I do kind of wonder if there’s a way to reverse engineer this process to determine what sort of information and biases a person has to have to act the way they do, by doing the same process on an LLM that acts like they do