ChatGPT is done with supervised and reinforced learning - RLHF
It basically means that people have voted millions of answers for whether chatGPT responded correctly to teach it to behave properly and not say dangerous and abusive responses.