3 ms·
How is that any different from a clean room?
by xyzsparetimexyz 17d ago
How is that any different from a clean room?
- preisschild 17d agoWith clean room you just have access to the api specifications, not the internals.
- simoncion 17d agoYep. An entity in the "dirty room" reads the thing to be reimplemented and produces a document that thoroughly describes its behavior. That document is passed to the entity in the "clean room" whose only knowledge of that system is through that document. Reverse engineering is legal, plagiarism is not. Despite the fact that the raw output of the system is incomprehensible to humans, scanning a photograph of Mickey Mouse and running it through a lossy compression system like JPEG doesn't suddenly make it not a picture of Mickey Mouse. Similarly, running the code for a system through the lossy compression system that is LLM "training" doesn't suddenly obliterate that data and make that LLM a "clean room". If one has any doubt of that, remember that they are known to reproduce their inputs, even after all these years of work to make them not do that. [0] [0] <https://news.ycombinator.com/item?id=49727685 https://news.ycombinator.com/item?id=49727685>
- mitxela 16d agoPlagiarism is legal, actually. Copyright infringement is not.
- simoncion 16d agoFair-ish point. I was considering plagiarism in the context of computer programs, which is -at best- dreadfully difficult to do without also violating copyright and/or software licenses. I'd also rephrase your first sentence as "Plagiarism isn't illegal, actually.". Unless you're rich and/or very influential, plagiarism is a seriously bad thing to do.
- xyzsparetimexyz 16d agoOne llm writes the api spec, another writes the api from the spec. No shared context. Easy.