also sampling, top-p, creativity setting
A setting that controls how much randomness the model uses when it picks each next token. Low values give consistent, repeatable output; higher values give more varied wording, and more varied mistakes.
You have done this if
Support answers changed wording every time the same question was asked, until you lowered the temperature for that feature.
Say it in a review
Extraction and routing run near zero temperature; only the drafting feature is allowed variety.
On the AI Application map Model