GLOSSARY
Temperature
For supported language models, temperature is a setting that affects the selection of possible next tokens. It can influence variation in outputs. Its availability and effect depend on the model; a low value is neither a measure of truth nor a general promise of complete reproducibility.
Distinguish between sales tasks
Different wording can help when exploring possible conversation openings. When transferring a confirmed delivery condition, preserving the content matters more. These tasks need different substantive criteria even if one application handles both.
A technical setting should therefore not be equated automatically with creativity, reliability or intelligence. A consistently wrong answer remains wrong. Varied wording may also communicate the same useful meaning. The relevant question is which differences are acceptable in the business process.
Compare settings with suitable cases
Run repeated versions of the same task and assess both variation and correctness. Record the model version and other settings so that changes can be interpreted. This distinction is especially useful for probabilistic methods. It does not automatically establish semantic fidelity: conditions, figures and commitments still need comparison with the source. Teams should choose settings based on that evidence and review them when the underlying model changes.