So, interacting with an AI trying to get domain specific answers where you already have domain knowledge, basically teaches you how awesome human minds are at interpreting context and intent and providing an answer that contributes to the perceived goal.
It also teaches you how woefully sloppy and imprecise your own thoughts and verbalisations are.
So we're basically back at Postel's law: be strict in what you send, liberal in what you receive.