What is Chain-of-Thought?
TL;DR
A prompting technique that improves answer accuracy by guiding AI through step-by-step reasoning.
Chain-of-Thought: Definition & Explanation
Chain-of-Thought (CoT) is a prompt engineering technique that dramatically improves accuracy on complex problems by having the LLM explicitly work through a step-by-step reasoning process. Published by a Google research team in 2022, it gained widespread attention. For example, when solving a math problem, prompting the model to 'first calculate X, then determine Y, and finally compute Z' produces more accurate results than asking for a direct answer. OpenAI's o1/o3 models and DeepSeek-R1 are designed as reasoning-specialized models that leverage CoT internally. Even the simple instruction 'Think step by step' has been shown to be effective, making it one of the most widely used basic prompt engineering techniques. Extensions of CoT, including Tree-of-Thought and Graph-of-Thought, are also being actively researched.