Pro-AI types like myself like to distinguish learning/assimilation from verbatim reproduction, and I think we’re right that LLMs mostly do the former.
But I wonder if we feel that learning itself poses no threat because humans have never been able to assimilate several million books. +
But I wonder if we feel that learning itself poses no threat because humans have never been able to assimilate several million books. +
Comments
I would still be able to say “well, my use is nonprofit,” but it’s a weaker +
https://copyright.columbia.edu/basics/fair-use.html
I don’t think they add up to a simple escape clause for nonprofits. Eg if Meta establishes that there is a market for “training AI on a corpus” it tends to support +
Les selfishly, I also think it’s a bad precedent to establish for the future of open inquiry that all intellectual content is property and any form of learning from a book requires a payment to the creator.
but would also keep more diffuse forms of learning free & fair use
But technology & law don’t care at all about my moral feelings.
Instead of trying to compensate people whenever they’re echoed (i.e, always), we might focus on the (rarer, clearer) moments of actual innovation.
Neat thought experiment! Lots to think about.
But knowledge also counts. The sheer *breadth* of LLMs is a place where they are already superhuman and may be opening new perspectives.