The proposed handling of LLM in Debian in the latest "Bits from the DPL" is a bit concerning.
-
@zerodogg I hate that the conclusion people jump to from "I'm not sure" is "I'll just go ahead and do the thing that may or may not be bad until I'm sure" instead of erring on the side of not doing the thing.
@thomasjwebb @zerodogg I'm actually a big fan of judicious inaction. There's good reason for the old saying "act in haste, repent at leisure."
-
@thomasjwebb @zerodogg I'm actually a big fan of judicious inaction. There's good reason for the old saying "act in haste, repent at leisure."
@SteveFoerster @zerodogg I’ve weirdly gotten out of a lot of trouble by being broke/thrifty or lazy. I think fomo is making people reckless.
-
The proposed handling of LLM in Debian in the latest "Bits from the DPL" is a bit concerning. It misses the mark by acknowledging issues with LLM usage, and then dismisses them all by saying that "As a society, we rarely respond with categorical refusal. Instead, we regulate, reflect, and take responsibility for how we use them." and suggesting absolutely no regulation or reflection (and no particular responsibility other than that which comes with any contribution).
I'm not a Debian developer, just a longtime user (and upstream for an unimportant package). But if Debian isn't the principled, ethical one, then I'm guessing no one will be.
https://lists.debian.org/debian-devel-announce/2026/03/msg00001.html
@zerodogg this line feels a bit cherry-picked at the expense of the larger point. The more relevant line, I think, is "Simply refusing to engage with widely used tools does not make them disappear; it only reduces our ability to shape how they are used within our project."
And I think this is a good summary of the issue. Debian may well end up voting for total abstinence from AI tools within the project, but that isn't e.g. going to stop upstream package sources from using them. If the Linux kernel is already incorporating AI-assisted code, what is the meaningful alternative? For a project that is basically just a middleman for distributing software, I don't think it's possible to avoid software "tainted" with AI in a meaningful way. That would effectively require large-scale human-only rewrites, for which Debian does not have the resources.
-
@zerodogg this line feels a bit cherry-picked at the expense of the larger point. The more relevant line, I think, is "Simply refusing to engage with widely used tools does not make them disappear; it only reduces our ability to shape how they are used within our project."
And I think this is a good summary of the issue. Debian may well end up voting for total abstinence from AI tools within the project, but that isn't e.g. going to stop upstream package sources from using them. If the Linux kernel is already incorporating AI-assisted code, what is the meaningful alternative? For a project that is basically just a middleman for distributing software, I don't think it's possible to avoid software "tainted" with AI in a meaningful way. That would effectively require large-scale human-only rewrites, for which Debian does not have the resources.
-
@zerodogg I dislike that this post uncritically reproduces Nussbaum’s false equivalence.
He wants to make this about using any ‘tools’ at all, as if any GenAI critics have also blanket-opposed automated testing, continuous integration, compilers, and syntax highlighting (a genuine example he gave in his thread).
@benjamineskola @zerodogg Absolutely. I haven’t seen anyone complain about LLMs being used for code analysis. It’s using them (and other people’s code) to generate code that’s the problem.
-
@benjamineskola @zerodogg Absolutely. I haven’t seen anyone complain about LLMs being used for code analysis. It’s using them (and other people’s code) to generate code that’s the problem.
-
@foolishowl @zerodogg what is the objection to specifically, though? Large language models as a technology, or specific vendors? Debian is never going to hand out licenses for an Anthropic or OpenAI product simply on the basis that they're proprietary software. But it's not like the project has ever *banned* the use of a paid, proprietary IDE to support one's work—how would it even know?
There exist LLMs trained on public, open data sets with public weights that can run on a personal machine, and would appear to be suitable for inclusion in Debian—are these tools also objectionable?
If the goal is "completely halt use of LLMs as a technology used in any form", I don't think it's realistic to expect Debian Developers to be able to accomplish this.
-
@foolishowl @zerodogg what is the objection to specifically, though? Large language models as a technology, or specific vendors? Debian is never going to hand out licenses for an Anthropic or OpenAI product simply on the basis that they're proprietary software. But it's not like the project has ever *banned* the use of a paid, proprietary IDE to support one's work—how would it even know?
There exist LLMs trained on public, open data sets with public weights that can run on a personal machine, and would appear to be suitable for inclusion in Debian—are these tools also objectionable?
If the goal is "completely halt use of LLMs as a technology used in any form", I don't think it's realistic to expect Debian Developers to be able to accomplish this.
-
The proposed handling of LLM in Debian in the latest "Bits from the DPL" is a bit concerning. It misses the mark by acknowledging issues with LLM usage, and then dismisses them all by saying that "As a society, we rarely respond with categorical refusal. Instead, we regulate, reflect, and take responsibility for how we use them." and suggesting absolutely no regulation or reflection (and no particular responsibility other than that which comes with any contribution).
I'm not a Debian developer, just a longtime user (and upstream for an unimportant package). But if Debian isn't the principled, ethical one, then I'm guessing no one will be.
https://lists.debian.org/debian-devel-announce/2026/03/msg00001.html
@zerodogg surprisingly Gentoo of all distributions took a hard line against AI
This may wildly end to me running Gentoo at this rate.
-
@zerodogg this line feels a bit cherry-picked at the expense of the larger point. The more relevant line, I think, is "Simply refusing to engage with widely used tools does not make them disappear; it only reduces our ability to shape how they are used within our project."
And I think this is a good summary of the issue. Debian may well end up voting for total abstinence from AI tools within the project, but that isn't e.g. going to stop upstream package sources from using them. If the Linux kernel is already incorporating AI-assisted code, what is the meaningful alternative? For a project that is basically just a middleman for distributing software, I don't think it's possible to avoid software "tainted" with AI in a meaningful way. That would effectively require large-scale human-only rewrites, for which Debian does not have the resources.
@ehashman I agree that it probably won't be possible to avoid software tainted by LLMs. But Debian can't only be about technical issues. Even this exact post provides plenty of examples that Debian *isn't* just about technical issues. Debian is also about community, and about ethics and freedom.
At this point, much of LLM-critique is about ethics. It's about how workers that are being abused to train the models. It's about them systematically undermining free software licenses by feeding them into the LLM grinder as fodder for generating code. As he mentions, it's also about the environment.
I hold Debian to a higher standard than others. Perhaps that's unfair. But it's also why I choose Debian. Sure, the distro is excellent on technical merits, but the difference from many others is the community and their thoughtful approaches to ethics and real-world issues.
Debian can't dictate what others do. But Debian can lead by example, and make principled stances. I've sort of come to expect that from them.
That said, I'm not one to dictate what Debian does. I don't get a vote when it comes to that. But it makes me sad, and I feel a bit hopeless, when even Debian doesn't take a principled stance.
-
@zerodogg this is the theme I keep seeing. Acknowledge issues, and then say "so we have to watch out for that" ignoring that no processes exist or could practically exist for the watching of the out.
@aburka Yeah, I'm seeing the same. Just plain dismissal of any issues as either "someone else's problem" or "this is going to happen regardless, might as well join in".
-
@zerodogg surprisingly Gentoo of all distributions took a hard line against AI
This may wildly end to me running Gentoo at this rate.
@gourd Yeah, I'm kind of impressed by them. Never really considered running it before, but this just might convince me to give it a try.
-
@gourd Yeah, I'm kind of impressed by them. Never really considered running it before, but this just might convince me to give it a try.
@zerodogg Gentoo has always struck me as a Bit Much even as someone who ran Arch for twenty years before tiring in my old age of my 30's and switching to Debian Stable, but I have to go with the
distro less susceptible to slopcoding if it comes to it. -
@zerodogg Gentoo has always struck me as a Bit Much even as someone who ran Arch for twenty years before tiring in my old age of my 30's and switching to Debian Stable, but I have to go with the
distro less susceptible to slopcoding if it comes to it.@zerodogg but Gentoo is a distribution where pre-compiled package repos are a new thing as of 2023
-
@zerodogg but Gentoo is a distribution where pre-compiled package repos are a new thing as of 2023
@gourd Well, silver linings, I do like my precompiled packages, and that must mean it has had time to stabilize
-
@gourd Well, silver linings, I do like my precompiled packages, and that must mean it has had time to stabilize
@zerodogg I don't mind compiling some stuff from source if necessary but compiling LLVM or Chromium-based shit is a nightmare I never want to do if I can avoid it.
-
@ehashman I agree that it probably won't be possible to avoid software tainted by LLMs. But Debian can't only be about technical issues. Even this exact post provides plenty of examples that Debian *isn't* just about technical issues. Debian is also about community, and about ethics and freedom.
At this point, much of LLM-critique is about ethics. It's about how workers that are being abused to train the models. It's about them systematically undermining free software licenses by feeding them into the LLM grinder as fodder for generating code. As he mentions, it's also about the environment.
I hold Debian to a higher standard than others. Perhaps that's unfair. But it's also why I choose Debian. Sure, the distro is excellent on technical merits, but the difference from many others is the community and their thoughtful approaches to ethics and real-world issues.
Debian can't dictate what others do. But Debian can lead by example, and make principled stances. I've sort of come to expect that from them.
That said, I'm not one to dictate what Debian does. I don't get a vote when it comes to that. But it makes me sad, and I feel a bit hopeless, when even Debian doesn't take a principled stance.
@zerodogg perhaps you can read through this post and let me know what you think. I'm not sure what the goal here is, other than asking for some sort of principled public statement https://cloudisland.nz/@ehashman/116178358384455284
-
@benjamineskola @zerodogg Absolutely. I haven’t seen anyone complain about LLMs being used for code analysis. It’s using them (and other people’s code) to generate code that’s the problem.
@mathew @benjamineskola @zerodogg I've complained about LLMs used for code analysis. Mostly because, if the one used as my day job is representative, they massively suck at it.
-
@benjamineskola @mathew @zerodogg breaking news: "lint" considered harmful /s
-
@mathew @benjamineskola @zerodogg I've complained about LLMs used for code analysis. Mostly because, if the one used as my day job is representative, they massively suck at it.
@beeoproblem @mathew @zerodogg Yes, precisely. In my experience they produce a lot of output and when you dig through it all there's not actually very much of value to it.
And, besides which, if the argument against them is that they're based on plagiarism, that still applies if they're being used for review/analysis and not generation of new code.
