What is chain-of-thought prompting?

What is chain-of-thought prompting?

Chain-of-thought prompting is a prompting approach that supplies or requests intermediate solution steps before a final answer. The original research used worked examples to show a language model how to handle tasks involving several connected operations.

The aim is to support problem solving, rather than assuming a direct answer will always capture the required relationships.

A worked example can clarify a method

Suppose a word problem says that three boxes contain four pencils each and two pencils are removed. A useful demonstration identifies the total as 3 × 4 = 12, then subtracts 2 to obtain 10.

Showing a comparable example can help a model apply that structure to a new problem. The chain-of-thought prompting paper reports improvements on selected arithmetic, commonsense, and symbolic reasoning tasks.

Explanations can still contain errors

A model can make an incorrect assumption, miscalculate, or produce a plausible rationale for a wrong answer. The presence of intermediate steps does not establish that the result is correct or that the explanation faithfully describes internal computation.

Check the premises, arithmetic, and final answer independently.

Use explanations for review and learning

For many applications, request the result plus a concise explanation of the method, assumptions, or evidence needed to verify it. There is no need to recover a model's private internal reasoning.

Prompting strategies vary in effectiveness across models and tasks. Test whether an example-based prompt actually improves representative results, and use tools where exact computation or external evidence is required.