Advanced Reasoning Strategies / Multi-Path Reasoning

Analogical Reasoning

Advanced [4/5]
Reasoning by analogy Analogical prompting Transfer reasoning

Definition

Analogical reasoning prompts the model to recall or generate similar problems and their solutions, then apply that reasoning pattern to the current problem. This mimics human problem-solving where we draw on past experiences to tackle new challenges.

This technique leverages the model's training data to find relevant analogies that guide reasoning on novel problems.

Key Concepts

  • Source problem: A known problem with established solution
  • Target problem: The new problem to be solved
  • Structural mapping: Identifying corresponding elements between problems
  • Solution transfer: Applying the source solution pattern to target

Examples

Self-Generated Analogy
Recall Similar Problems
Problem: "A farmer has chickens and rabbits. He counts 20 heads and 56 legs. How many of each animal does he have?" Prompt: "Before solving, recall a similar problem you know and explain how you solved it. Then apply that approach here." Response: Similar Problem: "A parking lot has cars and motorcycles. There are 10 vehicles with 32 wheels total." Solution Pattern: - Let x = cars (4 wheels), y = motorcycles (2 wheels) - Equations: x + y = 10, 4x + 2y = 32 - Solve: x = 6 cars, y = 4 motorcycles Applying to Current Problem: - Let c = chickens (2 legs), r = rabbits (4 legs) - Equations: c + r = 20, 2c + 4r = 56 - Solve: c = 12 chickens, r = 8 rabbits
Provided Analogy
Explicit Analogy Prompt
Consider this analogy: Debugging code is like being a detective: - Bug = Crime - Error message = Witness testimony - Stack trace = Evidence trail - Root cause = Criminal - Fix = Arrest and conviction Using this analogy, explain how to debug this error: "TypeError: Cannot read property 'map' of undefined" Response: "As a code detective, let's investigate: 1. Witness testimony (error): Something is undefined when we try to use .map() 2. Evidence trail: The variable we're calling .map() on is undefined 3. Investigation: Find where this variable should be defined 4. Root cause: Data not loaded, or typo in variable name 5. Arrest (fix): Add null check or ensure data is loaded first"

Interactive Exercise

✎
Create an Analogy

Create an analogy to help explain this problem:

Problem: "How does a neural network learn?" Create an analogy to a familiar concept.

Pro Tips
  • Let the model generate its own analogies for better fit
  • Ensure structural similarity, not just surface similarity
  • Verify the analogy doesn't mislead on key differences
  • Combine with chain-of-thought for detailed reasoning

Related Terms

Chain of Thought Few-Shot Prompting Self-Consistency Example Selection