AI Glossary Term
What does “Temperature” mean?
Definition
A sampling setting controlling randomness in generated outputs.
How Temperature works
Example
At a low temperature, asking for a name for a bakery tends to return the same safe option each time. At a higher temperature, you get more varied and unusual suggestions, with a greater chance of odd ones.
Why it matters
It lets you trade consistency for variety. Tasks needing accuracy generally use lower values.
Common questions
Does temperature 0 give identical output?
Usually very close to deterministic, though small variations can remain in some systems.
How does it work?
It rescales the probabilities of next tokens before one is sampled: lower sharpens the choice, higher flattens it.
Does high temperature cause hallucinations?
It can raise the chance of odd or incorrect output, though models can be wrong at any setting.