What does “Chain-of-Thought” mean?
A reasoning style where an AI model decomposes a problem into intermediate steps.
How Chain-of-Thought Works
Question: 23 apples, use 20, buy 6
Step 1: 23 - 20 = 3
Step 2: 3 + 6 = 9
Answer: 9.
Practical Example
Asked a word problem, a model first writes out intermediate steps, such as subtracting then adding, before stating the final number. Research has shown that prompting for step-by-step reasoning can improve accuracy on math and logic problems compared with answering directly.
Why It Matters
Showing steps can improve results on multi-step problems and lets people spot where an answer went wrong. The written steps, however, may not always faithfully reflect how the model reached its answer.
Common Questions About Chain-of-Thought
Q: Do I have to ask for chain-of-thought?
Often a simple instruction like 'think step by step' helps. Some newer models are trained to reason in steps automatically.
Q: Does chain-of-thought guarantee a correct answer?
No. A model can write convincing steps that contain errors, so results still need checking.
Q: Does it cost more?
Yes, usually. Writing out steps uses more tokens, which increases cost and response time.