Brian Keene, Abdi Nazemian and Stewart O#39;Nan said their works were part of a dataset of about 196,640 books that helped train NeMo to simulate ordinary written language, before being taken down in October “due to reported copyright infringement.”