In my experience code generation is most affected by the local context (i.e. the codebase you are working on). On top of that a lot of code is purely mechanical - code generally has to have a degree of novelty to be protected by copyright.
If you had a contributor that plagiarized at a 2-10%, would you really go “eh it has to have a degree of novelty to be a problem” rather than just ban them? The different standards baffle me sometimes.
So do you want to legally review every line by an LLM to see if it meets the fair use criterion, since you have to assume it was probably stolen? And would you do this for a known plagiarizing human contributor too…?
So what does the signed-off-by magically solve here, that doesn’t require either you or the contributor to legally review every line by an LLM? If you’re not a lawyer, is your contributor going to be one?
They don’t have to be. They know what they asked the LLM to do. They know how much they adapted the output. You usually have to work to get the models to spit out significant chunks of memorised text.
Imagine how broken it would be otherwise. The first person to write a while loop in any given language would be the owner of it. Anyone else using the same concept would have to write an increasingly convoluted while loop with extra steps.
Where are you seeing the 2-10% figure?
In my experience code generation is most affected by the local context (i.e. the codebase you are working on). On top of that a lot of code is purely mechanical - code generally has to have a degree of novelty to be protected by copyright.
If you had a contributor that plagiarized at a 2-10%, would you really go “eh it has to have a degree of novelty to be a problem” rather than just ban them? The different standards baffle me sometimes.
You can find various rates mentioned here: https://dl.acm.org/doi/10.1145/3543507.3583199 and here: https://www.theatlantic.com/technology/2026/01/ai-memorization-research/685552/
If the 2-10% is just boilerplate syscall number defines or trivial MIN/MAX macros then it’s just the common way to do things.
So do you want to legally review every line by an LLM to see if it meets the fair use criterion, since you have to assume it was probably stolen? And would you do this for a known plagiarizing human contributor too…?
No, that’s why the author asserts that with their signed-of-by. It’s what I do if I use any LLM content as the basis of my patches.
So what does the signed-off-by magically solve here, that doesn’t require either you or the contributor to legally review every line by an LLM? If you’re not a lawyer, is your contributor going to be one?
They don’t have to be. They know what they asked the LLM to do. They know how much they adapted the output. You usually have to work to get the models to spit out significant chunks of memorised text.
I don’t have much more to say other than I doubt the data backs up what you’re saying at all.
Imagine how broken it would be otherwise. The first person to write a while loop in any given language would be the owner of it. Anyone else using the same concept would have to write an increasingly convoluted while loop with extra steps.
Sounds like an origin story for recursion.