I mean, that’s a slightly different thing than writers not using LLMs and there are reasons like the environmental impact and putting people out of jobs for not using LLMs, but the biggest one is that it will only ever produce derivative retreads.
Some books have always been derivative slop of course - they were called ‘pulp’ for a reason - and really the only reason for not using LLMs to create more is that even hacks deserve bread on the table (speaking as a hack).
LLMs are only ever going to come up with the most likely version of the next word given its instructions, and that probability is based on what’s out there already. So, always derivative shit. That’s why not to use AI.