Hmmm.... so it now appears that AI (& LLMs) 'memorise' much of their training data than had been previously thought... to the extent that they can largely reproduce verbatim novels that were used in there training process.
As you can imagine this further complicates the claims about copyright infringement & the fair use defence that have been deployed by Big Tech.
Its more evidence of the theft at the heart of the (so-called) AI 'revolution'!
h/t FT
Follow
It is funny that if you give the starting paragraph of a novel and ask it to continue the text, it does it if it is public domain, but it doesn't if it isn't, giving an explanation instead.
That means it has a copy of the text inside; otherwise, how can it determine it is copyrighted?
@jgg
exactly