OpenAI now tries to hide that ChatGPT was trained on copyrighted books, including J.K. Rowling’s Harry Potter series::A new research paper laid out ways in which AI developers should try and avoid showing LLMs have been trained on copyrighted material.
The thing is that, for the AI to be trained on Harry Potter, someone must have bought the book.
I wouldn’t be opposed to use an AI and tell it “here. Read this book that I bought. Summarize it for me.” That would be fair game.
But if I ask an AI “did you read Harry Potter? What can you tell me about Voldemort?” That means the AI consumed the book.
Granted, someone on the internet could publish an essay about Voldemort and bake it available to everyone, so, hmmm… you may have a point. May.
Sure. That follows. I guess the point I’m considering is more broad. If you’ve sold some one a copy of a book, you’ve given up ownership of the information in the book. If you truly don’t want that information out there, don’t sell the book. Its an extreme take, but I don’t believe that ideas can be property in the same way as a shoe (or rather, I believe maybe they are in-fact the same kind of property as a shoe. You can’t own the principal of a shoe, simply the shoe its self.)