Contents Chain-of-Thought PromptingReferences Chain-of-Thought Prompting 作者发现,通过 chain-of-thought prompting 给 LLM 提供 few-shot chain of thought demonstrations 就可以使得 LLM 在推理时生成推理的中间步骤 chain of thought (for sufficiently large LLMs),从而显著提升 LLM 在复杂推理任务上的能力,此外也提升了 LLM 的可解释性 References Wei, Jason, et al. “Chain-of-thought prompting elicits reasoning in large language models.” Advances in neural information processing systems 35 (2022): 24824-24837.