It’s not possible as things currently stand. It’s worrying how often people don’t understand this. AI proponents hate the “they just predict the next token” approach, but it sure helps a lot to understand what these things will actually do for a particular input.

I think the only way I could see it happening is if you were to build an entire reversal layer with like LangExtract, tried to determine the user's intent from the question and then used that as middleware for how you let the LLM proceed based on its intent... I don't know, it seems really hard.