I believe FreeBSD currently doesn’t allow LLM’s either, at least for now due to licensing concerns (I get the vibe they’d be open to it if it was deemed ‘safe’ from a licensing standpoint, so if the court ever made a decisions that makes them feel safe, they’d probably open it up to it).

Realistically though, both BSD and Haiku would be a pretty drastic downgrade in overall usability from reduced software and hardware support compared to Linux, and at least at this point, that’s a bit too big of a sacrifice to make :(

I plan to start probing other DE’s who have yet to make a public stance on LLMs in hope of finding at least a few that can be considered a safe haven.

Agreed, but this is a video based community, and unfortunately Peertube is not yet used enough to completely replace YouTube.

In the meantime, there are ways to avoid YouTube’s tracking, such as Freetube, Invidious instances, NewPipe, and Grayjay (grayjay is not GPL, but it is source available, and can sometimes work when the others are temporarily blocked by youtube and need to update a new solution).

Freetube videos cannot be linked to, it’s up to each individual to use manually on their end.

Invidious instances can be linked to, but unfortunately instances die fairly frequently, and links using them often die with them after a couple months or more, requiring the poster to search out new instances to replace them (having to comb through my old posts to replace the links with working ones eventually turned me off using it, I’m afraid).

KDE unfortunately is also accepting AI/LLM code contributions, as is LXQt and Gnome.

This news has made me realize it’s not going to be trivial to avoid AI slop FLOSS projects, we’re likely going to need to collectively collate a list of known anti-AI projects, and probe projects that haven’t taken a public stance to find out what’s safe and what’s not.

An unfortunate update; a user from the other thread on this went and asked LXQt developers their stance on LLMs, and they responded they accept AI contributions, and will not ban the practice. :(

Ah man, that absolutely sucks. I went and read all of their responses. You did a fantastic job presenting a compelling case to them, but for the most part they seemed to not want to engage with the meat of your points, or acknowledge your distinction between code output and AI bug searching. And odd of one to equate AI code as equivalent to code coming to them in a dream, yet never address the copyright angle. Thanks for taking the time to engage with them to find all that out.

I’m starting to see that avoiding slop even in the FLOSS world isn’t going to be trivial, and more are jumping on board than I anticipated… I’m thinking it may be necessary and worthwhile to probe as many widely used FLOSS projects as possible on their stance to then compile the results into a list on codeberg (so others can contribute their findings as well), to make it easier for people to find software that refuses to use AI.

Maybe calling it the Slopless Software Project?

Gloomwood is the only close spiritual successor, but its been in early access for a long time (though it does still get substantial updates).

My crumpet

Agree 100%.

Honestly, as a silver lining, this is likely to finally make studios pull back from graphics being the end all be all, and instead focus a bit more on art styles that run well on weaker hardware.

The endless pursuit of graphics has had drastically diminishing returns since the PS3, it’s about time the industry focused on optimization and doing more with less for once.

It’ll let our hardware stay relevant, and it’s better for the environment if we make it last as long as possible anyway :)

Down with consumption, and bring on the age of Permacomputing, I say!

This comment in a different thread from someone who used to work for them gives some great background as to why.

The other day I tried installing Star Wars Dark Forces 2 from disk, complete with a ‘windows 95/98 compatible’ blurb on it. Heroic launcher got the installer running without issue and within a few minutes it was playing.

I ended up using the Openjkdf2 engine instead for some nice quality of life improvements, but it was still quite something to see Linux run a piece of software from 1997 so effortlessly.

I’m not sure a high temperature oven with a convection fan can truly be “buy it for life”

This community also allows for things that won’t literally last forever. Just lasting significantly longer than average is enough :)

It is very unfortunate there wasn’t more standardization around compositors between all the major desktop environments, that part of it does suck.

This is a direct quote from the linked blog:

Occasionally I’ll have it write more substantial code for me, usually for bug fixes, and occasionally for small features. I find it does a pretty decent job here, but I keep it on manual mode and often ask it to write the code in a different way, or take a different approach.

He is generating code and using it directly. AFAIK, being on manual mode in Claude just means approving or rejecting the each step of generated code, but it is generated and the dev appears to be using that generated code, which is introducing the risk of unknowingly using and shipping non-open source copyrighted code that the LLM trained on.

Whether he’s doing it blindy or paying close attention, or telling the AI to re-write it or code it in a different way, does not minimize the copyright infringement risk of the output of the code that is used. Only rewriting the output by hand would minimize that risk.

Manual mode just means he’s approving, rejecting, or asking it to do something different for each step, but the output is still ultimately from the LLM.

But he is still generating decent amounts of code with it as well, as he states:

Occasionally I’ll have it write more substantial code for me, usually for bug fixes, and occasionally for small features.

Generating that code is injecting copyright risk into the compositor.

Thanks for sharing your experience, LXQt is definitely looking the most promising :)

I haven’t heard of it happening and until it does it’s basically a non issue.

Projects that operate on that assumption are essentially sticking their head in the sand and hoping for the best. LLM output is on particularly shaky ground regarding copyright and it’s vulnerability to being sued for plagiarism, see here: https://slrpnk.net/comment/23533549

I think it’s a particularly dangerous gamble for projects based in the US, where the court system is extremely pro-corporation, making it likely to be much harder for a small FLOSS project to defend itself successfully.

Using the internet in any capacity directly helps the financi situations of the big tech companies

If someone gets their internet in the US, it is likely supplying a mega corporation with a monthly revenue stream, as internet providers are an oligopoly here. In other countries, this is often not the case.

Internet access in the modern day is, for the most part, extremely difficult to avoid needing in order to function easily in society.

In comparison, Corporate LLMs are extremely easy to avoid, and all of society got along fine without it just a few short years ago.

Is it not worthwhile to lower your fossil fuel use where possible, despite many in the US needing a car due to terrible public transport? Is it not also worth lowering animal meat consumption where possible for both animal welfare concerns and environmental concerns? I think there’s equal merit in depriving as much capital (and thus power) from pro-fascist tech corporations as possible, even if it is not total.

I agree, that would be good to know.