1.5: Chain-of-Thought
What You'll Learn
- Why "just give me the answer" produces shallow or wrong conclusions
- How to trigger chain-of-thought reasoning, and why defining the steps beats "think step by step"
- How to use structured reasoning for analysis and decision-making
- Thinking frameworks you can hand to the AI: pros/cons, red team, first principles, and more
Key Concepts
The Problem with Jumping to Conclusions
Asked for a direct answer to a complex question, AI will produce one, confidently, even when it skipped steps. You can't see where the reasoning went wrong because there's no reasoning to look at.
Chain-of-Thought Prompting
Asking the AI to work through a problem step by step before concluding changes the output. "Think step by step" is the simplest version. The stronger version is structured CoT, where you spell out the steps: first identify the key figures, then compare them, then list the risks, then make a recommendation.
Why It Works
- Better accuracy: the AI catches its own errors as it goes.
- Transparency: you can see and check the reasoning.
- Debuggability: if the conclusion is wrong, you can find the step that broke.
- Control: you can redirect the analysis partway through.
Use it for analysis, decisions, comparisons, math, and anything with multiple steps.
The Exercises
Working from a set of raw meeting notes in the course files, you'll design your own reasoning steps and run an analysis of a real business decision, watching each step build on the last. Then you'll try a thinking framework on a decision from your own work.
How to Start
This lesson is taught interactively by your AI. Download the course materials, open your AI tool (Claude, ChatGPT, or similar), and say:
start lesson 1.5