noun · AI setting
Temperature
also called sampling temperature
A dial for how adventurous the AI is. Turn it down and it plays it safe, turn it up and it takes more chances.
A sampling setting that controls how random token choice is: lower values make output more predictable, higher values make it more varied.
Before a token is chosen, the model's scores are divided by the temperature. A low value sharpens the gap between likely and unlikely tokens, a high value flattens it, so unlikely tokens get picked more often.
Real-life example
You ask a model to name a pet goldfish. With a low setting you get the same handful of names each time. With a high setting you get odd and surprising ones, and sometimes a bad one. For extracting data from an invoice you would want the low setting, because there is one right answer.
Playground
Slide the temperature and watch the odds of each next word change, then draw words to see what the dial really does.
Settings
My favorite pet is a …
Entropy of the choice: 2.03 bits. Each score is divided by the temperature, then turned into odds with softmax.
What comes after “My favorite pet is a”
The grey number is the raw score (logit) before temperature.
Press “Pick 100 times” to see how often each word wins.
Illustrative: made-up scores for one example sentence. A real model scores tens of thousands of possible tokens at every step.
Where the analogy breaks
-
Temperature is not a creativity or intelligence dial. It only changes how much chance is in the choice between candidates the model already finds plausible.
-
Zero does not guarantee identical outputs every time. Small differences in how the model is served can still change the result.
-
Allowed ranges and defaults differ between providers. Check the documentation for the model you use instead of copying a number.