Skip to content
Navigation
Dashboard
🎬Video•35 min

Chain-of-Thought Reasoning

Learn how to make LLMs show their reasoning.

Chain-of-Thought (CoT) Prompting

What is CoT?

Ask the model to explain its reasoning step-by-step before giving the final answer.

Zero-shot CoT

Simply add: "Let's think step by step."

Surprisingly effective for reasoning tasks!

Few-shot CoT

Provide examples with reasoning:

"Q: If John has 3 apples and gives 1 away, how many left? A: John starts with 3 apples. He gives 1 away. 3 - 1 = 2. John has 2 apples.

Q: [new problem] A: Let me think step by step..."

When CoT Helps

  • Math problems
  • Logic puzzles
  • Multi-step reasoning
  • Complex decisions
  • Variations

    Self-Consistency: Generate multiple CoT paths, vote on answer Tree-of-Thought: Explore multiple reasoning branches Plan-and-Solve: Plan first, then execute

    Reasoning Models

    OpenAI o1, DeepSeek R1 have built-in CoT:

  • Extended thinking time
  • Internal chain-of-thought
  • Better on complex problems
  • 🎯 Key Takeaways

    • ✓CoT makes models show their reasoning
    • ✓"Let's think step by step" often helps
    • ✓Self-consistency improves accuracy
    • ✓Reasoning models have built-in CoT