Controlling how output is generated.
WHAT TEMPERATURE CONTROLS
How much randomness is in the model's word selection.
Low: predictable, consistent, conservative output
High: more varied, more creative, less reliable
WHEN TO USE LOW
Factual tasks Extracting structured data Anything requiring consistency Code
WHEN TO USE HIGHER
Creative writing Brainstorming Generating varied options
WHY THE SAME QUESTION GIVES DIFFERENT ANSWERS
Randomness in selection. At a low setting, outputs are more consistent but rarely identical.
OTHER COMMON SETTINGS
Maximum output length, which caps the response Top-p, an alternative way of controlling randomness Stop sequences, telling the model where to end
WHERE YOU SET THEM
In API access, and in some interfaces offering advanced options.
Most chat interfaces set sensible defaults and do not expose these.
THE PRACTICAL ADVICE
If output is too erratic, lower the temperature. If it is repetitive and dull, raise it.
For most business writing, a lower setting produces more usable results.
