Is it possible to train a text generative model on someones message history and then have a pretty good estimate on how they would respond to something written to them?