Definition
Chain-of-thought prompting is asking an AI to spell out its reasoning steps before answering, which improves accuracy on multi-step problems.
At a glance
- You tell the AI to think out loud and show its work, not just hand you an answer[3].
- The easy version: add a phrase like “Let us think step by step” to your request. No setup needed[2].
- 2022 research showed sharp gains on math, logic, and commonsense tasks[1].
- Helps most on complex problems; on simple ones it just adds clutter.
How to use it
Zero-shot: add “Let us think step by step” to your request. Few-shot: include one or two worked examples showing the kind of step-by-step reasoning you want, then ask your real question[4]. Both push the model into a more careful mode.
When it is worth it
Use it for problems with several moving parts: multi-step calculations, comparing options, or logic puzzles. Skip it for quick questions. Newer top models increasingly reason this way on their own, so the trick matters less than before, but it stays a cheap thing to try when an answer looks shaky[5].
Bottom line
Ask the AI to show its work: it costs one sentence and pays off most on multi-step problems.
References
- Chain-of-Thought Prompting Elicits Reasoning in Large Language Models — Jason Wei, Xuezhi Wang, Dale Schuurmans, Quoc Le. arXiv arxiv.org
- Large Language Models are Zero-Shot Reasoners — Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, Yusuke Iwasawa. arXiv arxiv.org
- What Is Chain-of-Thought Prompting? Explained. Amazon Web Services aws.amazon.com
- What is chain of thought (CoT) prompting? IBM www.ibm.com
- Prompting Science Report 2: The Decreasing Value of Chain of Thought in Prompting — Lennart Meincke, Ethan R. Mollick, Lilach Mollick, Dan Shapiro. arXiv arxiv.org
Comments
Questions, corrections, and links welcome. Be specific and civil.