Lesson 33 · 2 min
How chatbots "reason" step by step
Reasoning models write out working notes before they answer, which helps on problems with many steps.
A plain chatbot writes its answer straight away, one word after another. For a hard maths or logic problem, that is like answering without scrap paper.
Reasoning models are trained to first write a chain of thought: break the problem into parts, try an approach, check it, and fix mistakes. Only then do they write the final answer. Some apps show a summary of this thinking, others hide it.
This extra thinking takes more time and computing power, so reasoning modes are often slower. They help most with planning, maths, code and tricky comparisons. For a quick email, they add little.
The working notes are useful but not a perfect window into the model. They can still contain errors, and the final answer still needs checking.
Do it now
Compare quick and careful
- 1. Take a small puzzle, like the one in the prompt below.
- 2. Ask a chatbot in its normal mode, then in a thinking or reasoning mode if it has one.
- 3. Compare the answers and how long each took. Check the maths yourself.
A shop sells pens in packs of 6 for 20,000 dong and packs of 10 for 30,000 dong. I need exactly 38 pens. What is the cheapest way to buy them? Show your steps.
Quick check
Keep going
Next →What AGI meansWords from this lesson
- Reasoning model
- A model that spends extra time working through a problem internally before replying, which helps with maths, logic and planning.
- Chain of thought
- Asking an AI to work through a problem in steps before answering, which often makes its answers more accurate.