What's the feedback loop here for ChatGPT?

For it to get better, it needs to know outcomes of its diagnosis.

Are people just typing back to ChatGPT saying "you're wrong / you're right"?