With Love to KDE: Take a Moment

I've been using KDE Plasma for four and a half years. The community is sweet and the software is stellar, and I see a bright future for it. I want it to be the best it can be! So, I'd like to talk about a small incident that I want KDE to lean away from.

TL;DR: Please look at adopting an "AI" Policy similar to Servo's. Other projects' policies, like Asahi Linux's (they love KDE!) and Bevy's, may also be worth a look. Even Forgejo has its "AI" Agreement, though in my opinion it's a bit watered down.

Grzesiek11 writes: Do you seriously accept AI-generated code into KDE software? That’s concerning. Let’s even ignore ethical debates and code quality questions, models used for this are trained on a lot of proprietary code. We do not know whether this constitutes a derivative work. Proprietary projects like Windows obviously do not care about copyright (unless it’s theirs), but libre projects should hold themselves to a higher standard in my opinion, as we really do rely on the code being owned by the people who write patches. Nate Graham responds: There’s an interesting parallel with KDE’s “real name” policy. In the past we were very firm about requiring one for contributions. But… how do we know the name someone picks is a real one? If it’s obviously fake like “uwu I’m a Kawaii Dragon” or “Max Mustermann”, then sure, we’ll know. But otherwise, it’s impossible. At a certain point we realized we were penalizing people for being honest about desiring anonymity and started accepting patches no matter the name. It’s a similar situation with AI, I think. There’s no way of knowing unless the use is obvious and not hidden — and at that point rejecting it would be be penalizing people for honesty. — A light exchange; or, *the incident*.

Before Nate made his response, I was also thinking about their old Real Name Policy. I thought it was kinda funny that KDE rejected psuedonyms for years for provenance reasons — and then felt they should accept LLM contributions even though a scraping LLM cannot ever have provenance.

Nate's reply then emulsifies these two positions. It seems his takeaway was not that pseudonyms have a negligible impact on provenance, but instead that provenance is impossible and so KDE should give up.

I find this odd? The logic doesn't sit well with me.

He's turned We can't know that someone's using a real name, so we must openly accept fake names.
Into We can't know that someone's not using an LLM, so we must openly accept LLM contributions.

But these statements don't evaluate the worth of pseudonyms or LLM code, and are instead purely defensive — "How practical is it to guarantee we avoid X?" (Which for almost any given X, the answer is "We can't guarantee much at all". People can lie on the internet!)

My 2¢ is that there are other reasons to consider not accepting something. For instance, it would be bad to say,
We can't know that someone's not a nazi, so we must openly accept nazi contributions.

Excuse the invokation of Godwin's Law. Obviously, I don't believe this is a position KDE would hold. I'm just underscoring the need to actually think about whether having X in a project is good, and what ought to be done if we find instead that it's bad.

So, are LLM contributions bad?

LLM Contributions Are Bad

As mentioned, LLMs trained on scraped data cannot ever give complete attribution. It's how they work; it's the party trick of a black box. It's also non-consensual use, and it's plagiarism.
- Occasionally, an LLM will regurgitate/resynthesize author credit. Sometimes these authors are not real, or are unrelated to whatever content is attributed to them. And unless the output is a 1:1 match for their work, it's incomplete credit and still plagiarism.
- Hypothetically, one could train a language model to only use public domain or consensually granted data, such as code you've written yourself. But, these tend to give poor results.
LLMs bring downward pressure on code quality, lost productivity, and maintainer abuse. LLM contributions are often accompanied by an erroneous, nonsensical, or blathering description. The contributor also tends to have little-to-no understanding of the contribution and cannot personally answer questions from the review process. This is a waste of maintainer time and labour, is disrespectful, and can lead to burnout.
Scrapers are a scourge on the open web. So many FLOSS projects have been struggling with DDOS attacks, draining labor and money. Information sources are being diluted in a flood of spam. It's an extractive and corrosive force capitalizing on the digital commons.
There's the environmental impact. The KDE Eco project is probably displeased by the power usage of "AI" datacenters, and by the increased reliance on coal and natural gas plants to provide that power.
- Hypothetically, a small local model could use less power than, say, playing a video game. But, these tend to give poor results.
And, importantly, LLMs are abetting fascism. These are scary times, and I like KDE because of its potential for being a reprieve from the power that tech wields over people. In constrast, "AI" normalization empowers what is increasingly clearly a tech oligarchy. "AI"'s greatest strength for fascism is the excuse; it's an excuse for thievary, bailouts, surveillance, discrimination, erosion of labor rights, and waving away responsibility. And that's to say nothing of its role in the disinformation machine.

I understand KDE has had some prior run-ins with "AI", such as Kdenlive's optional Whisper integration and a few in-progress chatbot clients. I'm not terribly fond of these, but right now I'd just like to see a plan for an "AI" contributions policy.

I'm not a decorated developer nor an expert on how not to feed into fascism, so please reach out to others to discuss. Reach out to the folks at Servo or Krita, or Bevy and Asahi Linux. Reach out to David Revoy. To Ed Zitron. To Andrew Roach. Anyone. See what people have already said, recap the ground we've already tread. Heck, I'm sure Brodie Robertson could talk about similar projects who've had to wrestle with an "AI" Policy.

Anyway, thanks for taking the time to read. This is important to me, and you'll find it's important to many. Take care and best wishes!