“Harvey AI has developed a pipeline for creating synthetic datasets for every legal use case, which it uses for expensive post-training of its models.”