Comment by gruez

>You've already done it, got caught, now a fine. It does not mean that the LLM model has to be tossed out and destroyed with a new version trained up without that data. It just means can't is a very stupid word to imply here.

Yes, because most courts have ruled that training is legal as long as the source material was acquired legally. The AI companies were made to pay for the wrongs they did when acquiring the books, but it makes little sense to destroy all works that were built off the infringement, when they would be in the clear if they paid $15 (or whatever) for each book. It'd be like you torrenting college textbooks and getting caught, and then the book publisher demanding that you start over your college degree from scratch.