chain of thought
/ˌtʃeɪn əv ˈθɔːt/
Prompting technique that encourages step-by-step reasoning
chain of thought in a sentence
“Chain of thought prompting improved the model's problem-solving accuracy.”
Origin of chain of thought
From Latin catena (chain) + Old English þōht (thought), from þencan (to think)
What does chain of thought really mean?
Chain-of-thought means the intermediate reasoning steps generated while a model solves a task. Classic prompting elicited those steps with worked examples or phrases such as "think step by step"; modern reasoning models may generate them internally and expose only an answer or summary. A visible explanation can be useful, but it is not guaranteed to be a faithful record of the computation that produced the answer.
The story behind chain of thought
Named in a 2022 Google Research paper (Wei et al.), which showed that simply demonstrating worked reasoning in the prompt unlocked abilities larger models already had. The now-famous shortcut "let's think step by step" arrived the same year, and the idea has since been built into models themselves as extended or hidden reasoning.
How to use chain of thought
Use it as a noun or modifier: "chain-of-thought prompting," "the model's chain of thought." Hyphenate when it modifies. Do not use it as a synonym for every explanation, and do not assume that requesting raw reasoning improves every model. For verification, ask for a concise rationale, cited evidence, or checkable calculations and evaluate the result itself.
Related Words
retrieval augmented generation
Combining search results with generative AI for grounded responses
multimodal
AI capable of processing multiple types of input like text and images
zero-shot learning
Performing tasks without any task-specific training examples
few-shot learning
Learning from just a handful of examples
in-context learning
Learning patterns from examples provided in the prompt
alignment
Ensuring AI behavior matches human values and intentions