A model writes one token at a time, and every token it writes becomes part of the text it reads next. That means a model that jumps straight to an answer has no room to work, while one that writes out its steps first gives itself something to build on. Asking for step-by-step reasoning, often called chain-of-thought prompting, can noticeably improve results on multi-step problems like math, logic and policy decisions.
The lessons show the technique at work, from the simple trigger phrase "let's think step by step" to worked examples that demonstrate the reasoning you want. They also cover its limits: it adds little on simple questions, costs extra tokens, and makes errors visible rather than making them disappear. Many newer models already reason on their own, so it is worth testing whether the extra instruction actually helps.
Watch the lessons in order, then continue on to Structured Outputs.
A short walkthrough of the simplest version: add a trigger phrase to your question and the model reasons through the problem before answering, with no worked examples needed. Uses a fruit stand problem to show how skipping a step leads to a wrong answer.
Video 2minCodeEmporium explains how language models can reason with chain-of-thought prompting, why it is useful, how it connects to few-shot examples, and what performance benefits it brings.
Video 8minWhy asking for steps works, compared on a discount and tax calculation, then applied to a support agent that reasons through policy before replying. Also covers when it is worth the extra cost, why it exposes errors rather than removing them, and how to combine it with examples.
Video 6min