Chain of Thought

CoT

Part of: Advanced Prompting Patterns

Ask a model "What is 17 times 24?" and it often returns a wrong number. Add four words, "Let's think step by step", and the same model works it out and lands on 408. The model itself did not change. What changed is what it wrote first. What it is Chain of thought (CoT) is a prompting pattern where you ask the model to write out its intermediate reasoning before giving a final answer. Instead of jumping straight to a conclusion, the model produces a visible trail of steps, and the answer falls out at the end of that trail. This works because of how models generate text. A model can only "think" by writing tokens; there is no hidden scratchpad. When you force the answer on the first token, the model has done no computation toward it. When you let it write steps first, each step becomes context the next token can build on. The reasoning is not decoration; it is the computation. How it works There are two flavors: - Zero-shot CoT. You append a trigger phrase like "Let's think step by step." No examples needed. The phrase nudges the model into a step-by-step continuation because that pattern is common in its training data. - Few-shot CoT. You show one or two worked examples that include

Challenge: Audit the Scratchpad