Improving LLM Response Quality in Mathematical and Logical Reasoning Tasks with Chain-of-Thought Prompting

Chain-of-Thought (CoT) is a prompting technique that enhances reasoning in large language models. Its effectiveness has been more evident in logical, commonsense, and mathematical reasoning tasks. Reported to be more effective with older and smaller LLMs, its ability to improve performance in models like GPT-3 is evident in the works included in this comparison. It is however noteworthy that CoT is an active research direction with yet only few research explorations. It is therefore a direction to further explore towards response improvement in generative AI models.

思维链（Chain-of-Thought，CoT）是一种可提升大语言模型（Large Language Model，LLM）推理能力的提示工程技术。其有效性在逻辑推理、常识推理与数学推理任务中表现得更为显著。有研究表明，该技术在体量更小、问世更早的大语言模型上效果更优，其在GPT-3等模型上的性能提升效果，在本次对比所纳入的研究成果中已得到验证。但值得注意的是，思维链仍是一个处于活跃探索中的研究方向，目前相关研究仍较为有限。因此，针对生成式人工智能（Generative AI）模型的响应效果优化，该方向仍有待进一步深入探索。