"Lossly storing IP in LLM itself, a" - that part I'm inclined to agree with.

But it's debatable if that's the case.

Google stores copyrighted content and produces in in their product.

Also - it's fair game to use snippets of things here and there, if the derived work is novel, which I think it is for LLMs, mostly.

I do agree though, that we ought to draw the line somehow.