Can AI models infer personal preferences from how people talk?

A user argues that AI models cannot infer a person's preferences even from unlimited, perfect context. They wonder why models can't pick up most of those preferences from the way people talk, and say context size and test-time training aren't the problem.