What is temperature?

Definition

Temperature is a dial that controls randomness in an AI model's output. At a low setting the model sticks to its most likely wording and gives similar answers each time. At a higher setting it takes more chances, which can be useful for brainstorming and worse for accuracy.

Chat apps usually set it for you. It's exposed mainly in APIs and some business tools.

Why it matters for advisors

For compliance-sensitive tasks (summaries, data extraction, rule explanations) consistency is what you want. If your firm builds a workflow through an API, a low temperature is a simple way to reduce variation.

It also explains why the same prompt can give different answers on different days. That's normal, and it's another reason to review output rather than assume it's stable.

Low temperature doesn't mean accurate. A model can repeat the same wrong answer consistently. Temperature controls variety, not truth, so it's no substitute for checking figures and facts against the source.

Example

A firm's developer sets a low temperature on an internal tool that pulls figures from statements into a template, and a higher one on a tool that suggests blog topics.

Watch out for

Low temperature doesn't stop hallucinations. It only makes the model's answers more repeatable.