My argument was, that is was not obvious.
And your clean room design has also issues that are not obvious. You might not break "copyright" under certain circumstances, but you will break trademarks and other legal barriers, like patents. Also, this is just to circumvent having used a "copy". Which too would be circumvented by using a llm.
So how would a clean room aproach look like for an image or a work of art? Oh, and no, you do not break copyright by writing down a poem you remembered. You break copyright by publishing it, because you distribute a "copy" and you did not have the "right to copy". But this interpretation is different world wide.
Using a work to create another work without permission is not a copyright breach, unless you use exact portions of that work. It might be a licensing issue or other legal stuff. But not copyright. That clean room approach was for example used to recreate functionality of software code, because, of course, afterwards there were accusations that portions of the code were used.
So in other words: as I said, you can't copyright the knowledge how something looks. Or what it does in case of software. You might give a patent if you have a broken patent law.
Aso a llm basically uses the hash value of a work. Some earlier version might have had portions of original works inside the database, but when you read about the complaints against those systems it was about the content of those databases and decidedly not the output of the llm.
I stand by my opinion. It is not obvious how to handle llm generative ai systems. Neither legally, nor in society, nor in art. Or games. The emerging consens among players seems to be, to prefer human made art. The emerging consens among software developers seems to be, yeah, another tool to play with that can churn out templates in a hurry and find semantical errors in a programming language.