How is it considered clean room implementation when the model was near certainty trained on the original code base?
The legal intricacies of this are of course interesting and relevant.
But what are the social ramifications if this kind of thing is deemed acceptable? It feels like it would effectively be the end of OSS licensing, because it's pretty straightforward to do this for any project.
Any company that wanted a proprietary copy of a program could in theory follow this same technique, with relative ease. That feels wrong.
So maybe we need to re-think the "copyrightable API" and "clean room" legal concepts. How? I don't know. But a world in which OSS licenses are easily sidestepped feels like the wrong direction.
Is everyone here just ignoring the fact that the 12 year maintainer of this project performed the re-write? Companies cannot do this with relative ease unless they hire these OSS maintainers
I think that's one of the most important details here.
From a moral perspective Dan wanting to relicense the project carries much more weight give his many years of contribution.
From a practical perspective the reason the rewrite went so well (significant performance boost, virtually no duplicate code) speaks to his skill and experience with the domain.
It's also a cause of friction here, because the fact that he knows the codebase so well makes him less credible as a clean-room implementer - his own biological neural weights are deeply biased by what he's learned from that existing code.
He’s not claiming the rewrite is a clean room implementation. In fact he’s explicitly saying it is not:
> However, the purpose of clean-room methodology is to ensure the resulting code is not a derivative work of the original. It is a means to an end, not the end itself. In this case, I can demonstrate that the end result is the same — the new code is structurally independent of the old code — through direct measurement rather than process guarantees alone.
That's a good note, yeah. Dan isn't claiming a pure clean room approach here.
If it is straightforward then the project is maybe not that interesting anymore? If an LLM can just write it?
no. A clean room implementation requires the programmer to have never seen the original source code. This is never the case with llms
Armin Ronacher has a more interesting take on this than I do: https://lucumr.pocoo.org/2026/3/5/theseus/
How is it considered clean room implementation when the model was near certainty trained on the original code base?
The legal intricacies of this are of course interesting and relevant.
But what are the social ramifications if this kind of thing is deemed acceptable? It feels like it would effectively be the end of OSS licensing, because it's pretty straightforward to do this for any project.
Any company that wanted a proprietary copy of a program could in theory follow this same technique, with relative ease. That feels wrong.
So maybe we need to re-think the "copyrightable API" and "clean room" legal concepts. How? I don't know. But a world in which OSS licenses are easily sidestepped feels like the wrong direction.
Is everyone here just ignoring the fact that the 12 year maintainer of this project performed the re-write? Companies cannot do this with relative ease unless they hire these OSS maintainers
I think that's one of the most important details here.
From a moral perspective Dan wanting to relicense the project carries much more weight give his many years of contribution.
From a practical perspective the reason the rewrite went so well (significant performance boost, virtually no duplicate code) speaks to his skill and experience with the domain.
It's also a cause of friction here, because the fact that he knows the codebase so well makes him less credible as a clean-room implementer - his own biological neural weights are deeply biased by what he's learned from that existing code.
He’s not claiming the rewrite is a clean room implementation. In fact he’s explicitly saying it is not:
> However, the purpose of clean-room methodology is to ensure the resulting code is not a derivative work of the original. It is a means to an end, not the end itself. In this case, I can demonstrate that the end result is the same — the new code is structurally independent of the old code — through direct measurement rather than process guarantees alone.
That's a good note, yeah. Dan isn't claiming a pure clean room approach here.
If it is straightforward then the project is maybe not that interesting anymore? If an LLM can just write it?
no. A clean room implementation requires the programmer to have never seen the original source code. This is never the case with llms
Related: https://news.ycombinator.com/item?id=47257803
Recent SCOTUS refusal to hear appeal could mean that clean room implementations may not be license-able at all.
If Ai produced content cannot be copyrighted, can it be licensed?