Chain-of-thought — writing out steps
Generating intermediate reasoning before the final answer. Each written token buys another pass of computation and becomes working memory for later tokens.
Reasoning models are trained to produce long chains with reinforcement learning on checkable problems (Module 22).