The one-sentence takeaway

Asking a large model to think step by step, rather than answer instantly, lets it use its own written steps as scratch paper and dramatically improves its reasoning, a free upgrade that needs no retraining but only switches on once models are big enough.

Next: Chapter 2, ReAct, where the model stops just thinking and starts acting in the world.