Temperature

Think of temperature as how much freedom a chef has. At a low temperature, they follow the recipe exactly, and it tastes about the same every time. At a high temperature, they improvise: sometimes amazing, sometimes a disaster. This is part of why you can get different answers when you ask the same question twice.

In AI coding tools, this setting is usually already tuned by the tool. Regular users don't need to change it, and may not be able to. It mainly comes up when you call a large model API yourself to build a product feature. For tasks that need to be accurate and consistent, like writing code, extracting data, or classifying, use a lower temperature. For tasks that need variety, like naming, writing copy, or brainstorming, you can turn it up.

Note that a low temperature doesn't mean no mistakes. The model can still hallucinate. Its output is just more consistent.