chain-of-thought

Definition

Having a model write out intermediate steps before its final answer. Prompting for it improves accuracy on problems that need multi-step work, and because it requires no weight updates it spread quickly; later models were trained to do it by default. One important caveat: the written trace is itself generated text and need not faithfully describe the computation that produced the answer.

Explainers using this term

• Chain-of-Thought from Scratch — What Changes When You Make a Model Show Its WorkChain-of-Thought Prompting Elicits Reasoning in Large Language Models
• Test-Time Scaling — How Models Get Better by Thinking LongerChain-of-Thought Prompting Elicits Reasoning in Large Language Models
• The Science of Prompt Engineering — What Is Proven and What Is FolkloreChain-of-Thought Prompting Elicits Reasoning in Large Language Models