Course 6, lesson 57 of 100, Ages 12+
Step-by-step reasoning
Why thinking out loud helps
Like I’m 5
When you solve a tricky sum, you write down the steps. Some AIs do better when they think step by step too, instead of blurting out an answer.
The big idea
For multi-step problems, models are more accurate when they work through intermediate steps before answering. Asking 'think step by step' or giving worked examples can help.
Newer reasoning models are trained to spend extra effort thinking before they reply, trying approaches and checking their work. That improves maths, coding and planning, though their steps still need checking.
Examples
- Word problems: Breaking 'If 3 pens cost 45 rupees...' into steps.
- Planning: Listing what to pack before a trip, item by item.
- Checking: A model re-solves a problem a different way to confirm its answer.
How it works
- Break the problem into smaller steps.
- Solve each step and write down the result.
- Check the final answer against the question.
Check your understanding
- Why does thinking step by step help AI?
- Options: Multi-step problems are easier to solve in pieces; It makes answers shorter; It uses fewer words.
Answer: Multi-step problems are easier to solve in pieces. Working through intermediate steps reduces mistakes. - Should you still check an AI's reasoning steps?
- Options: Yes, steps can contain mistakes; No, they're always right; Only on Mondays.
Answer: Yes, steps can contain mistakes. Reasoning helps, but errors can still slip in.
Remember
Step-by-step thinking helps AI with multi-step problems, but steps still need checking.
Talk about it
Solve 'What is 15% of 80?' out loud, step by step. How does it help?
Go deeper
Chain-of-thought prompting (Wei et al., 2022) showed large gains on reasoning tasks. Reasoning models are trained with reinforcement learning to use more test-time compute before answering.