A prompting technique in which a large language model is encouraged to produce intermediate reasoning steps before its final answer, typically improving performance on multi-step arithmetic, logical, and commonsense tasks.
In practiceChain-of-thought prompting can be elicited through worked examples (few-shot CoT), through a simple instruction such as let us think step by step (zero-shot CoT), or through a separate scratchpad that is hidden from the end user. Practitioners must remember that the intermediate text is a generation artefact, not a faithful trace of the model's internal computation, so explanations should not be presented to users or regulators as causal evidence of how a decision was reached.
A tax-classification assistant prompted with three worked examples that each break down the rule application into sub-steps produces materially better answers on a held-out edge-case set than the same model prompted to answer directly.
This definition is maintained by Moweb partners and used in live client engagements. For how Chain-of-thought applies to your estate, or to challenge a working definition, speak to a partner.