I would love to hear your opinion on something I keep thinking about. There’s the whole idea that these LLMs are training on “available” data all over the internet, and then anyone can use the LLM and create something that could resemble the work of someone else. Then there’s the people calling it theft (in my opinion wrong from any possible angle of consideration) and those calling it fair use (I kinda lean more on this side). But then we have the side of compensation for authors and such, which would be great if some form for it would be found. Any one person can learn about the style of an author and imitate it without really copying the same piece of art. That person cannot be sued for stealing “style”, and it feels like the LLM is basically in the same area of creating content. And authors have never been compensated for someone imitating them.
So… What would make the case of LLMs different? What are good points against it that don’t end up falling into the “stealing content” discussion? How to guarantee authors are compensated for their works? How can we guarantee that a company doesn’t order a book (or a reading with your voice in the case of voice actors, or pictures and drawings, …) and then reproduces the same content without you not having to pay you? How can we differentiate between a synthetic voice trained with thousand of voices but not the voice of person A but creates a voice similar to that of A against the case of a company “stealing” the voice of A directly? I feel there’s a lot of nuances here and don’t know what or how to cover all of it easily and most discussion I read are just “steal vs fair use” only.
Can this only end properly with a full reform of copyright? It’s not like authors are nowadays very well protected either. Publishers basically take their creation to be used and abused without the author having any say in it (like in the case of spot if unpublished a artists relationship and payment agreements).