Facebook and Stanford researchers design a chatbot that learns from its mistakes

Original paper: https://arxiv.org/abs/1901.05415