Its successor, ChatGPT, is a model especially designed to excel at conversation, but this required supervised learning—humans were asked to manually label thousands of input phrases with their expected answers and also rank different answers by quality, and this data was fed into the model to refine it.
