Anybody tried to train a LLM, or seen a LLM that instead of providing "authoritative" answers to queries, would lead a user to a discovery through a series of questions and hints? Would such thing even be possible?
It's definitely possible. The founder of Khan Academy demonstrated something like this with GPT-4o: <a href="https://www.youtube.com/watch?v=IvXZCocyU_M" rel="nofollow">https://www.youtube.com/watch?v=IvXZCocyU_M</a>