The article also mentions the idea that the model is merely extracting (uncopyrightable) facts, which is interesting, but might be a tough one to prove since LLMs have no way of establishing what are facts, and don't return facts by design.
The article also mentions the idea that the model is merely extracting (uncopyrightable) facts, which is interesting, but might be a tough one to prove since LLMs have no way of establishing what are facts, and don't return facts by design.