Oh well. I’m too used to the “/s” for recognising sarcasm.

Genuine? Almost all his arguments are strawman and he contradicts himself constantly.

One way I could see this being enforced is by mandating that AI models not respond to questions that could result in speaking about a copyrighted work. Similar to how mainstream models don’t speak about vulgar or controversial topics.

But yeah, realistically, it’s unlikely that any judge would rule in that favour.

Alsup? Is this the same judge who also presided over Oracle v. Google over the use of Java in Android? That guy really does his homework over cases he presides on, he learned how to code to see if APIs are copyrightable.

As for the ruling, I’m not in favour of AI training on copyrighted material, but I can see where the judgement is coming from. I think it’s a matter of what’s really copyrightable: the actual text or images or the abstract knowledge in the material. In other words, if you were to read a book and then write a summary of a section of it in your own words or orally described what you learned from the book to someone else, does that mean copyright infringement? Or if you watch a movie and then describe your favourite scenes to your friends?

Perhaps a case could be made that AI training on copyrighted materials is not the same as humans consuming the copyrighted material and therefore it should have a different provision in copyright law. I’m no lawyer, but I’d assume that current copyright law works on the basis that humans do not generally have perfect recall of the copyrighted material they consume. But then again a counter argument could be that neither does the AI due to its tendency to hallucinate sometimes. However, it still has superior recall compared to humans and perhaps could be the grounds for amending copyright law about AI training?