meanwhile at alpine/pmOS conference 2026...

and...

guess alpine is woke now

@ariadne Holy shit alpine is based as fuck

@freyalikesgirls @ariadne sadly it seems like Alpine might go with a pro AI slop* policy, so it seems more like multiple forces where some are based and some less based

*I use slop to refer to any sort of LLM code that has a dubious source which seems to all of today's available LLMs

@ell1e @freyalikesgirls as someone who has been vocally involved in the process, while it won't be a ban, it won't be as pro-AI as you may think. the policy we have arrived at, through very rigorous debate, is LLM-neutral, but anti-slop.

@ell1e @freyalikesgirls i think the policy is actually good because it is not specific to LLM, but rather aims to preserve the creative process, and explicitly sets expectations that preserve this. it is more agreeable than where Linux and Debian have landed, while remaining technology-neutral.

@ariadne @freyalikesgirls I simply don't believe in "technology-neutral" when the technology has such inherent issues.

@ell1e @freyalikesgirls @ariadne honestly, just like all “ai” art and writing is referred to as slop, it is the same for code. anything coming out of a generative model is definitionally “ai” slop

i reject the notion that there exists “ai” output that is not slop

tho i know i am more so arguing about definitions here x3

@lumi @ell1e @freyalikesgirls

the policy will be published soon, but the tl;dr is that generative algorithms of any kind for code *completion* is OK, but the actual creative elements of the output must be human-authored, and we reserve the right to reject patches where we believe they do not reflect genuine human authorship.

this means, in practice that "vibe coding" in the way that most are doing it is banned.

agentic development, in practice, is also banned by this policy.

using agents to design software changes, such as authoring design documents, commit messages and other related work is also banned.

using LLM technology as a substitution for competency is also banned.

using automated processes to generate patches is allowed, as long as there is sufficient evidence to support that creative elements reflect deliberate human authorship.

in practice, this makes the project unattractive to slopists, while setting ground rules for automated development in general.

@ariadne @lumi @freyalikesgirls I just don't think a sensible differentation between auto complete and creative elements exists.

Here is an example of auto complete of "function isEven": https://github.com/mastodon/mastodon/issues/38072#issuecomment-4105681567 (Video clip involving EU lawyer's commentary)

Anyway, I don't need to convince you. I just want to express why this policy doesn't seem good to me, in hopefully a way that people find insightful - whether they agree or not.

@ell1e @lumi @freyalikesgirls respectfully, none of us are lawyers, and the lawyers we have asked have given us the advice that we are pursuing, based on the questionnaire process and in-person conversations at the conference.

@ariadne @ell1e @freyalikesgirls interesting. would love to see it

@ariadne @lumi @ell1e ill have to thinn abt that one
does it essentially enable vs intellicode/jet brain type shit to exist in a valid way at a policy wide level? or am i misrepresenting ideas in a stupid way

@freyalikesgirls @lumi @ell1e intellicode and turning descriptions of algorithms into code completions, are the *only* LLM aspects that are allowed for creating new contributions, under this policy.

LLM technology is otherwise read-only.

@ariadne @freyalikesgirls @lumi I don't like it but I highly appreciate the clarification. Thank you for answering our questions! ❤️

@ell1e @freyalikesgirls @lumi like all compromises, nobody will be entirely happy, but we can continue to live with it and work together.

@ariadne @ell1e @freyalikesgirls while i am against all genai, i think it’s an okay comprise, kind of at the upper bound of what i am comfortable with it

@ariadne @freyalikesgirls @lumi Well, some clearly will be willing to live with it, for sure.

But any policy will have people walk away. I guess the question is which groups you have walk away the most.

Anyway, I'm rambling at this point and I don't really want to delve deeper into this.

@ell1e @freyalikesgirls @lumi i think the techbros who want to let claude loose on the project are the most likely to walk, and the project will be better for it.

@ariadne @freyalikesgirls @lumi That wouldn't be my guess. Anyway, I don't want to debate hypotheticals.

@ell1e @ariadne @freyalikesgirls in my experience, those who go for full slopcoding tend to feel very offended when they can’t do that

so i am thinking they would rather fork than accept that

@lumi @ariadne @freyalikesgirls for clarification's sake, that wasn't the type of fork I had in mind 😅

@ariadne @lumi @ell1e i knew it…. slop dev…… /j

@freyalikesgirls @ariadne @lumi I just think it doesn't matter which avenue (auto complete vs prompt window vs agent) you use to slop a project. The difference seems merely superficial to me. But clearly many in this thread have other opinions...

@lumi @ariadne @ell1e my problem is auto complete can be on without user realizing which is annpying, especially in current day high profile ides…
i guess the answer is “dont use them” but as someone who uses vs on windows, thats a little fucky….

@freyalikesgirls @ariadne @lumi there are IDEs with auto complete based on compiler semantic analysis without any LLM use. so there would be alternatives if you wanted to!

@ell1e @freyalikesgirls @lumi

yes. but in practice, we have no way of knowing if you use an LLM or an LSP to do the auto-completions.

also, some of us (myself included) are getting to the point in our lives where we physically cannot sit at a computer every day and type for 8+ hours per day.

so, that is the motivation for the narrow carveout for using automation.

@ariadne @freyalikesgirls @lumi Prefacing this with not attempting to convince you, but merely explaining my reasoning:

The reason I don't agree with that angle is that a ban is typically still useful even if it has enforcement problems. otherwise we wouldn't have any laws forbidding anything, for example.

I understand the accessibiltiy angle, but LSP auto completion remains an option.

(It would also be possible to e.g. train an LLM on opt-in projects as training data or other such ideas for auto completion, that may never be explored if there's no community pressure to make it happen.)

@ell1e @freyalikesgirls @lumi

yes, we understand that you are an idealist, and that you want a total ban.

this is basically as close to a total ban as any large-scale distro will ever do. i am sorry if you do not like it.

also sharing this slide

@ariadne having "put people first" be the last point on the slide is kinda funny ngl xndtdty

@ezri as someone who has left the project before because of the environment, yes. but in recent years, there has at least been an attempt to improve the vibe.

@ariadne "turning descriptions of algorithms into code completions" - ah, I was wondering why LLM needs to be involved in simple code completion. This makes a bit more sense - though this is also the kind of projects university students might get assigned to do as part of learning an algorithm, so I feel slightly worried letting that in.

I guess it's not "creative", but it reminds me of the distinction between software "engineers" and site reliability "engineers" at some companies, where the latter might not be expected to understand the algorithms they use, and... that seems like a shortcut at times.

@freyalikesgirls @lumi @ell1e

@michel_slm @ariadne @freyalikesgirls @ell1e imo it’s not ideal to allow an llm to puke out a full algorithm, but it’s kind of at the upper bound of what i will tolerate

as long as it doesn’t turn into a whole conversational chatbot, i can … somewhat tolerate it

will i use it? no. will i like it? no. but eh i will go after more egregious issues

@lumi @ariadne @michel_slm @freyalikesgirls wouldn't that basically allow in 1:1 unattributed copies of entire functions? or am I misunderstanding what you're referring to?

@ell1e @ariadne @michel_slm @freyalikesgirls if the llm pukes that out, yeah

technically it’s all plagiarised because every single token comes from training input gathered without consent

(whether that is relevant for copyright, idk. but i don’t tend to argue from the perspective of copyright because i don’t think copyright should exist)

@lumi @ariadne @ell1e @freyalikesgirls I recall Ted Ts'o arguing that this kind of coding is not copyrightable anyway (and for straightforward algorithm implementation he's probably right, but a lot of coding doesn't fall into that category)

I also agree that other issues are more important, but I sympathise that it is hard to do a complete ban too.

And a bit envious as this is way tighter than what Fedora and Debian allow...

@michel_slm @ariadne @ell1e @freyalikesgirls i feel something like what is proposed here is a good start

we can keep pushing the opinion window to be more genai-hostile, and hopefully projects will get more strict with it

but this is at least a tolerable starting point, imo

@lumi @ariadne @michel_slm @freyalikesgirls i highly disagree 🤷‍♀️ but i also don't want to bore everyone by going on about this forever. i've already explained why

@lumi @michel_slm @freyalikesgirls @ell1e i do not have the energy to continue with this, and have done my best to push back on LLM use, resulting in the conditional policy. it is already impacting my mental health negatively, and having people shouting at me over a community-wide decision is exhausting. please tag me out of this and future conversations on this topic, thanks.

@ariadne @michel_slm @freyalikesgirls @ell1e i’m happy there are people pushing back; thank you for that

i don’t mean to sound like i disapprove of the policy; i think you did a good job

will stop tagging you now

@ariadne @lumi @michel_slm @freyalikesgirls i should have thought to untag you, i apologize 😞