Empathic Inference — LessWrong