An anti-LLM policy?

daemonspudguy · 2026-06-15 15:43:17

Hello. I am curious if Xfce can adopt a policy against accepting LLM contributions (colloquially AI). In addition to the never ending copyright concerns with the way these systems inherently create, they also have tons of detrimental environmental effects and inherently create codebases that are impossible to audit because you can't explain the technical decisions you didn't make.

ell1e · 2026-06-15 16:30:44

I'm in the same boat with having LLM concerns.

I'm looking for a desktop to switch to, and I've had some somewhat difficult experiences with software where LLM code submissions (not just LLMs consulted for bug finding) was allowed and common. Generally, the move to higher rates of bugs and lower quality code done by most of the LLM-using projects worries me a bit, and how attributing other projects when taking snippets from them has seemingly gone out the window. Sadly, KDE seems to be going down that route, too.

Generally I've had the impression that pro LLM policies seem to often coincide with a climate where code quality isn't valued as much, but also more short term and less forward looking planning for the project's future which I find slightly concerning.

So I was wondering how XFCE handles that as well. There doesn't seem to be any AI or LLM policy mentioned here: https://docs.xfce.org/contribute/dev/st … pment_team

Last edited by ell1e (2026-06-15 16:53:52)

organicode · 2026-06-16 09:01:26

It is probably a very bad idea to use the tools of the predator/epstine class to do anything against the predator class. We don't really know what resources they will allocate to tools like libre software that defies their imperialism. There is an old saying, "you can't use the master's tools to bring down the master's house". All LLMs I've come across are served by techno fash predators, the C.A.G.e.M.A.F.O.A. etc (cloudFlare 'scAm'azon goog, elon-gated musk-reant, microshaft, appl, falsebook oracle and akamai) including outright contractors to murderous, regressive, orange-faced regimes. I wouldn't touch them with a barge-pole for pretty much anything.

If anyone is foolish enough to entrust their computer to these systems I have a feeling they will fail, either immediately or in the near future. Including failing to pass code reviews based on quality alone. Communities would do well to adopt a clause, No LLM was used in the making of this X, where X can be anything from docs and man pages, to webpages artwork and software, I think it would be healthy, to do this, to signal that it is not okay to use those abusive systems. It is impossible to prove beyond doubt, of course, this is just about making ones ethical position clear.

I think if someone has a fully free and open-source LLM and runs it themselves, it doesn't have the same ethical issues. Machine learning has been around for a long time. To me, the ethical issues center on how the predator class have scraped and withhold training data, masked weights, overdevelop in terms of datacenters, and overconsume water and/or electricity in pursuit of data collection, surveillance, profit, and ~~humanity~~ (VERY LATE EDIT: at the *expense of* humanity).

Last edited by organicode (2026-06-20 22:04:38)

lvsl · 2026-06-17 08:12:17

I'm not a good programmer myself, nor I have skills to program for a living, but if I need some ideas or even sample code, I go for AI (they use LLM anyways) for ideas. I can't say if this is a reliance of LLM or whatsoever, but we really cannot completely avoid the use of it now. Probably somehow directly if I never touch it, but someone in some website might of have use it.

This means we might be at anytime reading articles or some inspiration that comes from LLM. And we use them without realizing it.

Edit: But of course, if you do the research and reading yourself, it make you understand it more, and ask better questions. You are helping yourself in better using of AI

Last edited by lvsl (2026-06-17 08:14:12)

organicode · 2026-06-21 01:51:24

lvsl wrote:

cannot completely avoid the use of it now

People may be tricked into accepting LLM slop, but that is a far cry from choosing it.

Be wary of throwing up one's hands in defeatism, if people did that in mid 1980s we wouldn't have GNU and the free license software we enjoy today. It takes only a few thousand people coming togather out of billions to make positive impact. There are very simple ways to end the scourge of AI and tech fash more broadly:

One is to simply publishing code with a Copyleft license and not using M$GitHub servers to host (and subsequently steal) that code. According to Richard Stallman (Nexa Wednesday, 2025), G(Ewww)gle and others throw a lot of funding at developers to not use Copyleft for a reason (they intend to run away with it).

Another is to end uranium mining/export. Of what is publicly released information (since Feb, Oct and Oct 2024 respectively) M$haft, g(Eww)gle, and sc'amazon all have hundreds of millions set aside for nuclear power for their datacenters, poisoning the planet and its people[1]. Call on your so-called leaders if you live in Australia or Africa, to end the mining and export of uranium. There are affluent places in the world that do this by diluting the uranium in a slurry that they deposit back over the land.

[1] If Iran was allowed to pursue nuclear energy, then MAYBE we could give other countries the benefit of the doubt, but now that we know that disfavored nations are not allowed to pursue it because, as detractors to nuclear energy have always said, the depleted uranium ends up in the tips of munitions to penetrate armored vehicles because it is so dense but also poisonous and the dust that emerges from such munitions is tantamount to chemical warfare. In Iraq the genetic damage causes on their population is said will last "a generation". Honestly, I don't even know how some claim it affects only one generation, I posit that once genetic damage occurs it probably can't be passed to any subsequent generations.

Germany recently appear to be making g(Eww)gle liable for the damaging hallucinations their slop generators spurt out.

The orange-faced regime does insider trading a lot. The orange-faced one can be put on trial for corruption and even treason as it was admitted by this individual that the money was received to do the bidding of a foreign regime; a regime that has serious International Court of Justice (ICJ) arrest warrants out for its leaders too btw. The penalty for treason is apparently quite severe in the country referred to. No more "Drill, baby, drill". smile

I personally block C.A.G.e.M.A.F.O.A. (see above) servers at the IP address level, so I have zero difficulty avoiding their slop machines. I also block javascript, as detailed in this thread, and that seems to chop out a lot of it too.

Moving off the legacy web and into the P2P networks, like i2p and Peertube seems to be another way to crush the slop. They have difficulty measuring the importance of information in such spaces. On that note, dont show Like counts and other counts on content, by default. Counts are not only easily manipulated and cause behavioral and societal problems, they also give info to slop farms. Good fediverse (decentralized social) servers don't show counts by default. One must expressly click on a post to see those metrics.

We could pressure so-called leaders to throttle the connection to CAGEMAFOA servers, if you think those leaders aren't already compromised.

Its totally possible to avoid the use of their slop, but its like gambling, smoking, alcohol, and stock trading, you can give people the information as to why that stuff is rank, but ultimately its up to the individual whether they want to give themselves over to it. We could pressure "representatives" to do more, but I personally think broad-based education is key. Think physical zines.

lvsl wrote:

I'm not a good ( X ) myself, nor I have skills to ( X ) for a living

I understand this as "Gellman's Amnesia". Not an expert in a topic X so I'll forgo my understanding that the newsmedia or slop generator gets things spectacularly wrong in the fields that I do have some expertise in, and will trust it when its telling me things outside my expertise. Well, I'm happy to say there's plenty of ways to do things outside one's area of expertise that don't involve tech predators, and all of them offer opportunities for growth and discovery. We all have thousands of things we are not experts in. Imagine if we all had a mindset of talking to a abusive predator for all the things we don't fully grasp. How easily do you think people will be gaslit?

The BigTech predators have contracts with very bad regimes, if they inject obfuscated code to place your computer or modem into a botnet or other vulnerable position that they can exploit, would you notice? If they think they can gain from turning a camera doorbell or fridge into a weapon be assured they will. We saw the results of weaponisation of the entire supply chain that made electronics that would go to Lebanon less than two years ago in the "Pager Attack" which allegedly included also laptops and other communication devices, not just "pagers". Reports of likely hundreds (minimum of 99) 'Unit 8200' operatives at G/M$/Amz/Fb alone, is bad enough. Remember also, that tech predators exploited gamification tactics and people's childhood memories, to turn people into unpaid spies (see PokemonGO), and they turned hapless website visitors into unpaid data-labellers (see Recaptcha,HCaptcha) (LGMs, 2024), with whistleblwer Zach Vorhies recently coming out to say the mapping data extracted by g(ewwww)gle was likely used to end the lives of school children in the aggrieved nation. Vorhies was right about CloudFlare being a MITM Attack on the entire internet too, if memory serves. At what point do we say I will not talk to the IP address of Cloudflare or G(eww)gle anymore? At what point do we recognize an abusive relationship and end it?

It is incredibly liberating to be able to do a simple IP address lookup (or IP-to-ASN lookup) and be able to say, "I see you are on the servers of X corporation and I will not even give you the light of day". If even just 1-in-50 people did this and were forthright and vocal about it, I have no doubt the positive effect it would have. The predator class depend on 'network effect', so if the most perceptive, forward-thinking 2% did just this and were VOCAL about it the entire epstained class would evaporate fairly instantly. We would get our compute back overnight.

Sometimes its the simple things like boycotting, that has the greatest impact.

Last edited by organicode (2026-06-21 03:02:48)

scm · 2026-06-21 21:11:15

organicode wrote:

The BigTech predators have contracts with very bad regimes

Is there a such thing as a good regime?

organicode wrote:

It is incredibly liberating to be able to do a simple IP address lookup (or IP-to-ASN lookup) and be able to say, "I see you are on the servers of X corporation and I will not even give you the light of day". If even just 1-in-50 people did this and were forthright and vocal about it, I have no doubt the positive effect it would have. The predator class depend on 'network effect', so if the most perceptive, forward-thinking 2% did just this and were VOCAL about it the entire epstained class would evaporate fairly instantly. We would get our compute back overnight.

I like this idea, but when I tried this with my linux distro. It came back as "cloudflare". I'm not changing distros. I thought of switching to Arch, but I'm 100 % systemd free now.

organicode wrote:

Sometimes its the simple things like boycotting, that has the greatest impact.

I am so happy you did not say voting has the greatest impact.

Last edited by scm (2026-06-21 21:11:50)

organicode · 2026-07-05 02:05:53

scm wrote:

I am so happy you did not say voting has the greatest impact.

Vote everyday, vote with your wallet, physical notes, physical coins and physical cryptographically generated numbers, physical choices, physically spoken words. Unfortunately, voting in political elections is 97% ineffectual, it may have worked at very local levels once upon a time but now that communication systems are consolidated and distilled by tech predators, they are pretty much useless, agree.

scm wrote:

I like this idea, but when I tried this with my linux distro. It came back as "cloudflare".

Sorry to hear your distro uses CloudFlare. Maybe try a polite message explaining the issues. You never know, maybe maintainers are secretly hoping people start requesting better, so they have an actionable reason to exit the CAGeMAFOA extortion racket. There is also an, albeit slim, chance they don't know their domain was absorbed by, and encryption outsourced to, CF. Although I use distros that don't use CAGeMAFOA, they are differently troubling, but worth the trouble. smile

daffyduck · 2026-07-05 09:07:48

daemonspudguy wrote:

Hello. I am curious if Xfce can adopt a policy against accepting LLM contributions (colloquially AI). In addition to the never ending copyright concerns with the way these systems inherently create, they also have tons of detrimental environmental effects and inherently create codebases that are impossible to audit because you can't explain the technical decisions you didn't make.

I always against such blanket and generalist restrictions.

Today, best spell-checkers and copy-editing assistants are based on LLMs. Should we prohibit that?
Should LLMs be prohibited from finding spelling errors in code comments and names of variables?

If you order an LLM to verify some piece of code for issues and suggest improvements, should that also be prohibited?

A less extreme counter-examples:
If a programmer orders an LLM to improve a single line of code in some way he specifies, and then he verifies what and why the LLM has modified in that line, should that also be prohibited?

Judicious use of AI in programming can greatly improve productivity, and such effects are already well known. It would certainly be beneficial for projects such as XFCE, which is generally understaffed.

daemonspudguy wrote:

they also have tons of detrimental environmental effects

A better cure for that issue is: everybody stops making children, and stop buying cars and drive on bikes. Also, don't go to travel-vacation this year because it has huge "detrimental environmental effects" at global scale, probably at least 20 times greater than all current LLMs combined.

treeview · 2026-07-05 10:07:11

organicode wrote:

murderous, regressive, orange-faced regimes

Sorry, but I'm a slow GNU/Linux user, could you provide example(s) of what a "murderous, regressive, orange-faced regime" looks like?

Last edited by treeview (2026-07-05 10:08:53)

daffyduck · 2026-07-05 14:30:36

treeview wrote:

organicode wrote:
murderous, regressive, orange-faced regimes
Sorry, but I'm a slow GNU/Linux user, could you provide example(s) of what a "murderous, regressive, orange-faced regime" looks like?

Please, less politics here. I get enough of it everyday, already.

ell1e · 2026-07-15 15:38:44

At the very least, it might be a good idea to not allow LLM code to be committed into XFCE. See e.g. here why: https://lore.kernel.org/lkml/e12330b9-c … e64.org/T/

That wouldn't rule out LLM reviews or consulting an LLM for tech questions.

As for LLM spellcheckers, I honestly find them to be worse than what we had before. But your mileage may vary, I guess.

daffyduck · 2026-07-17 00:55:24

ell1e wrote:

At the very least, it might be a good idea to not allow LLM code to be committed into XFCE. See e.g. here why: https://lore.kernel.org/lkml/e12330b9-c … e64.org/T/
That wouldn't rule out LLM reviews or consulting an LLM for tech questions.
As for LLM spellcheckers, I honestly find them to be worse than what we had before. But your mileage may vary, I guess.

I watched the very interesting full video and I red the text. Interesting indeed. Although, in many examples the AI was clearly baited.

But anyway, in my opinion, it all depends on how the AI is/was used.

If you write a human language description of what a piece of code should do, and that description begins with "write one or two lines of code that implement this description", then the chance of any problems, whether correctness or copyright related, is much less likely.

Last edited by daffyduck (2026-07-17 00:57:33)

Xfce Forum

Sub domains

#1 2026-06-15 15:43:17

An anti-LLM policy?

#2 2026-06-15 16:30:44

Re: An anti-LLM policy?

#3 2026-06-16 09:01:26

Re: An anti-LLM policy?

#4 2026-06-17 08:12:17

Re: An anti-LLM policy?

#5 2026-06-21 01:51:24

Re: An anti-LLM policy?

#6 2026-06-21 21:11:15

Re: An anti-LLM policy?

#7 2026-07-05 02:05:53

Re: An anti-LLM policy?

#8 2026-07-05 09:07:48

Re: An anti-LLM policy?

#9 2026-07-05 10:07:11

Re: An anti-LLM policy?

#10 2026-07-05 14:30:36

Re: An anti-LLM policy?

#11 2026-07-15 15:38:44

Re: An anti-LLM policy?

#12 2026-07-17 00:55:24

Re: An anti-LLM policy?

Board footer