The Multi-Tower Study Strategy — LessWrong