Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
Jason Wei Xuezhi Wang Dale Schuurmans Maarten Bosma Brian Ichter Fei Xia Ed H. Chi Quoc V. Le Denny Zhou Google Research, Brain Team
Abstract
We explore how generating a chain of thought—a series of intermediate reasoning steps—significantly improves the ability of large language models to perform complex reasoning. In particular, we show how such reasoning abilities emerge naturally in sufficiently large language models via a simple method called chain-of-thought prompting, where a few chain of thought demonstrations are provided as exemplars in prompting.
中文速览
让大型语言模型（LLM）做复杂推理一直是个难题，单纯堆大模型规模并不够用。研究者提出了一种叫"思维链提示"（chain-of-thought prompting）的简单方法：在给模型的少样本示例中，不只给出问题和答案，而是附上一步步的中间推理过程，引导模型在回答时也生成类似的推理链条。实验在算术、常识和符号推理等多类任务上测试了包括PaLM、GPT-3、LaMDA在内的多个大模型，发现这一方法能大幅提升推理准确率——例如仅用8个示例提示PaLM 540B，就在数学应用题基准GSM8K上超越了经过专门微调的GPT-3。值得注意的是，这一能力只在参数量约达1000亿以上的模型中才会涌现，小模型并不受益；而对于人类来说，这项工作说明只需在提示词里加入自然语言推理步骤，就能无需任何额外训练地激活大模型潜藏的推理能力，既省成本又灵活通用。
原文 arXiv:2201.11903;中英对照 + 大白话阅读 https://aha.fim.ai/paper/2201.11903v6