Knowledge Maxxing← Back to topic library
AI MaxxingBeginner60 sec
Temperature
Temperature is a setting that controls how random or predictable a language model's output is when choosing its next word.
Prerequisite: Next-token prediction
What matters
A low temperature makes the model pick its most likely next token almost every time, producing consistent output; a high temperature lets it pick less likely tokens more often, producing more varied output.
Example
Asking a model to write a poem at high temperature might yield a different, more surprising poem each time, while low temperature gives nearly the same safe answer every run.
Common misconception
Temperature does not control how 'smart' or accurate the model is; it only controls output variability, not correctness.
Recall question
What effect does raising the temperature have on a model's output across repeated runs?
Keywords and related concepts
temperaturerandomnesssamplingoutput variability