Hi Lucas,
On 8/12/26 5:41 PM, Lucas Nussbaum wrote:
I'm not trying to dismiss the topic of climate change. However, if we
are discussing the environmental impact of Debian development, I think
that we should do it more globally. Singling out AI on this sounds like
AI-bashing, not like a well-reasoned discussion about the impact of Debian
on climate change.
Some things deserve to be bashed. ;-)
I agree that we need to look more globally, and that is precisely the
problem I have with this response.
The standard cannot just be "it emits CO2, therefore stop it", as Debian
development has a clear positive impact on the world. We should consider
the environmental impact of Debian practices, compared to their
outcomes. For example, other Debian practices could be changed:
On the contrary. The standard cannot be "The contribution to this
problem that Debian makes through its LLM requests (ignoring the
training that was used to make this possible) is relatively small,
therefore don't worry about it". We are discussing to make a position
statement: a message to the world about a subject that we care and know
a lot about.
If we look at the effects of LLMs on the world, I'm not sure how anyone
can ignore all the problems. They've been listed before, but just to
repeat a few: it uses so much energy that residents near data centers
asked to limit their use, otherwise the network cannot take it. Same
thing for drinking water. They convince governments to subsidize the
data centers, which leads to reduced services for the local population,
because they don't have the money. People are using chat bots instead of
talking to other people, which is resulting in less community, more
isolation, and it makes some people go crazy. Every creative platform is
overrun with slop.
On the pro side we have "with a lot of effort (that most people are not
going to put in), it is now possible to write good code a little faster"
and "we can more easily find bugs".
Comparing these, it is not hard to understand that when looking more
globally, LLMs are a negative influence on the world.
This is why I believe Debian should make a statement to the world,
telling people "you are not crazy, LLMs are indeed a huge problem". Not
because we personally cause all of that, but because we are experts in
this field and when we make such a statement, people may actually listen
to us. It may be used in hearings about building a new data center, for
example.
And of course, if we make such a statement, it also makes sense to want
our contributors to set an example; if LLMs are so bad, we shouldn't be
using them ourselves.
Why would Debian encourage me to travel to India, South Korea, Argentina
or Japan for Debconf, but at the same time, decide that it is
unacceptable to use AI assistance to make it possible to
develop/maintain udd.debian.org or debaudit.debian.net (at current
maintainer's capacity)?
I think it is reasonable to think about whether such things are a good
idea, indeed. The climate crisis is extremely serious and we should look
at every option to slow it. That doesn't mean everything must be
stopped, of course.
If we look more globally at the travel problem, I think it is true that
we should both limit the amount of travel (by allowing people to work
from home, for example) and also to limit the modes that pollute the
most (planes and cars). This is however not something that Debian has
expertise on, so I don't think it makes sense to issue a position
statement about it. Nevertheless, I would personally recommend everyone
(including everyone in Debian) to try and limit their carbon footprint.
But to bridge back to the subject: personal carbon footprints were
invented to pretend that climate change is a personal problem that
people have to solve individually. It's a deceptive tactic by the fossil
fuel industry. And while it is good if people try to minimize it, actual
solutions come from systemic change. And that is why I believe Debian
needs to make a position statement condemning LLMs: because such a
statement may help to achieve such systemic change.
Thanks,
Bas
