思维链 (CoT)
定义
思维链（CoT）提示要求模型输出中间推理步骤 before the final answer. This often improves accuracy on math, logic, and multi-step tasks.
它是 one of the simplest 推理 patterns: 无工具或搜索，仅使用提示. 当…时使用 the task benefits from explicit steps (例如 arithmetic, deduction) and you want to avoid fine-tuning. For exploring multiple solution paths, see tree of thoughts; for tool-using agents, see ReAct.
工作原理
你给模型一个问题（或任务），让它逐步推理。模型产生步骤1、步骤2、… (intermediate 推理) and then the answer. Zero-shot CoT: add “Let’s think 逐步” (or similar) to the prompt. Few-shot CoT: include example (question, steps, answer) triples so the model mimics the format. The model generates the sequence in one pass; you can optionally parse the steps and verify or score them. Quality depends on prompt engineering and model capability.
应用场景
Chain-of-thought 在任务受益于明确的中间步骤时最有用 (数学、逻辑、代码).
- Math and arithmetic where intermediate steps improve accuracy
- Logic puzzles and multi-step deduction
- Code or 设计 推理 where showing steps aids debugging
外部文档
- Chain-of-Thought Prompting (Wei et al.) — CoT paper
- OpenAI – Prompt engineering — Includes 推理 and step-by-step guidance