Hi Jonathan,
Thanks for your thoughtful message.
On 11/7/26 16:05, indieterminacy wrote:
In many respects, taking any agreement formalising tolerance towards AI
is a nuclear option - taking a utopian viewpoint on a technology which
has very far reaching costs and risks with an assumption of the
capabilities to restrain such issues being able to continue in perpetuity.
In my utopia, we would indeed have genAI. Because I think we can figure
out how to use genAI ethically, and I believe we will find good use for
it.
Maybe that is naive optimism. So I do try to heed warnings like yours.
But it is that same optimism that drives me (e.g. to Guix).
(As for your analogy; maybe it is fitting because it demonstrates the
risk of both sides: A powerful, but dangerous technology pops up, some
terrible experiences make us commit to never use it, we fail to apply
other solutions, and now we risk extinction.)
Perhaps I feel that betting entirely on developing those other solutions
is also unwarranted. Nevertheless, I'm optimistic about that too, so if
the rest of the project wants to take that risk, then I can live with that.
While you assert it as a question of artificial competition, I assert
that its about opportunity cost.
The opportunity cost reasoning works both ways though. Currently I
cannot use Guix for most of my projects because of the missing and
outdated packages; I'm still mostly using podman/dnf/pip.
If I could use an agent to help with the package maintenance, then I
would have more opportunity to improve the infrastructure (and I would
have a usable system while doing so).
Yes, in the past no-one had a working system. So it is fine to put in
the work required to make Guix working for more people. But we can only
do so much work each, so I prefer to do work that a machine can't do.
Nevertheless, while I still don't find the arguments compelling enough
to do everything by hand, I'm willing to do so out of respect for the
community.
As such, Id posit that Guix will need to be mindful of AI Steganography,
either from the perspective of avoiding abuses but to avoid patent
trolls upending legal foundations.
FWIW, I'd say that using genAI to create code that is large and complex
enough to have room for steganography will probably be problematic for
other reasons. E.g. it would indicate that there might not have been
enough 'human in the loop', or that the used model/agent was not free
enough.
Hugo
P.S. just to not be misunderstood again: I do like doing the work by
hand. I even enjoy working with 'ed', because it forces you think in a
different way. I'm also enjoying using genAI, because that is yet
another way to approach our craft. They complement.
Actually, programming with a line editor or with an agent is similar, in
the way that they both ensure that you do the programming 'in your
mind', instead of 'in the text editor'. That's probably why I like
both; as the thinking part (or model building) is the most important.
