Back to papers
    chain-of-thoughtreasoningpromptingagents

    Chain-of-Thought Prompting Elicits Reasoning in Large Language Models

    Fei XiaJason WeiDale SchuurmansMaarten BosmaBrian IchterEd H. ChiDenny ZhouQuoc V. LeXuezhi Wang
    📅 2022🏛 Advances in Neural Information Processing Systems (NeurIPS) 2022, arXiv:2201.11903
    Problem

    This paper demonstrates that by prompting LLMs to generate intermediate reasoning steps (chain-of-thought), their performance on multi-step reasoning tasks dramatically improves.

    Method

    The authors use chain-of-thought prompting, where the model is encouraged to produce a sequence of reasoning steps before giving the final answer.

    Finding

    This method significantly improved accuracy on math, logical reasoning, and complex QA tasks, even for smaller models.

    Limitations

    Chain-of-thought can suffer from hallucinations and may require careful prompt design; it also increases computation cost.

    Practical application

    This technique is widely used in autonomous agents for step-by-step planning and reasoning. In the software agent project, chain-of-thought helps the agent logically design the sequence of actions needed to execute a Photoshop command.

    📇 Summary flashcard — 13 analytical fields for this paper

    خلاصه

    Chain-of-thought prompting shows that encouraging LLMs to produce intermediate reasoning steps greatly improves performance on complex multi-step problems.

    نمای سریع

    Enhancing reasoning by chaining thoughts.

    یافته‌های کلیدی

    This method significantly improved accuracy on math, logical reasoning, and complex QA tasks, even for smaller models.

    هدف

    To improve the reasoning capabilities of large language models on multi-step tasks.

    روش

    The authors use chain-of-thought prompting, where the model is encouraged to produce a sequence of reasoning steps before giving the final answer.

    نتایج

    Higher accuracy on mathematics and logical reasoning benchmarks.

    نتیجه‌گیری

    Chain-of-thought is a simple and effective technique to unlock latent reasoning abilities in LLMs.

    مفاهیم کلیدی

    chain-of-thought، reasoning، prompting، agents

    مطالعه‌ی بیشتر

    https://arxiv.org/abs/2201.11903

    تحلیل

    This paper has had a significant impact on prompt engineering and the design of autonomous agents.

    محدودیت‌ها

    Chain-of-thought can suffer from hallucinations and may require careful prompt design; it also increases computation cost.

    کارهای آینده

    The authors suggested combining this method with external tools and feedback loops.

    کاربرد عملی

    This technique is widely used in autonomous agents for step-by-step planning and reasoning. In the software agent project, chain-of-thought helps the agent logically design the sequence of actions needed to execute a Photoshop command.

    References (this paper draws on) (0)

    No references recorded.

    Cited by (0) ▶

    No paper cites this one yet.

    Prerequisite learning path for this paper