Link: https://www.theregister.com/software/2026/09/24/talk-of-ai-in-kde-sets-the-community-ablaze/5298854

Graham sounds more and more like a grifter. Is KDE slowly doing the walk towards evil?

A proposal to make KDE an "AI-native" desktop has gone down like a house on fire: screams, flames, people running for safety. There may be no survivors.*

Last weekend was KDE's annual Akademy conference and it included a presentation proposing an AI-native KDE. We noted that it seemed likely to polarize the audience. It looks like this was a considerable understatement.

In the days since, a debate over proposed restrictions on LLM-assisted contributions descended into bans and a deleted thread. An outside campaign called for KDE to prohibit AI altogether, a GNOME developer proposed a similar policy for that project, and KDE developer Nate Graham apologized for his role in the uproar.

The Akademy talk was titled "A lovable, sovereign, AI-native KDE" and the slide deck [PDF] is now available. It ends by saying: "The question is not whether AI. It is how. You choose how much AI – and which."

It seems almost calculated to provoke an argument rather than invite discussion. We do not know if the authors intended this – we attempted to contact both of them, but they have not yet responded.

Graham opened a discussion on Invent, KDE's GitLab instance, about proposed restrictions on LLM-assisted contributions. Although we are not a member, we watched this with some interest over the weekend. It rapidly became heated. Moderators issued warnings, restricted further comments, and eventually removed the thread.

There is an archive of the discussion, but we warn you, it contains some highly offensive language, albeit censored by the member who quoted another member's tweets on X. As events unfolded, the participant who drew attention to the posts was banned first. The author of the offensive material was banned later after their identity was verified.

The discussion is gone, but the argument continues.

One response was an outside initiative called KDE for People, which called on the KDE project to adopt a No-AI policy. Around 250 people signed it before its organizers closed it to further signatures.

GNOME developer Jordan Petridis has also published The GNOME LLM Policy That I Want, proposing that LLMs be barred from creating or modifying anything submitted to GNOME or hosted on its infrastructure.

This mentions the recent ballot on AI usage in Debian. We reported on the developer referendum about a month ago, noting that the broad spread of anti-AI options in the ballot was likely to split the vote. (This likelihood was dismissed in the comments.) Well, as The Register's Asia-Pacific desk reported a few days later, Debian did not ban AI contributions.

Graham subsequently explained his involvement in a post titled KDE and AI, and you, and me. He opens by saying: "So I accidentally triggered an online shitstorm in the process of trying to craft a set of more restrictive LLM usage guidelines for KDE. Sorry about that."

He notes that the Akademy talk met "what I'm told was a fairly chilly reception" and that "the next day, a workshop was held about the topic, also receiving a chilly reception."

Every couple of years, the KDE project sets three goals for the next two years. Earlier this week, after Akademy, it chose its latest three: you can see them in the last column of the goal-setting Kanban board. The ones chosen are:

KDE for Enterprise and Deployments (Issue #5)

Better documentation (Issue #2)

Next Generation Styling for KDE (Issue #3)

We see no mention of AI in there. In context, that may come as a relief to parts of the community.

Bootnote

*A tip of the black Borsalino to the late great Terry Pratchett, for two different "house on fire" references we combined. ®

What part of this makes Graham a grifter? If it’s there, I am missing it. Also, it seems like KDE got the message from the community…at least for now.

I dislike AI. It's like the one ring. Stop trying to use it. But all these people think they're stronger than Boromir.

Doesn't Debian make use of the Condorcet method? How are their proposals supposed to "split the vote"?

I'm no luddite, have used it (not quite voluntary) at work. Gotta say has its use use cases but I don't think any of them are in FOSS right now. It seems hell bent on producing unmaintainable slop.

It would be fine for proofs of concepts and/or commercial code, where the time you get for tasks hardly ever is sufficient and most customers want a working solution and don't care what's under the hood. If there weren't the 3 giant red flags of environmental impact, cognitive offloading and dependence on big tech.

So why the fuck would you even consider using it in a FREE and OPEN source project, if it's none of those things?

Well said! And I’m stealing that red flag list.

As events unfolded, the participant who drew attention to the posts was banned first. The author of the offensive material was banned later after their identity was verified.

I've also noticed a tendency of censorship in KDE's forums, for posts that are simply inconvenient, even if written in a civilized way.

Funny that instead of learning from Windows 11's failure, some people just want to follow the same path...

I have this same sentiment, I took a long introspective journey to evaluate my bias. Even then LLMs have extremely limited use cases. But they can be useful, they just have a ton of baggage attached to them and condition.

Really feels like handling the ring from LOTR sometimes. Corrupting it's user even if they seek to use it for good

You can add AI allignment and questionable legal copyright status of produced code. :)

Yeah, that's a great analogy, it turns idealistic developers into empty slop-wraiths :D

The main reason open source projects are pushing to formally adopt a policy around (allowing) llm use is that they’re already using it.

If there was a unified front between the contributors and the community against llm use, how much code would have to be reverted and rewritten? Most likely a bunch. How long would that take? Most likely a while.

Now perhaps they handle it quickly or truly didn’t have any llm work in their code base, how will they handle the flood of legitimate bug reports from llm testing?

If you’ve ever worked in a trade where your role is split up in an assembly line you know that the person behind you getting access to a new tool that makes them faster necessitates you get a new tool that lets you keep up.

time to post this again:
a worthwhile read in regards to what pushing AI does to a contributor and a community:
https://blog.fallible.net/it-breaks-a-village/

I'm no luddite

You’re not a luddite. Because you’re a Luddite.

The actual Luddites weren’t opposed to technology. They were opposed to, in their words, “machinery hurtful to commonality”.

And well… You can’t spell “machinery hurtful to commonality” without AI.

You think (unpaid) FOSS developers have MORE time? O_o

Great piece, thank you for sharing!

Right‽

Heck, that guy who set up a camera in such a way to encourage a monkey to take a selfie didn't win his copyright case.

He likely cropped the pic afterwards and touched it up.

I don't see how that's fundamentally all that different than going back and forth with an AI and then tweaking it's output.

That's even before taking into account the copyrights of the original "training materials".

The camera guy did a lot more to earn the copyright than vibecoders, imo. Had the guy just started a video himself and then grabbed a still frame from when the monkey pressed the button again (or set up the camera to take pics every 1/4 second or whatever), he'd have a 100% valid copyrighted photo unlike the viber who'd at best have some pseudocode.

Aside: Are "AI Boosters" the Real Disadvantaged-Marginalised-Subaltern Party?

No, and this is an extremely suspicious framing. The largest structural and monetary support ever leveraged is being put behind "AI" rollout.
[...] Pushing back against it is what involves sticking out your neck.

Great article. Still working through it, but full of excellent, clearheaded points like this.

🤝
It should be posted in other interested communities on Lemmy (and Mastodon). I imagine it has already; yet this is the first time I heard of it.

I tend to agree, but I have occasionally seen my fellow AI skeptics behaving horribly towards people that decided to use LLMs in their own development of their own project.

The Lutris drama comes to mind, but I was also not very happy to find out that the creator of Booklore had deleted his repo after being fed up with all the harassment.

And there's plenty more.

Sure, it's been forked, but what the hell people, it's his project, which is for a self hosted ebook library anyway, nothing that's supposed to save the world. Could you all be cool? Please? At least in the github issues?

I even think you're right on the issue, but just because you're right doesn't mean you have to be such a dick about it.

Nothing. OP is stirring shit, attacking the other is part of that same tribalistic POV

The main value of LLMs in coding IMO is research and debugging not implementation

questionable legal copyright status of produced code

So many projects have LLM generated content now that ruling content from common LLMs illegal would be catastrophic to the economy

The gitlab discussion was opened by graham in order to implement some restrictions on AI use to help the maintainers with low-effort slop PRs. Unfortunately it completely derailed.

feel free to post it where ever it applies/adds to the conversation. i didn't write it, i just share it.

You know a lot and I mean a lot of Foss coders are paid by companies right. At least on the big projects.