PAPER 13 · Reasoning prompts

Chain-of-Thought Prompting Elicits Reasoning in Large Language Models

Wei et al. 2022 Paper

Showed that reasoning behavior can be elicited through examples and intermediate reasoning steps.

Core concept

Models often reason better when the prompt shows worked steps instead of only final answers.

Why it mattered

It revealed that prompt format can unlock latent reasoning behavior at inference time.

Visual shortcut · Show the work
question
reasoning steps
answer
hard tasks improve

Chain-of-thought prompting works because examples teach the model to decompose instead of guess.

How it works
Give examples with reasoning steps.
Ask the model to follow that format.
Let it break the problem into pieces.
Use the intermediate path to improve the final answer.

The quick digest

This paper is about a deceptively simple change: when giving examples to a large model, include the intermediate reasoning, not just the answer. For arithmetic, commonsense, and symbolic tasks, that can materially improve performance.

The nontechnical intuition: the model has seen many explanations and solution paths during training. If your prompt demonstrates that the task should be solved by decomposing it, the model is more likely to follow that pattern instead of jumping to an answer.

This becomes the root of a major line of reasoning work. Later systems hide the scratchpad, train reasoning behavior directly, or spend more inference compute, but the seed is here: the shape of the answer can change the quality of thinking.

What to remember

One-liner
Showing the process can improve the answer.
Why it matters
Prompt format can unlock latent reasoning.
Builder instinct
Written reasoning is useful behavior, not guaranteed truth.

Read it like this

Build instinct

Run the same math and planning tasks with direct answers, short reasoning, and structured step-by-step prompts.

Read source → All papers
Previous12 · Direct Preference Optimization