I will say that large language models tend to become “forgetful” the longer a story becomes. Especially if you have it write multiple versions of the same chapter. It seems to forget which one is the best one.