- cross-posted to:
- linux@programming.dev
- cross-posted to:
- linux@programming.dev
You must log in or register to comment.
Yeah, LLM code is fundamentally incompatible with FOSS licensing because you fundamentally cannot vet the codebases that it was trained from. The fact that this is ignored in most other places speaks to how well the LLM propaganda and astroturfing has worked.
The main confusing (or confused?) actor in the FOSS space in relation to this probably really is the Linux kernel.
I’ve seen so many projects reference significant LLM patches as acceptable “because the kernel does it”. It continues to blow my mind that the kernel would accept such a risk.
The kernel absolutely does not do that. What theyre referring to is linus saying he was fine with people using claude for analysis purposes or to double check your pr, not submitting whole ass ai prs
It seems like they do though, because
-
it’s explicitly allowed
Code or other content generated in whole or in part using AI tools can be contributed to Linux Foundation projects.
and beyond that,
-
some numbers suggest it’s highly likely that it’s happening with no public concern or pushback from the kernel leadership.
I’ve also brought up these concerns on the mailing list, with apparently no response from the maintainers, even though Linus was CC’ed here by another kernel dev. Specifically, my suggestion to not allow AI code submissions resulted in no response.
I’m not saying I would be owed a response. But the implications of that seem pretty clear.
-
I am far and away not an AI enthusiast but I don’t understand specifically the copyright issue with respect to “trained on x data”
Because we humans who contribute to FOSS are also generally trained on proprietary data (gaining experience with non-foss systems at companies).
I get the “who owns it” argument since the person didn’t create it.
Fuck LLMs for drying up groundwater and stealing electricity from families though.
We humans are not Ai. Don’t mix that up. In example just because you can write a script to scrape data from all Github, does not mean you can use the scraped data as you like without giving credit and respecting the license. But that is what Ai programs do. LLMs are not human and have not the same human rights.
I don’t understand specifically the copyright issue with respect to “trained on x data”
It’s a question of whether the output of the LLM is a derivative work of the training data. From a copyright standpoint, if the work is derivative (and not fair use), then it needs to follow the terms set by any licenses you have to its parent works.
I don’t think it is just propaganda. Most people (especially non coders) do not understand licensee and copyright, or Open Source and GPL at all. Or don’t care anyway. Sure there is some propaganda, especially from big tech companies, who don’t want to talk about the issues because they (think) to benefit from the tech. I just think its worse than “just” propaganda, with the previous mentioned issues.
considering all the “no copyright infringement intended!!” you see on entire movies posted on youtube, yeah i don’t think 90% of people have the slightest clue how copyright works, it’s just magic
Also uploading an entire film or music and then saying its not their work, they just upload it and all credit goes <whatever credit it is>. I’m not sure if those people actually think how this works or if this is just a disclaimer. Even as a disclaimer this is not how they get a license. Most people just don’t understand their most basic rights.
Yeah, LLM code is fundamentally incompatible with FOSS licensing because you fundamentally cannot vet the codebases that it was trained from.
Doesn’t really have anything to do with what it’s trained on. The fact is, that you cannot copyright code written by an LLM, and copyright is required for copyleft licenses to be enforceable.
Sad to see that it’s accepted outside of the test cases at all. But good to see they’re severely limiting it, at least.
Why do I find even minor LLM changes sad?
-
Because it moves the goalpost about what is okay to copy. If you brought a one-line change to a FOSS project in the past that you took out of the leaked MS Windows source code, you would have been scolded for risking such an explosive origin for such low gain. Nowadays with LLMs people seem to be trying to make that the new normal. I don’t think that’s a good path to take for the ecosystem.
-
And because no project should have to think about what’s “copyright significant”. If you reach that state, I feel like you should perhaps just reject it and have somebody rewrite it cleanly.
-
get rid of the exceptions too pls
