How to learn from conversations — LessWrong