Finetuning Activates Verbatim Recall of Copyrighted Books in LLMs
arxiv.orgThere have been other papers that demonstrated LLM ability to reproduce copyrighted text as well.
Will it actually impact the legal landscape, though? It seems like there's enough money being thrown at this that it'll be deemed legal no matter what.
The important thing will be to ensure then that this is legal for everyone. Nit just big LLMs
It would be a feature not a bug if it could preserve attributions and "who said what" too.
This is the kind of thing that looks simple until you're three layers deep in edge cases.
You mean like everything else? There is no simplicity. Just a local minima in the waterbed.