文章目录AbstractIntroductionChain-of-thoughtpromptingArithmeticReasoningexperimentalsetupresultablationstudyCommonsenseReasoningexperimentalsetupResultSymbolicReasoningexperimentalsetupresultLimitationAppendixAbstract本文通过ChainofThoughts(CoT，即推理中间过程)，提升大型语言模型(LLM)推理能力。在三个LLM上证明了CoT能够显著提升算术、常识、符号推理能力。Intr