AI assistance when contributing to the Linux kernel https://github.com/torvalds/linux/blob/master/Documentation/process/coding-assistants.rst

Look like Linux foundation and big tech members like Microsoft, Google and others who are behind the Linux foundation drafted it. Basically it says that you can use AI/LLM, but you take full responsibility for your commits to the kernel and code must satisfy the license GPL-2.0-only. So more slop is allowed to add directly to the Linux kernel.

linux/Documentation/process/coding-assistants.rst at master · torvalds/linux

Linux kernel source tree. Contribute to torvalds/linux development by creating an account on GitHub.

GitHub
@nixCraft At the very least, they added a clause of a human reviewer, although not having any slop would be better.
@nixCraft doesn't requiring gpl-2-only licensing effectively rule out all the current models?
@nixCraft how can you put it under a licence if you don't have the copyright
@nixCraft Generated code is by definition not source, so it cannot satisfy the GPL

@criffer @nixCraft Does that depend on whether you can round-trip it?

Can you

* generate some code from a prompt
* fix the code manually
* tell the AI to adjust the prompt to generate the manually fixed version of the code next time?

@TimWardCam @criffer @nixCraft Perhaps, but then the word "generate" might be doing a lot of extra lifting?
@nixCraft it’s depressing, really. We’re fucked.
@nixCraft of course. Everything else would not make sense. The responsibility of the Linux kernel maintainers is not culture war. Also how would you prevent it? If someone would use an LLM to create a structure of a class, find a bug, fix a bug, write a function, write a comment - how would you know? Especially if its not only AI but combined with a developer? Would you also not accept source code because the IDE did some auto complete?
@nixCraft You can discuss where LLMs make sense and where not, but they are here to stay. Acceptance and adapting to the situation is more helpful then #hopium - and rules that demand #HITL are the right choice! Even if code reviewed happen AI assisted, especially for critical software like kernel modules, someone with a brain should be the one to take responsibility and approve the pull request.
@dag hey now buddy, that sounds an awful lot like a rational, pragmatic position. We don't much care for those around here.
@nixCraft hey @torvalds we dont want slop in the kernel
@nixCraft so keeping backup "pre-AI" copies of my favorite distro (and its repos) seems to make more and more sense...
@nixCraft Well well well... Let's see what will happen. Hopefully we still have BSD distros free from that AI slop
@nixCraft Cartels like ms and alphabet should be kept away from the Linux Foundation. They will destroy everything for their own interests. Community donations can keep free software alive
@nixCraft much as i hate AI, i have to admit a very, very vast majority of people are tripping over themselves trying to integrate it as deeply into their lives as possible. Such people will continue to attempt to bombard open source software with slop PR’s regardless of whether the project accepts those or not. This method of requiring human review and clear attribution for AI generated content is the only way to make sure people clearly mention when they used AI. If the kernel outright rejects AI slop, the AI-bros will just stop attributing generated slop. So maintainers won’t know if submitted code is slop or not without thorough manual review and maybe not even then.
@nixCraft is everyone forgetting Torvalds himself has talked about LLMs being an useful tool? this is not surprising at all
@nixCraft Can't we all switch to another kernel? A fork or something compatible?