Temperature

A low temperature makes the model pick the most likely words, so results are more consistent. A high temperature gives variety, useful for ideas and creative drafts. For data extraction, classification or answers that must repeat, use a low temperature. A low temperature does not make the model more truthful; only more consistent.