It's an unfortunate outcome. Now to be a big player in AI, you have to have enough capital to buy your own library worth of books and digitize them. (Fun fact: a pallet of books is called a "gaylord," and they buy hundreds of gaylords.)

I created books3 to help settle the question of whether AI companies should be allowed to train on books. The outcome of "it's okay to pirate books as long as you're only training on them" was a long shot, but it would've let individual hackers train their own AI models (assuming access to sufficient compute, which you can get e.g. via https://sites.research.google/trc/about/).

Now we're in a world where you have to have dozens of millions in capital to do substantial work.

I heard at one point Eleuther was gathering public domain training data. I wonder if they ever built a corpus large enough so that training on books doesn't really matter...