One main thing that this brings to mind is if an LLM can ever actually create a clean room implementation of a piece of open source software, given that there i

One main thing that this brings to mind is if an LLM can ever actually create a clean room implementation of a piece of open source software, given that there is a near certainty that the software was used in its training data. Therefore it has seen it and remembered it, and could if appropriately prompted recreate the code verbatim. This can also apply to people, either if they have seen the code previously and therefore are ineligible to write the code for a clean-room implementation, or it gets murky when the same person writes the same code twice from their own knoeldge, as in the Oracle Java case. Coming from a professional programming perspective I can totally see the desire to have more libraries written in permissive licences like BSD or MIT, as they allow one like myself to include them in commercial closed-source products without needing to open source the entire codebase. However I find myself agreeing with the article in so far as this LLM generated implementation is breaking the social contract for a GPL/LGPL based library. The author could have easily implemented the new version as a separate project and there would not have been an outcry, but because they are replacing the GPL version with this new one it feels scummy to say the least.