Acknowledging these LLMs are not human, but isn’t this kinda what humans do? Take in lots of different examples and produce something similar but distinctly different and not paying royalties or being considered plagiarism.
To give another "scale makes all the difference" analogy. Very early on when Google Maps Street View was originally released, there was debate about whether it was OK to show individuals faces. In the US at least, the argument went something like "People are out in public, they should have no expectation of privacy. If it's legal for a person to take a picture of someone else on the street (which it is in the US), why should Street View have any privacy concerns?"
The difference is that while I may expect that other people may see and even take a picture of me if I'm outside, that's different from making my picture searchable, geolocated, instantly available to billions of people across the world and online forever. And I think it's totally rational to think these 2 different situations require different approaches.
Lots of our previously common, intuitive notions of what is OK can change greatly when large scale automation enters the mix.
Mechanizing that process so that it can be wielded at scale and automated is a pretty significant "change to the contract", and worth exploring what compensation is fair under these new circumstances.