People get way too hung up on what current copyright law is and whether it supports or condemns LLMs trained on copyrighted material. IMO, who cares what the current law is? Existing laws were made for past technologies and for the current limitations of humans. It's nonsense to apply the same rules to new devices that don't have the same limitations.

I can pay to read all of the GoT books and then tell anyone who asks who kills Dumbledore or whatever. That's information you would have to pay the author for, but anyone can get it for free from me because I already read the book. This is acceptable because we assume that #1 I paid to read the books (or my library paid to acquire a copy) and #2 I won't be able to literally regurgitate the entirety of the books to every person in the world simultaneously. For everyone in the world to get that 2nd-hand enjoyment, enough people would have to pay to read the books that the author would make enough money to be happy.

The situation with LLMs is clearly different. The ratio between the amount an LLM compensates an author vs. how much many people it can share derived content with is off the charts compared to the same for a person.

IMO, there's no need to argue whether LLMs are being treated differently from humans w.r.t. copyright. LLMs have different capabilities than humans. Copyright at present is optimized for humans, and should be updated to address the implications of LLMs' capabilities.