What is Chain-of-Thought?
A prompting technique where the model is asked to reason through a problem step by step before producing a final answer. By making intermediate reasoning explicit, chain-of-thought significantly improves accuracy on tasks requiring logic, math, multi-step analysis, or nuanced judgment. First demonstrated by Wei et al. (2022) at Google.
My perspective
In practice
A reliable way to get better AI output on complex tasks. Simply adding "think step by step" or "reason through this carefully" to a prompt can noticeably improve accuracy. More advanced versions include structured reasoning (numbered steps), self-critique ("now check your reasoning for errors"), and tree-of-thought (exploring multiple reasoning paths). It works particularly well when the model needs to weigh trade-offs or handle ambiguity.