Jargonautica
Controlling and Extending Behaviour at Run Time

Chain of Thought

Getting the model to work through a problem step by step before giving its answer, rather than jumping straight to a conclusion. It improves accuracy noticeably on anything involving reasoning, arithmetic or multiple stages, for much the same reason that showing your working helps a person avoid careless mistakes.

Technical view

Intermediate reasoning tokens give the model more computation to allocate to a problem and let it condition on its own partial results. An important caveat is that the stated reasoning is not a reliable account of the underlying process: models can produce plausible reasoning that does not correspond to how the answer was actually reached, so a chain of thought should not be treated as an audit trail.

Part of AI Terminology Index, section: Controlling and Extending Behaviour at Run Time.

How it connects

Chain of Thought links to 13 other terms, and the 12 closest are drawn here. The faint lines are links between those terms. See it in the full graph.

Reasoning Models and Test-Time ComputeReasoning Models an…Chain-of-thought FaithfulnessChain-of-thought Fa…ScratchpadScratchpadTemperature and SamplingTemperature and Sam…Thinking BudgetThinking BudgetProcess Reward ModelProcess Reward ModelInterpretabilityInterpretabilityTokensTokensPrompt EngineeringPrompt EngineeringZero-shot, One-shot and Few-shotZero-shot, One-shot…ReActReActTree of ThoughtsTree of ThoughtsChain of Thought