DataTerminology.comNEW
A prompting technique that elicits step-by-step intermediate reasoning, improving accuracy on tasks requiring multi-step inference.
DESCRIPTION
Chain-of-Thought encourages models to generate intermediate steps before final answers. It improves performance on arithmetic, commonsense, and symbolic reasoning tasks. The method can be applied with few-shot exemplars showing stepwise reasoning. Self-consistency sampling further boosts reliability by aggregating multiple chains. Techniques like Tree of Thoughts extend this idea to branching exploration. Care must be taken to avoid revealing sensitive reasoning steps in risky domains. CoT is a core tool in prompt engineering and evaluation.
ALTERNATIVE DEFINITIONS
arXiv
Chain-of-thought prompting elicits step-by-step reasoning by generating intermediate reasoning steps.
Defines CoT as eliciting stepwise reasoning via intermediate steps.
Google AI Blog
Chain-of-thought prompting encourages models to produce intermediate reasoning steps that lead to the final answer.
Emphasizes encouragement of intermediate steps for better answers.
RELATED TERMS
Prompt Engineering Few-Shot Learning In-Context Learning Self-Consistency Reasoning Deliberate Sampling Scratchpad Evaluation
COMPETING TERMS
Tree-of-Thought Direct Answer Prompting Program-of-Thought
RELATIONS
- Broader: Chain-of-Thought Prompting Prompt Engineering
- Referenced by: AIME
KNOWLEDGE AREAS
AI Data Science Testing & Validation
TARGET AUDIENCE
AI Engineer Data Scientist MLOps Engineer AI Product Manager
EXTERNAL SOURCES
- Wikipedia - Prompt engineering
- arXiv - CoT prompting paper
- Google - Accessible CoT overview
- arXiv - Self-consistency method
TOP QUESTIONS
When should I use Chain-of-Thought prompting?
Use it for tasks requiring multi-step reasoning, arithmetic, or logical inference where stepwise explanations improve accuracy.
Category: UseCases
How do I craft CoT exemplars?
Provide clear, concise step-by-step examples demonstrating reasoning patterns that lead to correct answers.
Category: How-To
What is self-consistency with CoT?
It samples multiple reasoning chains and aggregates answers, improving reliability by reducing variance.
Category: Concept
What are risks of Chain-of-Thought prompting?
Exposing sensitive or biased intermediate steps can be risky; consider hidden CoT or answer-only output for sensitive domains.
Category: Risk
How does Chain-of-Thought compare to Tree of Thoughts?
CoT follows a single path; Tree of Thoughts explores branches, trading compute for potentially better solutions.
Category: Comparison
Does CoT always help?
Not always; on simple tasks it may add latency without gains. Evaluate per task.
Category: Opinion
How to evaluate CoT effectiveness?
Run A/B tests with and without CoT, measure accuracy, latency, and consistency under realistic prompts.
Category: Evaluation
What is hidden Chain-of-Thought?
The model reasons step-by-step internally but only the final answer is revealed to users, reducing leakage risks.
Category: Safety
Where to start learning CoT?
Read the original paper and Google AI blog, then practice with few-shot examples on reasoning tasks.
Category: GettingStarted