deliberate vs. intuitive responses

There is a useful analogy from psychology between fast, intuitive thinking and slow, deliberate thinking. A model answering in one quick pass is like the fast mode: fluent, immediate, often right on familiar ground, but prone to the confident slip when a question is designed to trip the snap response. A model that writes out reasoning, considers cases, and checks itself is acting like the slow mode: more effortful and expensive, but far steadier on problems where the obvious answer is a trap.

Reasoning models and chain-of-thought are essentially ways of pushing a model from the fast mode into the slow one on demand. The analogy should be held loosely, since a model has nothing like human gut feeling and its slow thinking is still token prediction, not a different mechanism. Still, it captures a real and practical trade-off. Quick answers are cheap and usually fine; deliberate answers cost more compute and time but pay off when accuracy matters, and a large part of using models well is knowing which mode a given task deserves.