Chain-of-Thought Prompting: Step-by-Step Reasoning in Large Language Models

Chain-of-Thought (CoT) prompting is a technique that dramatically improves the reasoning capabilities of large language models by asking them to show their work step by step. Instead of jumping directly to an answer, the model is encouraged to generate intermediate reasoning steps, mimicking how humans solve complex problems.

This video demonstrates CoT using a simple "apples" word problem where standard prompting fails but chain-of-thought succeeds. The technique is then validated on real benchmarks: on the GSM8K math benchmark, PaLM 540B jumps from about 18% accuracy with standard prompting to approximately 57% with chain-of-thought—a 3× improvement that illustrates an emergent ability unlocked by model scale.

You can apply CoT today in any mainstream LLM: add a short instruction like "Let's think step by step" or "Show your work," and when possible include a compact example in the question → reasoning → answer format. CoT not only improves performance on multi-step tasks (math word problems, planning, rule following) but also makes the model's logic auditable so you can verify each step—not just the final answer.

Chain-of-Thought Prompting Elicits Reasoning in Large Language Models

Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc Le, Denny Zhou (2022)