bradley13 parent
Good. Reading books is legal. If I own a book and feed it to a program I wrote (and I have done exactly that), it is also legal. There is zero reason this should be any different with an AI.
I've co-authored a book that a lot of the models seem to know about. The models consistently get the names of the authors incorrect and quote the material with errors. If the canonical representation of our work is now embedded within AI models, don't we deserve to have it quoted and represented correctly and fairly? If you asked a human who had read the book, I think there is a fair chance they would likely give you the reference to the source material.
I do concede that the book does contain a distillation of material that is also available from other sources, but it also contained a lot of personal experience. That aspect does seem to be lost in this new representation.
I am not saying that letting AI models read the material is wrong, but the hubris in the way models answer questions is annoying.
If you charge me to use your program and it spits out unedited, copyrighted material then it should be illegal. I don't know the details of this case, but that's what's going on in the New York Times case. It's not always so cut and dry.
Which is amusing because NYTimes has fought in court a few times in favour of technology progress over copyright. Including recently when they got sued over collected a bunch of freelance writing into a database without consent.
https://harvardlawreview.org/blog/2024/04/nyt-v-openai-the-t...
I doubt the exact replica stuff will stand, as technically it was only achievable via advanced prompt engineering (hacking), not simply asking for a replica. So their 2 other arguments boils down to scraping a news database = infringement and LLM output = derivative works.