claimed to
> be produced in whole or in part by genAI **may be incorporated
> if the contributor has a valid copyright claim on the code**
> and if the contribution isn’t some extractive contributions.
>
> Where “Extractive Contributions” would be detailed in a specific
> subsection outside the Pledge. Well, something also reusing the
> Accountability from Mastodon
> <https://github.com/mastodon/.github/blob/main/AI_POLICY.md>.
>
Permit me to predict that it will soon be, if it is not already, a Turing-test
matter to exclude
autonomous AI-agents from becoming welcomed committers ;-)
The only obstacle for the metal-based that I see is the apprentice/sponsorship
process ;-)
Then it becomes who and what do you trust not to "misbehave" (except for
harmless fun :)
(Harmless? Whose idea of that ... ;-)
If you are radically strict about commits, what is the difference in requiring
human or algorithmic
"understanding" of the code they are submitting? Or requiring social manners in
word use in logs?
Why not just demand the same of Chat-whatever as you would of a human
equivalent, and see if it can do it?
For example, I'd like to see what Chat-whatevers come up with as proofs of
their legal right to publish their code. ;-)
Also, perhaps the algorithmic agent should prove that it knows enough, and has
acquired effective
behavioral restrictions such that it will obey Asimov's
<https://en.wikipedia.org/wiki/Three_Laws_of_Robotics>
for starters? (and not ignore decades of sci-fi movies exploring those issues?)
What prompts could humans give to force it to do that??
Maybe in trying to define robotic "understanding," we will learn something
about understanding other humans ;-)
But what I'm coming to here is:
I think that to imagine the current and rapidly coming versions of
Chat-whatever as
just producing stochastic remixes of training data selected by some
prompt-relevance filter,
(and further filtered to avoid too-literal repetition, then for spelling and
grammar etc.),
is to be distracted by history -- and IMO _MUCH_ less important than what we
can see evidence of in the pdf below.
I hope you will reread this pdf (easy to skip given it's non-informative file
name :-/ )
<https://lists.gnu.org/archive/html/bug-hurd/2026-02/pdffAOvuZusjj.pdf.>
by Brent Baccala ("and Claude Code") in its entirety, but think especially
about "1.2 MCP: Model Context Protocol,"
which IMO signals the arrival of a radically new and more powerful generation
of Chatbots.
Whatever, I expect we'll see a lot more Artificial Idiocy than Artificial
Intelligence before the dust settles. ;-)
--
All the best,
Bengt Richter
>
> Cheers,
> simon
>
