Chain-of-thought prompting asks a language model to produce intermediate reasoning steps before its final answer. It turns a direct question into a sequence of simpler steps, giving the model an explicit scratchpad during inference. section 7.5.3
The technique grew from informal prompts such as “Let’s think step by step.” It can improve multistep reasoning without changing model architecture or parameters. Few-shot chain-of-thought reasoning also acts as a scaffold: performance can improve, but reasoning remains brittle when rules change. section 7.5 section 7.5.3
A multistep word problem
For a farm with 23 cows and 17 chickens, the model can work through the animals’ legs before giving a total. This illustrates why an explicit sequence can help where a direct answer may fail. section 7.5
