Debian is debating the use of LLMs

https://www.debian.org/vote/2026/vote_002

The Debian project is engaging in a debate and vote on whether large language model (LLM) tools may be used in the development of Debian, its distribution-specific code, and the project's website. The general resolution is currently being discussed leading into a vote on the subject.

84 points · 39 comments · view on lemmy.world

39 Comments

valar@lemmy.ca · 63 pts · 10d (12 replies)

In summary,

Argument 1: LLM code is trained on stolen copyright, filled with errors, will mix up best practices, burns out our contributors, hurts the free software community, and threatens Debian's most important quality: its stability.

Argument 2: Some people find AI tools helpful.

lambalicious@lemmy.sdf.org · 17 pts · 9d (3 replies)

"just one more dpkg bro I swear we are going to get pulseaudio and systemd working this time bro just one more data lane, you have to do this or you are going to fall behind bro trust me"

I'm all for Argument 1. Also, it's been scientifically proven that any sort of extended use to AU lowers your intelligence, and that would be the last thing we need for a distro that needs to remain a bastion of Stable.

frog_reborn@mstdn.social · 4 pts · 9d (2 replies)

@lambalicious @valar

Scientifically proven? Any sort of AI usage? Just genereal intelligence?

There are definitely some initial studies out already that point in that direction, but I think your wording is a bit overly broad and definitive. It will be a while until we understand all the factors involved and what is going on there exactly.

Lettuceeatlettuce@lemmy.ml · 3 pts · 8d

One of the most dangerous things about modern LLMs is that they rarely mess up in obvious ways.

They are incredibly good at making very subtle mistakes that look correct to non-experts.

That, combined with how drastically they lower the barrier of entry to creating functional software, creates the ultimate Dunning-Kruger scenario.

The amount of effort it takes to properly vet LLM-produced code is an order of magnitude more than the effort it takes to produce, which makes maintainers' work basically untenable. The "solution" proposed by AI shills is for the maintainers to do the same thing, but obviously that suffers from the exact same problem.

It's better to be broadly right than precisely wrong, and LLMs are fantastic at the latter.

I think we're in the middle of another wave similar to the "code bootcamp" era. The market is getting flooded with a bunch of people who want to make software engineer salaries without having to put in the effort to learn proper software engineering skills or computer science theory.

We're constantly hearing a more extreme version of the coding bootcamp propaganda, "Why go into debt $60k-$100k and spend four years of your life getting a Comp Sci degree, when you could spend just $15k-$25k and 6-12 months getting our cert to land the same jobs?" Now it's, "Why spend all the time and effort learning how to write & understand code, when you could just prompt an LLM to build all your software for you?"

But in the end, people will learn that there actually is value in having a deep, technical understanding of the theory and practices of software engineering. They'll learn that you can't just vibecode the next AWS, Linux kernel, SAP, hypervisor platform, or firewall appliance. They'll learn that it's not enough to just tell a swarm of agents to maintain your code base forever.

But it's going to be a very painful lesson for a lot of people, and when the AI bubble finally pops, it's going to be a truly Biblical reckoning for the entire tech industry.

lambalicious@lemmy.sdf.org · 0 pts · 9d

I'm not bothered that much: the results are observable empirically with how idiocracy people in general have become in the last 2 years (not that we as a species were a beacon of understanding before that).

That said, in terms of "just general intelligence?" I'd say it's two intelligences that are hit the most: emotional intelligence and logical intelligence. Not necessarily in that order.

Lost_My_Mind@lemmy.world · 7 pts · 10d

You forgot argument 3.

Argument 3: FUCK THIS SHIT! FUCK AI!

Hi, join us over on !FUCK_AI@lemmy.world

Franconian_Nomad@feddit.org · -5 pts · 9d (6 replies)

People like Linus Torvalds?

I realize that some people really dislike AI, but this is an area where I'm willing to absolutely put my foot down as the top-level maintainer.

Linux is not one of those anti-AI projects, and if somebody has issues with that, they can do the open-source thing and fork it.

Or just walk away.

AI is a tool, just like other tools we use. And it's clearly a useful one.

It may not have been that "clearly" even just a year ago, but it's no longer in question today.

valar@lemmy.ca · 10 pts · 9d (2 replies)

I'm just summarizing the arguments made by the actual proposals by the Debian team linked in the OP

Choice 1 provides this position on AI

Debian has a well-earned reputation for stability. This stability is crucial to Debian's position in the free software ecosystem. It is our belief that widespread LLM usage comes from the "move fast, and break things" attitude that, while common in many parts of this industry, is contrary to what makes Debian Debian, and is inappropriate for Debian contributors.

In practical terms, LLM usage raises the following concerns:

Copyright

LLM output has very unclear legal status: it may be possible to copyright on its own merits, or not; it may be affected by all of the licenses and copyrights in the training data, or not. Debian Policy and the DFSG require absolute clarity for licensing and copyright[1][2]. Software and other contributions written conventionally by humans with unclear copyright or license status are not allowed in Debian; LLM output should not have a special exception to this.

Quality

LLM output has many well-known problems with accuracy.[3][4][5] A LLM can never "know" if its output is correct since it merely produces syntactically likely combinations of the training data. In some environments this is good enough. In Debian, it is not. For instance, in packaging, each Debian source package is unique. Since packaging syntax and best practices have changed over time, a LLM-produced package will have a mixture of contents spanning the age of the archive, with watch files that do not work, overrides out of context, imaginary copyright, and will generally be unfit for upload. A seasoned Debian contributor with packaging expertise may find some limited usefulness here, but a new contributor cannot, and would not know how to fix it. These same quality and accuracy concerns apply clearly to all of the areas listed in the scope of this proposal above. If Debian were a closed organization comprising only domain experts who never leave, this might not be an issue; however,

Community

Debian is a project that is more than just code: it is a community built on shared interests in free software and solving technical problems. Debian intentionally grows this community through many means, and new contributors are always encouraged to join. Allowing LLM contributions breaks this. New contributors submitting LLM output for review places an unnecessary strain on the reviewer, which can lead to burnout. Furthermore, LLM-dependent new contributors do not actually learn and understand the details of Debian packaging or processes, so they cannot come to replace a former burned out DD.

Ethics

LLM companies directly hurt the free software community as whole by scraping the whole web for training data without any regard for license, copyright, or even established conventions such as robots.txt.[6] This has had a major negative impact on Debian's public web resources, effectively a large scale and perpetual Denial of Service attack on sites that many users rely on. As a consequence parts of our infrastructure were not reachable at all, and JS-based checks had to be enabled. Many other projects were similarly affected. Furthermore, LLM training consumes a staggering amount of resources[7], and the user verification systems that we have been forced to implement as protection waste resources as well. This is blatant disregard for the internet as a public resource, wastes system administrator time, and although individual LLM sessions do not directly use massive resources or DoS the public web, the fact that they can be used at all is a direct result of these unethical behaviours by the LLM companies.

Debian has a Social Contract. [8] Our priorities are our users and free software. Debian is Stable. [9] Users and organizations choose Debian because it is reliable and secure.

Debian is not here to generate as much code as possible requiring manual review by a shrinking number of human volunteers, or to package every piece of software, or to rush new features, but these are what LLMs are used for.

In conclusion, allowing LLM contributions is contrary to the social contract and the common cause of creating a free operating system with a focus on quality and stability.

While Choice 2 only says

The Debian project recognizes that AI-assisted contributions raise many concerns, e.g. about the technical quality and maintainability of such contributions, and their legal status. AI itself also raises additional concerns, about its impact on society at large, on the IT industry and on Free Software; about its environmental impact; and the aggressive or non-compliant practices of AI scrapers.

Nevertheless, many Debian contributors find AI tools helpful when contributing to Debian, and ultimately for improving Debian.

Franconian_Nomad@feddit.org · 4 pts · 9d (1 reply)

Lol what? That’s their official phrasing? Holy shit!

„Are you in favour of introducing vanilla icecream in our product line?

Points against it: its an abnormal abomination created by the devil that tastes like frozen shit!

Points in favour: while admitting that vanilla icecream is an abnormal abomination created by the devil, some people like eating eating icecream.“

Lmao

AlligatorBlizzard@sh.itjust.works · 3 pts · 7d

Yeah that ain't even trying to be unbiased lmao.

It'd be absolutely wild if we end up at a point where the Debian team forks the kernel over AI. I doubt they've got the team to pull it off, but that statement screams that someone on the team is tempted.

rumschlumpel@feddit.org · 1 pts · 6d (2 replies)

The thread where Torvalds said this was about AI code analysis tools, not LLMs (aka generative AI). Big difference.

Franconian_Nomad@feddit.org · 0 pts · 6d (1 reply)

LLMs aka Large Language Models are the current standard of AI. (Despite all their problems and weaknesses)

While he said this in a discussion about AI code analysis (powered by LLMs) his statement was clearly meant broader. Don’t be delusional.

rumschlumpel@feddit.org · 1 pts · 5d

It certainly sounds like he would use LLMs if he found them useful for kernel development, but I didn't get the impression that he currently finds them useful for this purpose.

cupcakezealot@piefed.blahaj.zone · 30 pts · 9d (1 reply)

i swear to god fuck ai and llms can we just have one thing that's not enshittified by them

Swedneck@discuss.tchncs.de · 1 pts · 6d

i really do think this is something that should be brought up more, whatever ones opinion on AI there at least needs to be some things that don't immediately jump on the bandwagon, as a fallback and control group.

like, fuck, imagine if our earliest ancestors had found some funny looking red mushrooms with white spots and all had a taste at once, our species would have gone extinct!

rumschlumpel@feddit.org · 5 pts · 10d (2 replies)

The choices kinda look like they would both either explicitly or in effect disallow practically all LLMs. Nice.

Though that doesn't answer the question about code analysis tools that the kernel devs recentently talked about (with Linus Torvalds being firmly in favor of using them). They aren't generative AI, but are AI nonetheless and unless Debian already talked about it elsewhere it does need explicit policy IMO.

naught101@lemmy.world · 9 pts · 10d

Non-generative AI is just complicated stats.. There's no reason not to use ML or whatever if you can find a use for it (e.g. bug hunting, optimisation).

Generative AI is that plus random noise, copyright theft, environmental destruction, and fascism.. It should be off the cards for Debian

supermarkus@feddit.org · 2 pts · 9d

The choices kinda look like they would both either explicitly or in effect disallow practically all LLMs. Nice.

From what I see so far, there is a loophole. LLMs may not be used to generate source packages but if Claude generates a script that then does the work and the script stays private and isn't official Debian software, it's fine.

This may be an oversight or deliberate. No idea.

sem@piefed.blahaj.zone · 4 pts · 10d (2 replies)

Wondering who gets a vote

BartyDeCanter@piefed.social · 15 pts · 9d

These 1044 people: https://nm.debian.org/members/

mjr@infosec.pub · 5 pts · 9d

Some of the people who work on Debian.

GustavoM@lemmy.world · 0 pts · 8d (2 replies)

Feel free to put me in a cross and burn me alive for posting this, but I don't see a single problem with AI/LLM's being used in small things like bug testing a code and/or finding out ways to improve a code.

Hund@feddit.nu · 2 pts · 8d

I highly doubt that you don't know a single thing about AI, the climate crisis, the computer parts crisis, the datacenter/human health crisis and a lot of other things.

Swedneck@discuss.tchncs.de · 2 pts · 6d

i mean.. okay then, step up and stretch your arms out while i get the hammer and nails..

Serinus@lemmy.world · -16 pts · 10d (14 replies)

LLMs are not going to replace a large number of people.

But they are very helpful, especially in coding. If you're not using LLMs in normal business development, you're going to fall behind.

It's made me a significantly faster developer, largely that a lot of the tedious work feels less tedious, and my projects are more polished.

You just can't vibe too much or you'll come out with an unmaintainable mess. There's a difference between using these tools responsibly and just producing slop.

jtrek@startrek.website · 21 pts · 9d (2 replies)

Everyone I work with who uses LLMs more than a little bit hasn't gotten better at coding. They've just made bad code faster, with even less understanding.

It's something like a catch-22. The people smart enough to use them effectively don't need to. The people who aren't smart enough will misuse them.

Serinus@lemmy.world · -24 pts · 9d (1 reply)

So you don't code and you like pithy nonsense.

jtrek@startrek.website · 18 pts · 9d

I'm a professional software engineer. You seem to have drawn at least one nonsense conclusion

lambalicious@lemmy.sdf.org · 18 pts · 10d (7 replies)

Hacker News is over there →

Serinus@lemmy.world · -6 pts · 10d (6 replies)

This isn't tesseract. I'm allowed to have (professional, informed) opinions.

lambalicious@lemmy.sdf.org · 18 pts · 9d (4 replies)

Then do so, please.

hendrik@palaver.p3x.de · 8 pts · 9d (3 replies)

Burn.

Yeah, I don't think worrying about 450 new CVEs in the Linux kernel each day made me any faster... Or my life as an admin more comfortable... This and fine-tuning all kinds of spam filters, securing web forms, protecting the servers which all get hammered by scrapers... I mean we got a lot to do because of AI. But it all feels kinda unproductive to me. Like the myth of Sisyphus. Bur maybe some admins and devs got quicker shoving the boulder around thanks to AI?!

lambalicious@lemmy.sdf.org · 4 pts · 8d (2 replies)

Like the myth of Sisyphus

And like the myth of Sisyphus, it's all about self-inflicted pains.

AI is more like placing a downwards conveyor belt under your feet on the mountain. You get a quicker result, but that result doesn't have you going anywhere useful.

hendrik@palaver.p3x.de · 1 pts · 8d (1 reply)

Wasn't Sisyphus a clever rascal who evaded death, pissed off the Gods several times and that's why he got punished? I think it was sacrilege, not really self-inflicted pain. Unless we want to count rebelling against powerful entities as inflicting something upon oneself 😄

lambalicious@lemmy.sdf.org · 3 pts · 7d

I wouldn't say he's clever if he hasn't yet realized he has access to a kinetic weapon against any town or settlement downhill that he has been fruitlessly charging and resetting for ages. At best, I'd say he probably has built up mad upper body strength.

eugenevdebs@lemmy.dbzer0.com · 2 pts · 8d

Weird how you don't like Tesseract when it's made by one of the admins you go to bat for and have removed several comments you disagree with.

You're just like Patrick.

HubertManne@piefed.social · 8 pts · 9d (2 replies)

This feels like almost exactly like a comment I have seen before on llms. Are you cutting and pasting something?

Serinus@lemmy.world · -11 pts · 9d (1 reply)

Jesus Fucking Christ. Typing in full sentences does not make me an llm.

HubertManne@piefed.social · 9 pts · 9d

I didn't say that. I asked if you copy and paste the same reply on llm critical posts.