Back to full roadmap
topiccore
Chain-of-Thought (CoT)
Writing 'think step by step' boosts math/logic accuracy by 20-60%.
3 hours3 resources1 prereqs
CoT makes the model write out intermediate reasoning — focusing attention on those steps.
3 variants:
- Zero-shot CoT — "Think step by step, then give your answer"
- Few-shot CoT — show example question-with-reasoning pairs
- Auto-CoT — model generates its own examples
New 'reasoning' models (o1, Claude thinking, Gemini 2.5) internalize CoT automatically. But on older/smaller models, it's still critical.