In more detail
Chain of thought is a model reasoning step by step before answering, instead of jumping straight to a conclusion — the AI version of showing your work in math class. Sometimes you trigger it by asking the model to ‘think step by step’; reasoning models do it automatically.
It matters because the improvement is real. Writing out intermediate steps helps a model catch its own errors mid-stream, exactly the way it helps people. Problems that fail when blurted come out right when worked through — a rare upgrade that costs nothing but extra tokens.
Goes with