Nexuldra
// ai
Temperature (LLM)
Temperature is a numeric setting — usually between 0 and 1, sometimes up to 2 — that controls how randomly a language model picks its next word. Low values make output deterministic and consistent; high values allow more surprising, varied phrasing. For factual, repeatable tasks like support answers, temperature near zero is standard; creative drafting tolerates more variety.
Temperature matters because the same question should get the same answer when facts are at stake. Setting it near zero removes one source of drift between test and production: the bot you approved on Tuesday behaves identically on Wednesday. It does not make a wrong model right — a system missing the facts still invents them confidently at zero — but it makes a correct system consistent enough to ship and to keep shipping.
In practice: a customer support bot answering from a manual runs at 0 to 0.2, so replies are stable enough to evaluate and cite. Marketing copy generation might use 0.7 to 0.9 for variety across dozens of drafts, where inconsistency is the point and a human edits anyway. If you find yourself raising temperature to make a boring answer sound livelier, the better fix is usually the prompt, or a model better suited to the task.
Frequently asked questions
- Does temperature zero guarantee identical answers?
- Almost, but not absolutely: identical requests can still vary slightly because of how providers batch parallel requests on their servers. The difference is small and shrinking, but if your system needs bit-for-bit stability, verify it in code rather than trusting the setting.
- What temperature should a support chatbot use?
- Near zero — 0 to 0.2. Support answers should be stable, testable and consistent across users, and variety adds nothing. Reserve higher values for genuinely creative tasks, and even there, plan for human review of the output.
// go deeper
Full treatment in the guide: How to stop a chatbot from making things up