Humans also respond differently when prompted in different ways. For example, politeness often begets politeness. I would expect that to be reflected in training data.
I think that's what is happening. It's simulating a conversation, after all. A bit like code switching.
that seems like something you wouldn't want from your tools. humans have that and that's fine, people are people and have emotions but I don't want my power-drill asking me why I only call when I need something.
If I, a moron, hire a PhD to crack a tough problem for me, I don't need to go back and forth prompting him at a PhD level. I can set him loose on my problem and he'll come back to me with a solution.