Definition
Chain-of-thought (CoT) encourages intermediate reasoning traces — "let's think step by step" — before the final answer. It helps on math, logic, and multi-hop questions.
Some models are trained to produce reasoning; others respond to CoT-style prompts. Hidden or summarized reasoning is an active product design choice.
In simple terms
Showing your work on a math exam — the path reduces careless jumps to a wrong final number.
Where you see it
- Math and coding benchmarks improve with CoT.
- Agent planners break goals into steps.
How it works
1.Ask for steps
Prompt the model to reason explicitly.
2.Generate trace
Model writes intermediate conclusions.
3.Conclude
Final answer follows from the trace.
Why it matters
- CoT is a simple lever for harder reasoning without changing model weights.
Often confused
A long reasoning trace guarantees correctness.
Models can rationalize wrong answers fluently — verify critical results.