chain-of-thought prompting

Chain-of-thought prompting asks the model to write out its reasoning before giving the final answer, the way a student shows their working instead of blurting a number. For multi-step problems, like word problems, logic puzzles, or anything that needs several linked deductions, this single move can lift accuracy dramatically. The phrase let's think step by step, added to a prompt, became famous precisely because it nudges the model into this mode.

Why does writing the steps help? Each token the model generates is fed back in as it produces the next one, so the visible reasoning becomes extra context the model can build on. A hard problem gets unpacked into a sequence of small, easier continuations rather than one giant leap to the answer. In effect, the model is given room to compute on the page instead of all at once in its head.

A caution: the written chain reads like an explanation, but it is not a guaranteed window into how the model truly arrived at its answer. Models sometimes produce convincing reasoning that does not actually drive the conclusion, and a flawed step can still end at a correct answer or vice versa. Treat the chain as a working aid, not as proof.