Chain-of-Thought Prompting

What is Chain-of-Thought Prompting?

Chain-of-Thought (CoT) prompting is a technique that encourages language models to generate intermediate reasoning steps before arriving at a final answer. By prompting models to "show their work," CoT significantly improves performance on complex reasoning tasks like arithmetic, commonsense reasoning, and symbolic manipulation.

Key Concepts

Reasoning Process

CoT transforms direct question answering:

Traditional: Question → Answer
CoT: Question → Reasoning Steps → Answer

Example

Question: Roger has 5 tennis balls. He buys 2 more cans of tennis balls. Each can has 3 tennis balls. How many tennis balls does he have now?

Traditional Answer: 11

CoT Answer:

1. Roger starts with 5 balls
2. He buys 2 cans, each with 3 balls: 2 × 3 = 6
3. Total balls: 5 + 6 = 11
4. Therefore, Roger has 11 tennis balls

How CoT Works

Prompting Techniques

1. Few-Shot CoT: Provide examples with reasoning chains
2. Zero-Shot CoT: Use magic phrases like "Let's think step by step"
3. Auto-CoT: Automatically generate reasoning chains

Few-Shot Example

Q: There are 15 trees in the grove. Grove workers will plant trees today. After they are done, there will be 21 trees. How many trees did the workers plant today?
A: Let's think step by step.
1. Start with 15 trees
2. End with 21 trees
3. Trees planted = 21 - 15 = 6
4. Therefore, 6 trees were planted

Q: {new question}
A: Let's think step by step.

Benefits of CoT

Applications

Mathematical Reasoning

• Arithmetic problems
• Algebraic equations
• Word problems
• Mathematical proofs

Commonsense Reasoning

• Everyday problem solving
• Social reasoning
• Physical reasoning
• Temporal reasoning

Symbolic Reasoning

• Logical puzzles
• Algorithm execution
• Code generation
• Formal logic

Complex Decision Making

• Multi-step planning
• Strategic reasoning
• Game playing
• Resource allocation

Implementation

Prompt Design

graph TD
    A[Question] --> B[Reasoning Prompt]
    B --> C[Intermediate Steps]
    C --> D[Final Answer]

    style A fill:#f9f,stroke:#333
    style D fill:#f9f,stroke:#333

Best Practices

• Example Selection: Choose diverse, representative examples
• Step Granularity: Balance detail level in reasoning steps
• Prompt Formatting: Consistent structure across examples
• Error Handling: Include examples with common mistakes

Research and Advancements

Key Papers

1. "Chain-of-Thought Prompting Elicits Reasoning in Large Language Models" (Wei et al., 2022)
   • Introduced CoT prompting
   • Demonstrated 30%+ improvement on reasoning tasks
2. "Large Language Models are Zero-Shot Reasoners" (Kojima et al., 2022)
   • Introduced zero-shot CoT
   • Showed "Let's think step by step" magic phrase
3. "Automatic Chain of Thought Prompting in Large Language Models" (Zhang et al., 2022)
   • Introduced Auto-CoT
   • Automated reasoning chain generation

Emerging Research

• Multimodal CoT: Combining text with visual reasoning
• Tree of Thoughts: Exploring multiple reasoning paths
• Self-Consistency: Sampling multiple chains and voting
• Faithful CoT: Ensuring reasoning matches final answer
• CoT Fine-tuning: Training models for better reasoning

External Resources

• Original CoT Paper
• Zero-Shot CoT Paper
• Auto-CoT Paper
• CoT Guide
• CoT Examples
• Reasoning Benchmarks