EDIT: I misunderstood exactly what role "generative" AI is having in the Linux space, I'm sorry for misleading people. I got quite upset upon learning it was even within arms reach and failed to get a good enough grasp on it before writing this post. I'm still pissed, so I'm leaving the post up, but hopefully it's not factually inaccurate anymore. Let me know if I'm still missing something, please.
Sorry for my frequent posts here recently, I don't mean to spam at all, but I'm so mad about this. I heard a few rumors about AI being used for something involving the Linux kernel, but never enough about it to think it was actively happening, for some reason.
I switched to Linux last year once I realized Windows 11 would have revolting LLM shit built in. I had finally had enough of Microsoft and their exploration and spyware.
I went cold turkey, started using open source programs for everything, got involved in the Fediverse (as you can see), I altered all my creative workflows to use only FLOSS software, down to the individual plugins in my Digital Audio Workstation. It's been a big effort, but I'm happier with my life technologically then ever before. Now it appears that no matter what distro we choose, the developers of the kernel itself are attempting to incorporate LLM reviews into their workflows, for some damn reason.
I can't say this in a way that doesn't sound overly dramatic so I guess I'll just lean into it: how fucking dare they even consider something so verifiably harmful to society and the planet as a whole? It's total hypocrisy and goes against the reasons Linux exists. I genuinely feel betrayed. Why haven't I seen more outrage over this? istfg it's like nothing is safe anymore. Even most FLOSS projects are too cowardly to say they don't support "gen" AI explicitly, and I often have to ask them myself. Fuck.
Where have you seen LLM generated code being merged into the kernel?
https://codeberg.org/small-hack/open-slopware#operating-systems
Did you read Linus’ response on the mailing list? From that linked article
https://lwn.net/ml/all/CAHk-=wj3fQVEcAqy82JnrX2KKi4NjnEGGSH2Pf_ztnLCcveWkQ@mail.gmail.com/
Given his sole control over what gets merged, thats all that matters.
also note hes discussing ai as a tool for reviewing patches. nothing about ai actually writing code.
make of that what you will
Linux, of all projects, should be opposed to these kind of things as a whole. Sure, we could argue it's not as bad, but I'm not comforted by that statement. The fact that those in charge don't see or care about the obvious problems is shocking.
Im not trying to say one way or the other, or take anyone’s side here.
just putting into context “ai generated code in the linux kernel” isnt whats currently happening.
unless you have evidence otherwise.
edit - KDEs response maybe clarifies “ai” to me in this discussion.
LLM providers like OpenAI are scum. But the general technology around “ai” isnt as bad as that
Right. Sorry, I'll try to make the post clearer bc I'm not trying to mislead anyone, I'm just so upset that this is even being entertained by Linux devs, it's boiling my blood.
Just edited my reply for context. Hopefully it explains at least my personal view.
LLMs provided by billionaire techbros? Burn that shit to the ground.
The scientific idea and application of ai when its helpful and relevant I dont see a problem.
The difference being no vibe coded ai generated bullshit ends up in the kernel. The use of this technology elsewhere can be completely fine, if its treated correctly.
But Im totally on your side with “openai llm vibe coded slop should never ever land in the kernel”. And I trust linus on that. given his history with regular 100% human maintainers, he wouldnt let that garbage slide.
“ai” as a term has become synonymous with openai, anthropic, gemini. theyre just LLM products sold by companies. They should never be near real critical production systems. But the wider scientific/technology side could be applied ethically - without using those LLM products
I like your thought process, but struggle to agree with you because I don't believe there is a way for this kind of technology to be ethical without being rebooted (haha) from the ground up. Even so, about the only thing I think it should be used for is double checking what humans have already done, or perhaps printing text from human speech for quick and dirty subtitles. However, all that is assuming that the data it is working off of was not stolen, and that it is not able to "generate" anything "new" because that's just theft and exploitation.
Youre right in it needing to be rebooted, and thats sort of what im getting at. The technical side (the idea of how models work) is fine. Problem is big companies fed it with stolen content. And instead of optimising its performance they instead decided to burn even more of our environment.
I wholeheartedly agree that is evil.
But assume an entirely “untrained LLM” - its fed with truly proven open source, willingly contributed information, which then runs locally? I think thats ok.
Double checking the output - thats absolutely correct. I work in software development. I have dabbled with ai code tools. I would not let them ever touch my projects directly. Using one as a sounding board was slightly helpful. but knowing the cost and impact of these models currently, its not worth it.
Use gallons of water, and a dozen trees to feed the plagiarism machine to do my job for me? Never ever. fuck that.
I don't think it's particularly off brand for Linux.
Have you considered moving to a BSD?