The "fun" part about cases like this is that we don't really know what the contours of the law are as applied to training data like this. Illegally downloading a book is an independent act of infringement (to my recollection at least). So I'm not sure that it matters if you eventually trained an LLM with it vs read for your own enjoyment. But we will see! Fair use is a possibility here but we need a court to apply the test and that will probably go up to SCOTUS eventually.