I have also wondered if OpenAI are going to train a private model with all the ChatGPT history, and then use that to train a public model.