Comment by g-b-r
8 hours ago
The prospect of being sued because the LLM outputted training code you're not allowed to use, or to relicense, is exciting?
Or to be forced to change license to large portions of your project?
With Linux I meant the Linux kernel
Copyright is based on laws, and laws can be changed or clarified. Given the amount of overinvestment in AI, I expect them to change. The EU is already making moves in this direction: https://noyb.eu/en/ai-eu-member-states-approve-digital-expro...
I expect the overall copyright regime to be weakened over time due to the effects of AI. It’s clear that the intent is to replace copyright gatekeeping with model/compute gatekeeping. This is a battle between two groups I greatly dislike, and I don’t really side with either, but there may be a small benefit for the average person who just wants to “download a car”.
It could be, but:
- For now the laws are in force, and I don't even know of discussions to change them (that EU proposal only wants to allow training on personal data)
- changing intellectual property rights law would typically require changing international treaties
- however much companies are in love with AI now, I'm not sure they'd support abolishing legal protections on which their business relies
You're being pretty, I'm not sure the right word, scaredycat?, about licensing. Licensing doesn't matter unless somebody sues you, which they won't in most cases. The AI era has exposed this truth that you're extremely unlikely to get sued about copyright violation - so unless you're the kind of person who avoids car trips because you could crash, you can ignore it.
Uhm, you're being pretty, careless?
I'm not sure that that's how legal departments think.
> The prospect of being sued because the LLM outputted training code you're not allowed to use, or to relicense, is exciting?
The problem with this is that if someone sues you for this they have to somehow prove that you actually infringed. They can't just point to the code and say "See, mah, infringement!" The legal system does recognize that it is very possible for two people, completely independently, to arrive at a nearly identical solution.
Of course we're talking of code for which that's implausible.
I think in most cases twenty lines of identical code are more than enough for that to be the case (at least if it contains a few non-single-letter identifiers).
I mean, if it uses the exact same identifiers and maybe even comments, then sure, that's pretty good evidence. However, 99.99999999999 percent of the time this will not be the case. As in, the super-super-super-majority of LLM-generated code is customized for the specific use case it's being employed in. As such, claiming any kind of infringement, let alone proving it, will be insanely difficult.
1 reply →
How many cases of an LLM outputting training code have you heard of? let alone training code that you are not allowed to use or relicense?
Why are proprietary software developers entirely unbothered by this?
They did in 2023 and the AI thieves built logic to rephrase the plagiarism so it isn't immediately detectable.
Proprietary developers don't care because their code wasn't trained on. Is that a trick question?
They do care now that reverse engineering is possible.
Can you explain further? What was the code affected?
> Why are proprietary software developers entirely unbothered by this?
Because the chance of someone auditing the actual source code is near zero, but for open source software it's near 100% that someone or, worse something aka an AI bot looking for a revenue stream, will have a look at your code. And EVEN IF you get caught stealing by whatever random incident (say, your code repository being 0wned) - it's a matter of a settlement payment. If you are not a company, but an individual person... well, just look at what happened to Aaron Swartz RIP.
And for the WINE/ReactOS/console homebrew folks, it's vital to be able to claim no code contributor ever got exposed, even indirectly, to proprietary code.
On the other hand it could be very expensive if caught. It could be very difficult to remove the code too.
Your point about being caught is valid, but it depends on AIs producing identifiable chunks of code, and that is incompatible with your license. AI training is going to lean towards open source, because its available to train on.
I also think the widespread use of AI will mean accidental copying will become more acceptable. Even as it is, its going to be regarded as an accident.
I heard many cases of LLMs outputting their training data verbatim; they sure don't check that such data can be legally used by you before doing it.
As others said, for closed source software it's a lot harder to detect plagiarism.
And it's in general very hard to detect plagiarized code; there's some product that does it, but it of course has only a partial inventory of the existing open-source code.
I think that some big company does use those products, though.