PKD, KW Jeter, and others as well. Those two I know used, as a story element (sometimes throwaway, sometimes key) the idea of robots/androids/computers trained on historical figures and other people. In one of Jeter's books, the figures were used in education so you'd ask "Abraham Lincoln" questions about his life (it's been 20 years since I read it, can't remember the specific figures used or if it was just "historical figures").

Obviously this predated LLMs, and they probably called it "programming" and not training, but the idea has been present in scifi for decades.

Star Trek: The Next Generation did this several times with fictional, historical, and (on at least one occasion) contemporary people (to the story).

We could train an AI to reason like Elmer Gantry but we would probably just end up with the prosperity gospels we have now.