Writing-wise, LLMs are only a threat if the customer doesn't actually read/watch/listen to whatever it is they bought. I've read a lot of LLM stuff at this point and not only is the tech currently incapable of producing something with entertainment value that couldn't be reproduced in a few minutes with Mad Libs, it can't be developed to a point which could allow it to write something that is engaging or entertaining. People tend to vastly overestimate this technology and vastly underestimate the true depth of complexity any human mind has whether that mind is aware of its own complexity or not. The threat is from the creativity of some of the best marketers in the world presenting this tech as something it can never possibly be to extremely wealthy investors with no understanding of art, which has been very effective.

The tech could improve in that the information it provides is less likely to be incorrect. Creating a plot that a human person relates to and is engaged by is many orders of magnitude more difficult to do. A "neural net" is not nearly complex enough to replicate a concious and subconcious mind molded by billions of years of natural selection, using hardware we haven't even scratched the surface of understanding other than being aware of trillions of calculations at any given moment related to certain aspects of cognition, acting on and being acted on by the social and physical aspects of our reality. It can't understand. It's only capable of processing information and reproducing it in language which is similar to how people use language. The best is could do is summarize something that has the potential to be entertaining if an actual writer does something with it.