A while back I tried to get ChatGPT to recite the text of Harry Potter and the Philosopher's Stone to me, as a test of just how much copyrighted text it's willing to recite.

It got partway through the first sentence before freezing up, presumably due to a sensitivity to copyright.

So I suspect that at least OpenAI are taking significant steps already to prevent their systems from reciting copyrighted text verbatim.

But it doesn't copy full sentences. If they did, maybe they wouldn't be such black boxes. They build this utterly insanely huge matrix of data that are basically just weights for parameters and there's billions of parameters (which make up the entirety of what the LLM "knows" or "can know"). It's closest to text prediction. Even though it doesn't know full sentences, if the sentence was used enough times, it can predict the rest of it. It can even do that without having scraped a book, simply because it scraped something else (likely many something elses) that had the quote.

Even if I illegally download a book about carpentry, and learn how to build a doghouse from it, and go into the business of building doghouses, the extent of my liability to the copyright holder does not include my entire doghouse-building revenue.

Even if I subsequently teach other people to build doghouses, if I'm not further copying the actual contents of the book, I am not further liable for copyright infringement.

Copyright is actually pretty narrow, and should not be construed to give authors or publishing companies unbridled control over the ideas or knowledge contained in works.

Yes, but it isn’t legal to download a repository of pirated books to read and learn from.

Sure, but that still doesn't change any of the above statements. If I steal a book from a library, read it... You get the point. All you can get me for is for... What exactly? Cost of the book + maybe a civil penalty? This is going to be a nothing burger for these writers if they're hoping for a payday. Further how do we know what specific repository that the AI got it's content from? It could be that the content it got was from some forum of a person summarizing a chapter + a review for the book + . There's no evidence I've seen thusfar that any of these AI systems are accessing books illegally to begin with. Or that those books were the only source that it derives its responses from.

The AI isn't reproducing the book and thus isn't violating copyright as literally everything it will produce is derivative which is protected. Unless you can get the AI to recite a book back verbatim... Which I've not been successful in doing personally... and I've seen no evidence of anyone else doing either.