I think what it’s saying is that using the books for training is fine, but only if you legally purchased a copy. Downloading the books to avoid paying the author is definitely illegal. So that part of the ruling seems slightly misleading as we knew that part.
So can I buy a copy of, say, The Da Vinci Code, use it to train “DaVinci-GPT”, a model trained to recite all chapters of the book and distribute that model for free?
I think what it’s saying is that using the books for training is fine, but only if you legally purchased a copy. Downloading the books to avoid paying the author is definitely illegal. So that part of the ruling seems slightly misleading as we knew that part.
So can I buy a copy of, say, The Da Vinci Code, use it to train “DaVinci-GPT”, a model trained to recite all chapters of the book and distribute that model for free?
Correct. From the ruling you don’t pirate the book and from other rulings that I remember, AI output can’t be copyrighted on their own. Neat isn’t it?