Chain of Thought (CoT) Prompting is a technique introduced in 2022 by the Google Brain team. Ref: Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
It involves prompting a Large Language Model to generate a series of intermediate reasoning steps (“Chain of Thought”) before giving the final answer.
Mechanism
Instead of asking for a direct answer, the prompt includes examples of step-by-step reasoning.
- Standard Prompting: Input Question -> Output Answer. (Model often fails on complex tasks).
- CoT Prompting: Input Question -> “Roger started with five balls… 5+6=11” -> Answer is 11.
By showing the model how to think through a problem with examples, the model learns to apply this step-by-step logic to new, unseen questions. This is a key method for Inference-Time Compute Scaling because it forces the model to use more tokens (compute) to generate the reasoning path.
