I feel like the author is missing a huge point here by fighting this. The entire reason why GPL and any other copyleft license exists in the first place is to ensure that the rights of a user to modify, etc a work cannot be ever taken away. Before, relicensing as MIT - or any other fully permissive license - would've meant open doors to apply restrictions going forward, but with AI this is now a non-issue. Code is now very cheap. So the way I see this, anyone who is for copyleft should be embracing AI-created things as not being copyrightable (or a rewrite being relicensable) hard*.
No right to relicense this project
81–90 of 388 posts
Re: No right to relicense this project
#82Earlier quoted context omitted.
This seems right to me. If you ask a LLM to derive a spec that has no expressive element of the original code (a clean-room human team can carefully verify this), and then ask another instance of the LLM (with fresh context) to write out code from the spec, how is that different from a "clean room" rewrite? The agent that writes the new code only ever sees the spec, and by assumption (the assumption that's made in al…
How would a team verify this for any current model? They would have to observe and control all training data. In practice, any currently available model that is good enough to perform this task likely fails the clean room criteria due to having a copy of the source code of the project it wants to rewrite. At that point it's basically an expensive lossy copy paste.
You can always verify the output. Unless the problem being solved really is exceedingly specific and non-trivial, it's at least unlikely that the AI will rip off recognizable expression from the original work. The work may be part of the training but so are many millions of completely unrelated works, so any "family resemblance" would have to be there for very specific reasons about what's being implemented.
Re: No right to relicense this project
#83Earlier quoted context omitted.
So effective, LGPL means you freely give all copyright for your work to the license holder? Even if the license holder has moved on from the project? What if I decide to make a JS or Rust implementation of this project and use it as inspiration? Does that mean I'm no longer doing a "clean room" implementation and my project is contaminated by LGPL too?
The standard way of "relicensing" a project is to contact all of the prior code contributors about it and get their ok. Generally relicensing is done in good faith for a good reason, so pretty much everyone ok's it. Trickiness can turn up when code contributors aren't contactable (ie dead, missing, etc), and I'm unsure of the legally sound approach to that.
The legally-sound approach is to keep track of your actions, so you can later prove you've made "reasonable" efforts to contact them.
Re: No right to relicense this project
#84Earlier quoted context omitted.
Interesting case, IANAL but sounds legal and legit. The AI did not have expose to the backend it re-implemented. The API itself is public and not protectable.
OTOH as of yesterday the output of the LLM isn't copyrightable, which makes licensing it difficult
As of yesterday?
Re: No right to relicense this project
#85Sounds like they didn’t build a proper clean room setup: the agent writing the code could see the original code. Question: if they had built one using AI teams in both “rooms”, one writing a spec the other implementing, would that be fine? You’d need to verify spec doesn’t include source code, but that’s easy enough. It seems to mostly follow the IBM-era precedent. However, since the model probably had the original c…
This seems right to me. If you ask a LLM to derive a spec that has no expressive element of the original code (a clean-room human team can carefully verify this), and then ask another instance of the LLM (with fresh context) to write out code from the spec, how is that different from a "clean room" rewrite? The agent that writes the new code only ever sees the spec, and by assumption (the assumption that's made in al…
Somewhat annoyingly, there's been research that suggests that models can pass information to each other via (effectively) steganographic techniques - specific but apparently harmless choices of tokens, wordings, and so on; see https://arxiv.org/abs/1712.02950 and https://alignment.anthropic.com/2025/subliminal-learning/ for some simple examples.
While it feels unlikely that a simple "write this spec from this code" + "write this code from this spec" loop would actually trigger this kind of hiding behaviour, an LLM trained to accurately reproduce code from such a loop definitely would be capable of hiding code details within the spec - and you can't reasonably prove that the frontier LLMs have not been trained to do so.
Re: No right to relicense this project
#86Earlier quoted context omitted.
OTOH as of yesterday the output of the LLM isn't copyrightable, which makes licensing it difficult
As of yesterday?
I think they mean this: https://news.ycombinator.com/item?id=47232289
Re: No right to relicense this project
#87Wow that's hot. I was not aware that you need to be "untainted" by the original LGPL code. This could mean that... All AI generated code is tainted with GPL/LGPL because the LLMs might have been taught with it
Not a lawyer, but that always seemed naively correct to me.
However, the copyright system has always be a sham to protect US capital interests. So I would be very surprised if this is actually ruled/enforced. And in any case american legislators can just change the law.
Re: No right to relicense this project
#88Earlier quoted context omitted.
OTOH as of yesterday the output of the LLM isn't copyrightable, which makes licensing it difficult
That's a very incorrect reading. AI can't be the author of the work. Human driving the AI can, unless they zero-shotted the solution with no creative input.
The human is still at best a co-author, as the primary implementation effort isn't theirs. And I think effort involved is the key contention in these cases. Yesterday ideas were cheap, and it was the execution that matters. Today execution is probably cheaper than ideas, but things should still hold.
Re: No right to relicense this project
#89Isn't the real issue here that tons of projects that depend on the "chardet" now drag in some crappy still unverified AI slop? AI forgery poisoning, IMHO.
Why does this new project here needed to replace the original like that in this dishonourable way? The proper way would have been to create a proper new project.
Note: even Python's own pip drags this in as dependency it seems (hopefully they'll stick to a proper version)
Re: No right to relicense this project
#90Sounds like they didn’t build a proper clean room setup: the agent writing the code could see the original code. Question: if they had built one using AI teams in both “rooms”, one writing a spec the other implementing, would that be fine? You’d need to verify spec doesn’t include source code, but that’s easy enough. It seems to mostly follow the IBM-era precedent. However, since the model probably had the original c…
Not if the codebase was included in training the implementer.