Chaining Methods
Analogy: Solving a problem step-by-step
Chaining techniques direct AI through systematic procedures, similar to how people solve problems step by step. Examples include Zero-shot and Few-shot CoT.
Zero-shot Chain-of-Thought
Zero-shot CoT prompts AI to show remarkable reasoning skills without prior examples, arriving at logical solutions.
Few-shot Chain-of-Thought
Few-shot prompting efficiently directs AI with limited input-output examples, enabling pattern discovery and flexible responses.
Decomposition-Based Methods
Analogy: Breaking a complex problem into smaller sub-problems
These methods simplify complex issues into smaller components, allowing thorough analysis. Examples are Least-to-Most Prompting and Question Decomposition.
Least-to-Most Prompting
This method addresses easy-to-hard generalization, enabling models to solve complex problems by dividing them into simpler subproblems.
Question Decomposition
Divides complex questions into manageable subquestions, improving reasoning precision and model transparency.
Path Aggregation Methods
Analogy: Generating multiple options to solve a problem and choosing the best one
These techniques leverage AI’s capacity to consider multiple options and find the best solution. Examples are Graph of Thoughts and Tree of Thoughts.
Graph of Thoughts (GoT)
Enhances prompting capabilities by modeling data as a graph, combining different ideas for synergistic results.
Tree of Thoughts (ToT)
Preserves a hierarchical tree of ideas for forward-thinking planning and comprehensive problem-solving.
Reasoning-Based Methods
Analogy: Reasoning and verifying sub-tasks for accuracy
These approaches emphasize producing accurate solutions and confirming their correctness. Examples include CoVe and Self-Consistency.
Chain of Verification (CoVe)
Improves AI accuracy by evaluating responses through a structured series of inquiries and self-verification.
Self-consistency
Enhances CoT prompting by ensuring a more reliable response through multiple chains of thought.
External Knowledge Methods
Analogy: Using external tools and knowledge to complete a task
Similar to how humans use external resources, these methods provide AI with additional data or resources. Examples include Consortium of Knowledge (CoK) and Automatic Reasoning and Tool-use (ART).
Consortium of Knowledge (CoK)
Supports reasoning by building structured evidence triples from a knowledge base, ensuring factual truth and reliability.
Automatic Reasoning and Tool-use (ART)
Utilizes external tools and reasoning stages to solve complex tasks, pausing generation to incorporate outputs from external tools.
