David Chisnall (*Now with 50% more sarcasm!*) on Nostr: nprofile1q…vcj3j Stanford law professor Mark Lemley, who has represented AI ...

nprofile1qy2hwumn8ghj7un9d3shjtnyd968gmewwp6kyqpq7z6pyrpq0an5jc0vlqrtuymxeahl8245ha0q349x592pjzq3dkjsyvcj3j (nprofile…cj3j)

Stanford law professor Mark Lemley, who has represented AI companies in copyright lawsuits, told The Atlantic that he isn’t sure whether an AI model “contains” a copy of a book or can reproduce it “on the fly in response to a request.”

This is the kind of nonsensical defence that makes me really question the competence of the lawyers working for the copyright owners. In a black-box system, if the two are indistinguishable then why would it matter from a legal perspective?

If I take a book and XOR every bit in it with 1, then write it to my disk, my disk does not contain a copy of the book. But I can reconstruct the original book on demand.

There is no defence for books in LLMs that would not also apply to MP3 compressions of CDs, except that MP3s are a 1:1 thing. But that defence would apply if I take a CD library, compress each track as MP3, then use zstd in dictionary mode to compress the MP3s.