Modularity, signaling, and belief in belief — LessWrong