Login
You're viewing the y.diskcat.com public feed.

Replies

  • MaddieM4MaddieM4@raphus.social
    Oct 27, 2026, 2:03 PM

    Oh and I have to keep saying this, but I do not want to hear a fucking word about how "projects need committee leadership, that would solve these problems" when COMMITTEE-LED PROJECTS DO THIS CONSTANTLY.

    While usually being irreplaceable because of infrastructure investment in ways that individual-led projects are usually not.

    I hate to have this chip on my shoulder but I'm so done at this point.

    💬 0🔄 19⭐ 57
  • Gerardo Lisboagvlx@masto.pt
    Oct 27, 2026, 3:16 PM

    @MaddieM4
    Although the didn't say they weren't going to use it just "there will be no AI of any kind in the default installation".

    However, one thing sort of implies the other.

    If the committer cannot explain, in their own words, what and how they came to that solution, this should rejected.

    #noaislop #SoftwareDevelopment

    💬 0🔄 1⭐ 6
  • MaddieM4MaddieM4@raphus.social
    Oct 27, 2026, 3:21 PM

    @gvlx Right, which is where I was going with the "nuanced stances" jab.

    They knew they were making that implication. They relied on that implication for marketing. And they knew they had no intentions to actually keep the core codebase free of LLM contributions.

    The thing about setting that kind of slimy precedent is that now everything you say needs to be examined as legalese, rather than understood in terms of simple good faith sentiment. Even if it wasn't about LLMs - a thing I have a hard personal boundary about - I really don't want to do business with a sketchy org that requires a baseline distrust filter applied to everything they say.

    💬 0🔄 16⭐ 44
  • Ivan CarapovicChapz@mastodon.online
    Oct 27, 2026, 5:29 PM

    @MaddieM4 @gvlx From your own wording of what they said, I understood too they will not have any AI features in the app. And this at the moment still is true? However, you asumed/wished/expected that this means also no AI controbutions in code. They never said that.

    Kinda unfair to blame them, and also again, assume, they did that on purpose. Anyone could have asked them to clarify AI usage in code contribution, and seems nobody asked to clarify.

    💬 0🔄 0⭐ 3
  • Paco Hopepaco@infosec.exchange
    Oct 28, 2026, 2:54 AM

    @Chapz This is a really legalistic take. That (a) they will explain if asked, and (b) we should ask them before judging. I don’t think either is true.

    The point @MaddieM4 made, which is valid, is exactly this legalistic point: since their broad-brush proclamation about “No AI is a feature” can coexist with “AI is a contributor,” we all have to parse what they say far more finely. They lose trust.

    If that’s the ODF’s stance (and I know you don’t speak for them, it’s just hypothetical), then we shouldn’t assume anything more that the black letter of what they write (or give them the benefit of doubts) because they’re reserving the right to be tediously legalistic.

    @gvlx

    💬 0🔄 10⭐ 26
  • MaddieM4MaddieM4@raphus.social
    Oct 28, 2026, 3:34 AM

    @paco @Chapz @gvlx exactly. I _can_ deal with a legalistic org that has a track record of screwing people over by the letter of the law, but given any freedom of association at all, why should I? Why should anyone? That untrustworthiness becomes more work for everyone else and it's highly antisocial. Why shouldn't that behavior be limiting to their prospects? If orgs that behave that way aren't shooting themselves in the foot on a practical level, more will behave like that, and we'll live in an exhausting world full of selfish people.

    Orrr, hear me out, we could have the barest minimum of self respect and say "I don't have to tolerate that," and not reward skeeviness.

    💬 0🔄 9⭐ 30
  • Paolo Vecchipaolo@mastodon.opencloud.lu
    Oct 28, 2026, 8:12 AM

    Hi @MaddieM4, we discussed the issues that AI contributions are causing during the public part of Monday's Board meeting and we are looking to extend our Development AI Policy:

    wiki.documentfoundation.org/De

    You'll notice that the contribution you refer to is related to a test unit and 2 people have been involved with review and testing :
    gerrit.libreoffice.org/c/core/

    So it seems our views on AI aren't that distant. Maybe next time just ask instead of jumping to conclusions.

    @paco @Chapz @gvlx

    💬 0🔄 0⭐ 1
  • MaddieM4MaddieM4@raphus.social
    Oct 28, 2026, 11:14 AM

    @paolo @paco @Chapz @gvlx

    Okay, I'll ask you directly. Are there any LLM contributions to the codebase of LibreOffice? I need you to understand, that itself is a dealbreaker for a lot of people who have historically donated to LibreOffice in the past, and this is the start of your policy that you just linked to:

    > While the use of generative AI based on Large Language Models in software development has several issues, they are allowed in LibreOffice contributions with certain restrictions.

    EDIT: I would say that the line in the sand is really technically "LLM contributions merged into the main branch of the repository." Obviously you can't control what people send you. But the people responding this strongly to a screenshot of a commit co-credited to Claude Opus 5.5 are not "misinformed." They do know what they're objecting to. Your policy leaves money, minds and hearts on the table - really worse, it means people will disparage you on the internet.

    💬 0🔄 2⭐ 13
  • Paolo Vecchipaolo@mastodon.opencloud.lu
    Oct 28, 2026, 1:02 PM

    @MaddieM4
    Didn't I answer your question in the post you are replying to?

    LLM assisted contributions are not banned but have limitations to discourage them and must be clearly marked as such. Many submission get rejected as they are clearly fully LLM generated, not just LLM assisted.

    How would you suggest to improve the Development/AI Policy without pushing contributors to try to hide that the code was generated with the support of an LLM?

    @paco @Chapz @gvlx

    💬 0🔄 2⭐ 1
  • MaddieM4MaddieM4@raphus.social
    Oct 28, 2026, 1:07 PM

    @paolo @paco @Chapz @gvlx

    Plenty of projects have a more effective policy that is entirely enforceable. This is a solved problem.

    1. LLM contributions are not accepted here. We will politely reject them.
    2. If we find out that you tried to sneak one past us, you are banned permanently as a contributor.

    This is the baseline standard, and matches the way that copyright attribution problems have always been handled. See, for example, the libsodium AI policy.

    github.com/jedisct1/libsodium/

    💬 0🔄 3⭐ 19
  • Paolo Vecchipaolo@mastodon.opencloud.lu
    Oct 28, 2026, 1:24 PM

    @MaddieM4
    So you would ban also this contribution?

    gerrit.libreoffice.org/c/core/

    You think she should have hidden the fact that she's learning and got some support from an LLM to test her idea?

    Or this one on another unit test where the author has been told that we don't accept LLM generated description?

    gerrit.libreoffice.org/c/core/

    Just to understand if your line is that much far away from where we draw ours.

    @paco @Chapz @gvlx

    💬 0🔄 0⭐ 1
  • MaddieM4MaddieM4@raphus.social
    Oct 28, 2026, 1:28 PM

    @paolo @paco @Chapz @gvlx

    I would have told both people, "we're glad you have an interest in contributing to LibreOffice, but the final PR needs to contain zero LLM content in the code or description. What you do in private is your business, but what you submit to our maintainers' attention is ours. If you could come back with clean versions, they'll be accepted."

    Like I said, polite but firm. Not everybody reads the policy, so it's okay to explain gently.

    If someone tries to hide their LLM usage, *that is an untrustworthy person.* You are not losing anything of value by telling them to take their business somewhere else. You are saving yourself long-term strife.

    💬 0🔄 2⭐ 22
  • Paolo Vecchipaolo@mastodon.opencloud.lu
    Oct 28, 2026, 1:42 PM

    @MaddieM4 I'm totally with you when we talk about sizeable submissions which are anyway already getting rejected.

    When talking about LLM assisted submissions we are talking about this:

    gerrit.libreoffice.org/c/core/

    gerrit.libreoffice.org/c/core/

    Others aren't that much bigger.

    Without a transparent admission from the submitter would you have been able to determine that these submissions were LLM assisted?

    @paco @Chapz @gvlx

    💬 0🔄 0⭐ 0
  • MaddieM4MaddieM4@raphus.social
    Oct 28, 2026, 1:49 PM

    @paolo @paco @Chapz @gvlx

    Probably not. But much of the point of accepting submissions is to onboard submitters, right? You are investing in a human being and their ability to store and utilize institutional knowledge about your project.

    If someone repeatedly contributes to the project, in larger scales, it will eventually become obvious if they use LLMs - people who do so don't really "turn it off" some of the time, once they're acclimated to these tools. This is mentioned in your own AI policy - the skill atrophy problem. So while some problematic contributions are small enough to fall under a practical, perceptual and legal radar, and that is a fact of life, problematic contributors do eventually reveal themselves, and you deal with the problem then. That may or may not involve review of their previous submissions according to human judgment.

    Like I said, this is entirely analogous to previous copyright attribution problems.

    💬 0🔄 0⭐ 12
  • MaddieM4MaddieM4@raphus.social
    Oct 28, 2026, 1:52 PM

    @paolo @paco @Chapz @gvlx Let's also not forget that people who *do* try to sneak small LLM contributions past a strict No-AI policy:

    1. Are often doing so to make a malicious point.
    2. Can only make that point if they brag about it somewhere else.

    So frequently, on social media or other secondary locations, these people do openly brag about their behavior where they don't expect maintainers to see. At which point, if someone does come across that and screenshots it, you have evidence that wouldn't have been apparent from the micro-PR itself, and you can handle that appropriately.

    💬 0🔄 2⭐ 11
  • Paolo Vecchipaolo@mastodon.opencloud.lu
    Oct 28, 2026, 2:04 PM

    @MaddieM4 That's why as you can see from the commits, which was not evident from the screenshot you posted, we have a great team that checks thoroughly what is being submitted.

    The same teams onboards and follows many new potential contributors every month so of course we invest in people and we support them through their learning process.

    For some it will take a while to kick the LLM habit but with our and the community support they'll learn to be better than LLMs :-)

    @paco @Chapz @gvlx

    💬 1🔄 0⭐ 0
  • Paolo Vecchipaolo@mastodon.opencloud.lu
    Oct 28, 2026, 2:11 PM

    @MaddieM4
    We keep the situation monitored and see if we'll have to resort to a total ban in future but at present I believe it's better to let contributors be transparent about their use of LLMs so that we can deal with it.

    At least now you have a clearer idea about our approach and the actual type of contributions we are talking about.

    I hope this will help you reconsider your initial statement about people being "defrauded" as it is surely not the case.

    @paco @Chapz @gvlx

    💬 0🔄 0⭐ 0