Torvalds' statement was incredibly disingenuous and rather sad. Every single commit is making a legal and political claim. Each commit is saying "this code is my original copyrighted work, and I am licensing it to the world per the requirements of the GNU GPLv2" (or "this code is someone else's work which was licensed under a compatible license, and I am integrating it into this GNU GPLv2 project").

That's not social warrior, that's simple facts. We don't need to try and guess motives or reasoning for people's contributions - that doesn't matter. What we have is an explicit declaration by virtue of the license, that the code being provided can legally be included in the GNU GPLv2 licensed "Linux" kernel. But with LLM generated code, this legal claim is dubious at best. For trivial (!= simple) code, one can probably argue that the LLM code will not be violating anyone's copyright, however for anything remotely novel, unless you are 100% certain that all of the training data was GPLv2 or compatible, it is not legally safe to use - and aside from that, even if all the training data was compatible with the license, the LLM will virtually always fail to properly attribute the source of its output, which is a requirement of - as far as I can tell - every license compatible with the GNU GPLv2. I cannot share GPL'd code and misrepresent it as my original work, I still have to make clear that it was someone else's creation.

LLM generated code is a ticking time bomb, and is going to poison projects.

Yes, it is possible to do this already without LLMs, but not by accident. A developer would have to make a conscious decision to violate someone's license in the past. Now it can happen without anyone being the wiser.

And as someone who very much uses GNU GPL licensed software because of its license, this worries me for the future of software. On technical merits, it doesn't matter, but if you care at all about the artful expression of software, as well as honesty about who created what, this should concern you.

▲

supriyo-biswas 12 hours ago | parent [-]

> LLM generated code is a ticking time bomb, and is going to poison projects.

Given the recent Anthropic ruling and OAI settlements, and the general stance of courts to look the other way when it comes to LLM use, I think an exemption for LLM-based software will be handwaved into existence (to my regret.)

	▲	Magicrafter13 12 hours ago \| parent [-]
		That certainly seems to be the way the courts are going, which is also quite troubling, considering a plain understanding of software licenses and how LLMs work should clearly demonstrate the issues I outlined above. I don't want to accuse the courts of intentionally siding with big corporations (I have no evidence to support such a claim anyway), but the only other option I see is that the courts are uninformed on either (A) the technology at play, or (B) the requirements of many software licenses. Though, this should also apply to works licensed under variants of the Creative Commons (with Attribution), since once again, you can end up using parts of a work that requires attribution without even knowing that it's someone else's work (or at least without knowing its license requirements). There might be a better argument for generative AI images being transformative (though I think still a weak argument since you could almost certainly produce a 1:1 reproduction of a piece of training data, much like LLMs have already been pushed to doing with snippets from books).