I assume that google had legal access to those books. In the case of GPT-3 derived models, they contain common crawl and webtext2 corpuses, which may include la

I assume that google had legal access to those books. In the case of GPT-3 derived models, they contain common crawl and webtext2 corpuses, which may include large amounts of pirated content (books and magazines uploaded in random places, paywalled content that's been uploaded elsewhere).