We are a community of KDE users and contributors, calling for the KDE Project to adopt a No-AI policy for the KDE Plasma Desktop and other related components.
Recently, there has been a lot of uproar within the KDE Community and within KDE Projects. KDE Eco has already declared that a policy which allows for AI use within KDE is incompatible with KDE Eco goals and values.
Similarly, we want to make ourselves heard.
You must log in or register to comment.
At this point banning AI slop is worth it solely to eliminate techbros and preserve whatever social contract the project has.
I don’t even care about pros and cons anymore every pro AI person has been a cryptonazi.
I’ve always preferred projects take on an anti LLM stance for PRs as it pushes people to either write code themselves or at least audit their llm generated code thoroughly and ensure it looks like it was written by a human, at which point it has the same effect as writing the code themselves imo.
Yeah. I am not willing to accept the argument that it doesn’t matter because people can lie. Imagine if WINE decided the clean-room policy wasn’t necessary because people could lie about formerly working on proprietary Microsoft code. It is preposterous. No one pushing this argument is acting in good faith. They are all either wreckers, scabs, infiltrators, or Nazis.
Was KDE allowing AI generated code before this? I have only checked some KDE apps and not Plasma but stuff like Kate had no AI code. So I assumed there was a blanket AI ban in place. Now there is this proposed LLM policy and that weird presentation about integrating AI in desktop. I really hope this trend dies. A lot of it feels like fascists trojan-horsing into free software.
The large desktop environments like Gnome and KDE are an amalgamation of hundreds of projects. As far as I know, neither have embraced a blanket policy, but many sub-projects (individual applications, etc.) have, one way or another.
Yesterday I arrived at the conclusion that, due to LLMs being embraced as standard procedure in the corporate sector, there is intense pressure on core Free Software projects to lower the bar, so the labor practice of LLM-driven software development can be fully standardized across the board. This has been very successful for the corporations at the foundation level (e.g. Linux, Mesa3D), but there are still a lot of hold-outs.
I think projects like KDE should be much removed from Linux and Mesa in terms of feeling the effects of corporate trends because Linux and Mesa make a ton of money for big corporations but KDE doesn’t at least aa far as I can tell. The source of pressure for KDE should be either weird guys at the top or donors or both. I am not sure what exactly is the case with them.
deleted by creator
Another day another reactionary luddite take from the FOSS community :(
shut up
Care to elaborate a bit?
Using LLMs to assist development is a good thing. I’m not sure what elaboration is necessary outside of that. So what, adversaries can use AI to find CVEs in KDE but not allowed to send them out because LLM bad?
There are a number of things going on here. To start, the Luddites were not reactionaries. Luddism was a labor movement which identified that the capitalists were deploying a new technology designed and financed specifically to undermine labor power and intensify exploitation in the workplace. It is actually a very similar situation to the one unfolding in the tech sector today, but the reactionary force here is not the technologists who take pride in their craft, who care about things like determinism, safety, correctness, and sustainability. The reactionary force is the purveyors of a technology explicitly designed to transform intelligence into a commodity, while rewriting history and demolishing conventional methods of research. The reactionary force is the side that has had trillions of dollars invested into it, to construct a new epistemological paradigm where the richest men in human history can change history as simply as their minions can rewrite a system prompt.
Second, this extremely contrived example you provide is not what is being debated, nor what the draft policy states. I think if the policy stated specifically that machine learning algorithms can be applied strictly as a method of analysis, similar to linters, fuzzers, static analysis, and other familiar methods of identifying software defects, there would not be nearly as much of an uproar. The policy under consideration is much more permissive. It is a green light for slop, as long as the slop gets reviewed. There are several problems with this.
First of all, if you give these motherfuckers an inch, they will take a mile.
Second, if you allow the slop, someone has to review the slop. Nobody wants to review slop.
Third, if you amass a codebase of slop which nobody actually even wrote, it will become unmaintainable very quickly.
Fourth, this shit makes mistakes all the time, and someone has to do the work of cleaning it up.
Fifth, as people spend their time trying to figure out the best way to prompt Copilot to write their program for them instead of actually considering the problems at hand, their skills will atrophy. If your program deals with a problem which requires solving a system of differential equations, and you prompt Claude to write it for you instead of actually learning differential Calculus, you are truly lost on the void of slop.
Sixth, free software is a collective endeavor. The collective erodes completely when instead of humans interacting with humans, you have agents dumping shit into a queue.
Seventh, the underlying basis of copyleft software licensing is the idea that copyright ownership is the instrument which is used to force downstream to share their changes back to the community. It is not clear LLM code can be copyrighted (as it has no human author), and it undermines copyleft.
Eighth, LLMs can be instructed to simply reproduce copyleft software packages in another language (as to make it distinct), and this is used as an excuse to strip copyleft and re-license plagiarized work permissively. This has already happened.
Ninth, the commercial LLM models in use today are trained on the entire corpus of free software, and their output should be considered to be a derivative work. They reproduce code from their training data verbatim in many instances, mixed together under dozens of incompatible licenses with no attribution as required in most cases.
So okay, maybe you can use an LLM to identify bugs in some cases, but I do not trust anybody who is ready to rubber stamp their usage in light of all these issues. This shit is still a novelty. It should never pass muster for a serious Free Software project which is committed to building a community of professionals around the task of publishing reliable software under a rigorous legal framework.
I don’t give a shit if people use it in their personal projects, but of all places, it does not belong in core infrastructural projects of the free software movement.
I did not know that about the Luddites at all, I will take this new piece of knowledge into account in my future haha.
I don’t want to leave this comment on “read”, but I also don’t really know how I’d like to respond. I will absorb what you’ve said regardless and think about it all for sure.
