Chain of thought (CoT), breaking a problem down into a series of intermediate reasoning steps, has significantly improved the ability of LLMs to perform complex reasoning.