A Very Concrete Model of Learning From Regrets — LessWrong