AI Glossary
Chain of thought
chain of thought, CoT, step-by-step reasoning
Chain of thought is a technique in which a model works through a solution step by step before giving an answer. It helps with multi-step tasks such as arithmetic or logic.
- The model lays out successive steps before the final answer.
- It improves results on multi-step tasks: mathematics, logic, analysis.
- It makes the answer longer and more expensive, and the steps alone don't guarantee correctness.
Chain of thought addresses the fact that language models often go wrong on multi-step tasks when they try to give the answer right away. By asking the model to lay out the successive steps before answering, you break the problem into smaller parts, which usually raises accuracy in mathematics, logic and analysis.
The technique is triggered through a suitable prompt, sometimes backed by examples in the few-shot style. Keep in mind that written-out reasoning makes the answer longer and raises the cost, and that the record of the steps is not proof of correctness, because the model can justify a wrong result.
Related terms