Vibecoding gets Emacs patch rejected
11 points by puhsu
11 points by puhsu
What I learned from past conversation around LLMs, however, is that the doubts about LLM contributions are around them being “open enough” and “legal to use”.
When we’re talking about open-weight models, I find the argument about being open absurd.
I think the author might be misunderstanding what the "open" in "open weight" means. Just because the final matrix-mash is publicly available and can be somewhat fine-tuned, it doesn't mean the training material used to create it is/was open source too as far as I understand and OSI seems to agree. And if so, the question of copyright isn't at all resolved.
I dislike when people tell me I’m holding the stick wrong - especially when I work on something out of my own volition.
It's hard not to feel sympathetic towards someone who's trying to do good, expecting nothing in return, but GNU has its rules spelled out and if you go there and say "I only kinda broke them, here's my patch", don't be surprised if they turn you away.
It's not honesty that got OP rejected, it's doing what they were explicitly asked not to do.
The GNU Project has to consider global copyright concerns, not just US ones (which themselves aren't entirely settled). They want to be 100% certain that they own the copyright for significant contributions. The author's interpretation of the law here doesn't matter. The GNU Project is playing it safe.
That's not even considering the other concerns they likely have with LLMs.
And, honestly, the author's understanding of US law is incorrect based on current precedent[1]. In existing case law, LLM-generated code cannot be copyrighted, and thus also cannot be licensed--it's automatically in the public domain. Moreover, if a repo contains LLM-generated code but that code is not clearly delineated and marked as such, the entire repo is considered un-copyrightable and un-licensable.
In other words, if the author had lied in order to get the code committed, it would risk de-licensing the entire emacs codebase.
(all of that's aside from the arrogant-to-the-point-of-delusional stance of "IANAL but the lawyers are clearly wrong" while only looking at a single bit of relevant statute, while also somehow accusing the lawyers of hubris)
[1] - Current as of about two months ago, when I was having these conversations with my company's corporate counsel.
Moreover, if a repo contains LLM-generated code but that code is not clearly delineated and marked as such, the entire repo is considered un-copyrightable and un-licensable.
This seems... surprising to me. If I make a drawing, and then put the Mona Lisa in it, that doesn't make the entire thing public domain. Same for if I copy some existing public domain code into my own project without attribution.
This is the difference between the theory and the practice. If the code is not clearly delineated but it is known that it contains code which can not be copyrighted then the entire codebase has to be "treated" as un-copyrightable until you can clearly and definitively identify the parts that are and the parts that aren't. It's a keep yourself safe thing even if in theory parts of it are copyrightable it's not useful information when you can't identify which parts are and which aren't.
The issue is that it becomes difficult to prove that you didn't use LLMs for the rest of the project, so someone can argue that you generated the entire thing with LLMs.
I believe there was another article linked here at some point arguing for clear attribution for LLM-generated commits for this reason. Not only that, but also outlining exactly what you did in each commit and possibly including the prompts to prove substantial creative human input into the process.
Like most things related to LLMs, I don't believe the legal basis for this isn't firmly established yet.
"Prove that you didn't <fill in the blank>" is a tough thing. That's why, in sane circumstances, the burden of proof is on the accuser.
I don't believe using LLMs in one part of a project makes proving that specific negative any harder. I guess it may make some people want you to prove it to a greater degree.
Moreover, if a repo contains LLM-generated code but that code is not clearly delineated and marked as such, the entire repo is considered un-copyrightable and un-licensable.
That's a big [citation needed] from me.
I expressed it in a way that put the cart before the horse, apologies. And, to be clear, IANAL, just somebody who talked through this at length with lawyers.
I'm not sure "I could have lied and then it would have been accepted" is a meaningful argument.
Yeah, their exact same argument could be used on licensing. If you copy something from a proprietary codebase and lie that you wrote it, your patch will be accepted! But if you tell the truth, it's rejected. Therefore ... licensing is ... bad?
“…licensing is…bad?”
If you preface that with “non-copyleft” that’s kinda the FSF/GNU position, isn’t it?
I do not endorse the original post, I just think this was a funny choice of examples.
Haha, yeah; poor word choice I guess. You could just as well say "obeying the law is ... bad?" in this case.
Indeed. I hear that as an argument about no-LLM policies a lot. "But then people will just lie."
That's... not saying anything good about the people in question then, is it?
I don't think it does any good to harass maintainers for pro- or anti-LLM policies. They're doing the work. They get to choose which contributions to evaluate, accept, and reject.
But it's fine to complain, I suppose -- this person is voicing their position on their blog; it's how we hash things out. I only disagree with the positioning; it's not "honestly" that is in question here. Assuming the no-LLM policy has been announced, then it's on them that they chose to spend time with an LLM to try to come up with a contribution for a project that doesn't want such things.
That's no more about honesty than offering someone who's vegan a dish with meat or cheese and then complaining they wouldn't eat it because you were "honest" about the ingredients. "You wouldn't have known it had dairy unless I told you" is not a great look; neither is "you wouldn't have known I used an LLM unless I told you".
Agreed. I've suggested "Vibecoding gets Emacs patch rejected" as a new title. Honesty in admitting vibecoding is like honesty in admitting copyright infringement. It's not the honesty that's at the root of the rejection.
Feels like a textbook case of "remove hyperbole and moralizing" as suggested in the story submission guidelines.
Yep. Hopefully one or two more people will make the same suggestion, and the change will be applied.
Yet GNU believes that THEIR lawyers and THEIR opinion has the most weight. I won’t deprive them from the right of deciding for they own, but this lack of self-awareness is almost caricatural.
GNU (and the FSF) invest quite a lot in getting professional legal advise. This potential contributor is advising them to disregard that professional advice in favor of what some guy on the internet has to say.
If one is going to pay for professional advice, it's only sensible to follow that advice, and if you disagree with it, find another professional. Disregarding it in favor of random internet commentor advice would be stupid. Not "almost caricatural".