Proposed KDE LLM guidelines (#187) · Issues · Plasma / Plasma Workspace · GitLab
invent.kde.org
This is a continuation of this mailing list thread about "fully LLM-generated merge requests, preserved at https://mail.kde.org/pipermail/kde-devel/2026-September/004496.html There, I proposed...

Direct link to the LLM Policy Discussion: https://invent.kde.org/plasma/plasma-workspace/-/work_items/187

Example of a KDE dev not wanting environmental concerns to play any role, despite KDE having an environmental project:

Devs rejecting any notion of transparency by disclosing AI usage in code submissions (in direct opposition to advice from the FSFE):

And all of this is coming right after the debacle with an AI slop project being allowed to give a presentation at Akademy, and the person asking how this was allowed being blocked by KDE:

https://lemmy.nz/post/42029517

as well as the discussion thread bringing up legitimate concerns about AI usage being closed:

https://discuss.kde.org/t/sorry-to-bring-up-a-contentious-topic-kde-ai-llm-policy/46333/164

As someone who genuinely loves KDE, and has even donated to the project multiple times, I sadly have to accept that the majority of KDE have hopped onto the Pro-AI train, and they will not be dissuaded otherwise.

  • Aatube@lemmy.dbzer0.com
    13·
    4 hours ago

    the policy in question:

    The golden rule for LLM usage in KDE is Don’t be lazy:

    1. Don’t try to use a tool to replace your own judgment, interpersonal communication, or learning process.

    2. Don’t take unsustainable shortcuts.

    3. Don’t avoid growing as a person.

    The result will be poor-quality work that eventually becomes someone else’s problem.

    Nobody in KDE should know if you use an LLM — not because you’re concealing it, but because your usage of it produces output indistinguishable from what you could produce yourself. Contributions with obvious and lazy LLM usage may be ignored or closed.

    Using an LLM to generate contributions

    KDE follows a “human in the loop” principle: you must be making decisions and adjustments beyond mere prompting. The output must express your unique humanity in some way.

    1. Don’t be a “meat proxy”. For example:
    2. Don’t submit throwaway LLM-created changes as a “first draft or proof-of-concept for maintainers to improve”.
    3. Don’t submit “vibe-coded” changes you don’t understand and couldn’t make yourself.
    4. Don’t disclose LLM usage as a way of trying to excuse the potential errors or poor quality of a contribution that you haven’t put much effort into.
    5. Don’t add “Assisted-by: [some LLM]” tags to your commits; it’s just free advertising for the LLM’s provider.

    (emphasis original, except for the fourth bullet point)

  • esc@piefed.socialEnglish
    75·
    2 hours ago

    Actual proposed policy:

    Nobody in KDE should know if you use an LLM — not because you’re concealing it, but because your usage of it produces output indistinguishable from what you could produce yourself. Contributions with obvious and lazy LLM usage may be ignored or closed.

    Generally, don’t. What you are willing to write is generally the upper limit of what others are willing to read. LLMs, on the other hand, often produce long, rambling, soulless, corporate-sounding nonsense that nobody wants to read; don’t do it. This includes:

    Don’t use an LLM to “organize your thoughts” and copy-paste the output somewhere. Organize them yourself. Don’t use an LLM to generate commit messages and merge request descriptions. Write them yourself. Don’t use an LLM to generate a reply to someone’s question or comment, then copy-paste that as if it were your response. Understand and reply yourself.

    The one acceptable usage is to write text in your native language and then machine-translate it into English, with no stylistic or tonal changes.

    But of course reading is for the LLM, it’s far better to be outraged about ‘hopping on the Pro-AI train’.

    • ProdigalFrog@slrpnk.netOPEnglish
      11·
      2 hours ago

      That section only relates to descriptions of pull requests, not the AI code itself. They don’t want long rambly AI generated explanations of how the code works, but they are fine if the code itself is AI generated.

  • leftzero@lemmy.dbzer0.com
    121·
    4 hours ago

    Well, fuck.

    Any viable alternatives, or should I switch to BSD? (Gnome obviously doesn’t count, it’s designed for tablets, and unusable on desktops).

    • Kraiden@piefed.socialEnglish
      4·
      3 hours ago

      This. Dunno where I’m going to go once KDE starts collapsing under the weight of AI slop. XFCE maybe? Do they have an AI policy? What’s Cosmic looking like these days?

      • ProdigalFrog@slrpnk.netOPEnglish
        4·
        2 hours ago

        Xfce have embraced AI with their new Wayland compositor.

        Cosmic accepts AI code, but according to them, they take a preemptively hostile view toward AI code submissions and will rewrite some parts by hand.

        There’s some lists linked in the comments here of DE’s that reject all AI.

  • Richard@lemmy.world
    158·
    5 hours ago

    It’s a actually completely reasonable proposal

    It all comes down to one thing - understand the LLM generated code you want to submit, and use the LLM for nothing else than write code, like making other people having to read LLM text. This is a whole lot like the kernel’s LLM policy.

    Niko is right, the environmental corcern is not something this policy should go against - what can they do despite ban LLM usage altoguether (and per consequence get LLM generated code submitted anyway), they can’t have a blacklist for models, or force contributors to run them on their local hardware.

    • tyler@programming.dev
      141·
      4 hours ago

      what can they do despite ban LLM usage altoguether (and per consequence get LLM generated code submitted anyway), they can’t have a blacklist for models, or force contributors to run them on their local hardware.

      But they can? Like OP pointed out, wine forbids contributions from those that have seen Windows source code. How in the world are they supposed to enforce that? Turns out the open source community is pretty good at self enforcing.

      • Richard@lemmy.world
        25·
        4 hours ago

        I don’t think the average person who wants to contribute to wine has seen windows source code in the first place.

        It is much easier to attempt on enforcing something like this that to enforce a full ban on AI. The problem is that AI has a much higher exposure rate than the windows source code.

  • macniel@feddit.org
    363·
    10 hours ago

    this is quite… disappointing, and I say that as someone who uses primarily GNOME :(

    • ProdigalFrog@slrpnk.netOPEnglish
      11·
      10 hours ago

      I’ll likely be switching my gaming PC to Gnome due to all this :\

      (on a side note, the Gnome dev in that thread mentions that Gnome bans AI usage, but I haven’t been able to find anything that confirms that, and AFAICT Gnome doesn’t have an official AI usage policy, only Gnome Circle does for 3rd party Gnome Extensions).

      • woelkchen@lemmy.world
        16·
        9 hours ago

        the Gnome dev in that thread mentions that Gnome bans AI usage, but I haven’t been able to find anything that confirms that

        You can bet your behind that the developers employed by IBM Red Hat are 100% using LLMs to code. Chances are they are even mandated by corporate higher-ups to do so.

        • ProdigalFrog@slrpnk.netOPEnglish
          251·
          7 hours ago

          As it is impossible to completely avoid AI use with Linux, I must content myself with only minimization instead of complete elimination, as I won’t let the pursuit of perfection mean I just give up entirely. Any reduction is still meaningful, just as it’s still worthwhile to minimize fossil fuel usage and meat consumption as much as is reasonably possible.

            • ProdigalFrog@slrpnk.netOPEnglish
              17·
              8 hours ago

              you realize ai is just the tool and isnt a data center. data centers exist with out ai and ai exists with out data centers

              I do not think LLMs can be “just a tool”.

              Most LLM use is with corporate AI, and this KDE policy explicitly allows corporate AI usage. If they had restricted it to only open-source local models, I wouldn’t have nearly as many environmental concerns (though that would not solve the plagiarism issue).

              For a more complete look at all of the downsides of AI usage, I’d encourage you to take a look at this comment.

              • jasory@programming.dev
                2·
                7 hours ago

                “If they had restricted it to only open-source local models”

                Is this even remotely enforceable? Stylometry can pretty effectively determine between AI and human written text ( not sure about code specifically since those are much stricter languages), but can it tell the difference between models or even if the LLM was run locally? (almost certainly not in the latter case).

                • ProdigalFrog@slrpnk.netOPEnglish
                  4·
                  7 hours ago

                  Most contributors generally respect the rules of a project, and there are other rules put in place that are also not technically enforceable, such as asking a contributor to not submit copyrighted code (this would be virtually impossible for a reviewer to distinguish if it came from a closed source project), which also gives the project a stronger defense if brought to court, as they can then at least point to the rules and show where a contributor willfully broke them. In comparison, I suspect a court would not look kindly on a policy that says “Don’t disclose that you used a tool that has a 3 to 10% chance of introducing plagiarized copyrighted code with every prompt to avoid being shamed.”

                  As an example, the WINE project has a rule to not allow anyone who has even seen Windows source code from contributing to the project. This is to avoid the chances of someone subconsciously or consciously submitting a reproduction of what they saw in the Windows source code, and thus hopefully prevent the project being sued into oblivion from Microsoft (who would likely love to see that happen).

                  Adding a stipulation in the guidelines that contributors can only use more-ethical LLMs likely wouldn’t entirely prevent someone breaking that rule, but by in large, it is likely that most would adhere to it, making it an effective rule despite being difficult to enforce.

  • statelesz@slrpnk.netEnglish
    265·
    9 hours ago

    I do kinda get Niccolo’s point that a contribution policy is not the place to decide environmental concerns even though it feels contradictory at first glance. He’s generally very critical and nuanced when it comes to LLMs so I wouldn’t call him “pro-AI”. Just watch his YouTube content.

    • Otter@lemmy.ca
      6·
      6 hours ago

      I don’t really see why environmental policy shouldn’t be mentioned in the contribution policy. If it’s there as the justification for why a part of the policy exists and it is accurate about what the environmental impacts actually are, it seems to fit in fine within the policy? Not wanting the point there doesn’t make KDE pro-AI (since they might just want to mention it elsewhere), but it is valid criticism if they don’t acknowledge it anywhere.

      However I do feel that the other points in the post are weaker. KDE’s stance is in line with what the SFC recommends when it comes to AI in FOSS, specifically around contributions. I came across the SFC recommendations a while back and it’s held up well in similar discussions so far: https://sfconservancy.org/llm-gen-ai/llm-backed-generative-ai-recommendations.html

      Devs rejecting any notion of transparency by disclosing AI usage in code submissions

      I don’t think they rejected the transparency part? I agree that transparency is necessary. Rather their reply is telling the commenter to not shun people. Both of those points are reflected in the SFC recommendations (point 5 on disclosure and point 3 on shunning)

      And all of this is coming right after the debacle with an AI slop project being allowed to give a presentation at Akademy, and the person asking how this was allowed being blocked by KDE

      I get the feeling that KDE does not curate much of anything, for better or for worse. There is a good bit of content in the KDE ecosystem that is outdated or straight up shovelware.

      On top of that, Akademy is a conference that is intended for discussion, rather than a presentation about KDE.

      Akademy features a 2-day conference with presentations on the latest KDE developments, followed by 4 days of workshops, Birds of a Feather (BoF) and coding sessions. It is an opportunity to meet key KDE contributors, learn about the latest features and enjoy the great atmosphere. You will meet developers, artists, translators, upstream and downstream maintainers, users, and free and open software industry leaders from all over the world

      It wouldn’t work well if they had to curate every presentation at such an event. It is indeed deceitful to imply that the presenter is going to change the direction of KDE if in reality they rarely contribute and have little say in the direction of KDE. Also IMO it’s good that they let that guy participate in the conversation because it let everyone else raise points about why the idea is bad.

    • ProdigalFrog@slrpnk.netOPEnglish
      33·
      8 hours ago

      The contribution policy is the only place where anything meaningful can be done, as that is ultimately what will influence developers trying to contribute. Putting environmental concerns anywhere else is, IMHO, just toothless virtue signalling.

      I personally consider advocating for non-disclosure of AI use to be Pro-AI. Advocating for disclosure but still allowing it would be neutral.

  • ikidd@lemmy.dbzer0.com
    37·
    3 hours ago

    You may only contribute your time building the product I use for free in a manner I approve of.

  • Einhornyordle@feddit.org
    711·
    3 hours ago

    So when will we ban code snippets? And autocoplete? Actually, lets go back to punch cards already!

    I’m seriously getting sick and tired of this topic. AI is a tool, if you don’t like it, don’t use it, but let others do what they want. I hate vibecoders as much as I hate full anti-AI haters since both are equally wrong. AI is a powerful tool and the next step for assisted coding, but that’s it, do not let it do all the work and just allow it to work on it’s own without double-checking what it is actually doing. If you prefer to type everything yourself, letter by letter, go ahead. If you found an AI that produces perfect output that doesn’t need any adjustment/corrections, send me a link. If you have no idea what you’re doing, I do not want your code on my pc, no matter if you wrote it yourself or let an AI generate it for you and anything in-between. It’s that simple. Or at least it should be.

    And if the environment is your biggest concern, LM Studio is free, so is Qwen, Gemma, Mistral, GPT-OSS and a ton of other models that you can run locally on your PC. You will need a somewhat powerful PC and some more patience than with the top models running on almost unlimited resources in a datacenter, but they get the job done eventually nontheless. The job of assisting you, that it, no vibecoding! They’ll break your stuff even faster if not kept in check.

    • bignose@programming.devEnglish
      9·
      2 hours ago

      AI is a tool

      There is no such thing as “just a tool”.

      Even so, some technology (like the examples of “autocomplete” you give) are effectively harmless to anyone else. But some technology is harmful to others even if those others don’t use it.

      if you don’t like it, don’t use it, but let others do what they want.

      I’m happy to let others do what they want where it harms no one else.

      LLMs as currently deployed are not that. Their use, by anyone, is harmful to us, and we need to resist it for the sake of everyone.

  • statelesz@slrpnk.netEnglish
    1912·
    9 hours ago

    I find the proposed policy very nuanced and generally useful. It’s not banning LLM usage but still prohibiting mindless vibe-coded contributions. I don’t see an issue and your post seems to be blowing this out of proportion.

    • ProdigalFrog@slrpnk.netOPEnglish
      274·
      9 hours ago

      It makes no distinction between self-hosted LLMs or Corportate LLMs, and disregards the dangers of using LLM code in FLOSS projects due to copyright concerns and the environmental and social damage corporate AI causes. For more details:

      1. Simply using an AI that’s run on a corporate data centers encourages the construction of yet more data centers, with all of the environmental/climate negatives they bring, as well as local harms they induce on the people living near them, such as increased electricity rates.

      2. Using corporate AI directly helps the financial situations of those giant corporations (by boosting usage/user numbers, they are able to attract more investment capital), most of which are ran by right-wing CEOs who are more than willing to collaborate with and fund fascist governments to ensure that they are not regulated in search of both maximum profits. Some of these companies, such as Nvidia, Palantir and Oracle, genuinely appear to be seeking to use these tools for what would previously be considered crackpot conspiracy theory levels of public control and surveillance.

      3. I’m wary of the potential effects of AI usage resulting in declining or stagnating critical thinking based on some preliminary studies (example 1, example 2).

      4. I do not think LLMs can be “just a tool” as many claim.

      5. 99% of LLMs (the ones virtually everyone use) are trained on copyrighted code that is incompatible with GPL licensed projects. LLMs have a 3 to 10% chance to unknowingly reproduce the copyrighted code perfectly, introducing legal plagiarism into an open-source project, which could easily open up the developers to being sued in the future if companies scanned open source repos for copyright violations.

      6. Many courts are ruling that LLM generated code cannot be copyrighted at all, meaning that it also cannot be legally made GPL, losing the protections from corporate exploration the GPL normally grants. https://fsfe.org/news/2026/news-20260825-01.html

      • Richard@lemmy.world
        510·
        5 hours ago

        So, the broke CS students that want to help in their favourite DE while using LLMs to save time on their tight schedule should buy some overpriced homelab server before being able to contribute, is that it?

        • Funkt4st1c@lemmy.worldEnglish
          121·
          4 hours ago

          No, they should manually code to help with their studies, like we have for generations. What do you think we did before AI? Have more time in the day?

          Its so unbelievably simple: don’t turn on the coal burning sycophant machine. Read a book or something. If you arent passionate enough to do something manually, i dont want your automatic contributions.

        • ProdigalFrog@slrpnk.netOPEnglish
          61·
          4 hours ago

          You make it sound like FLOSS hasn’t been able to function for the last 30 years without code plagiarizing LLMs.

          Do you think the whole endeavor will just up and collapse without using fascist owned AI?

          • Richard@lemmy.world
            42·
            4 hours ago

            No, FOSS won’t die without AI, this was never a possibility.

            The real problem is that the ““AI”” industry is forcing us to take a choice: create policies to deal with it, or drown in the noise. This what an AI policy serves, to filter out the noise they created.

        • Linearity@piefed.zipEnglish
          41·
          5 hours ago

          The broke CS student that neither has money nor time shouldn’t be contributing.

            • Linearity@piefed.zipEnglish
              91·
              4 hours ago

              There’s no classism about this. You either contribute with labor or with money. The dev team could’ve already consulted an LLM themselves if that’s what they wanted. Meat proxies are useless.