alt Hacker NewsThis is shady. Copyrighters absolutely not get to control use of their copyrighted material when people mentally, sonically, or physically reproduce it for personal use.
It's absurd to say "you can't record this book to a friend or robot".
Nobody seems to actually reproduce the copyrighted materials.
High-dimensional eigendecompositions which underpin AI similarity are some of the most literally derivative materials of texts that you can imagine.
So you record a copy "for a friend" and then you sell lots of those copies as your business. All within your rights! What's mine is yours, my Comrade!
(my point being that it would be different if the product CommonCrawl provides were trained models, but this is not the case: its product is unlawful reproductions of copyrighted data for commercial use)