Temperature
Level 2
A setting that controls how predictable or varied a model's output is.
Term 5 of 8 in Language models
In plain language
At a low temperature the model plays it safe and picks the most likely next token. At a higher temperature it takes more chances, which reads as more creative and also more error-prone.
Think of it like this
A dial between "careful accountant" and "brainstorming session".
Why it matters
Knowing this setting exists explains why the same question can produce a slightly different answer each time.
Next terms in Language models
- Large language modelA model trained on enormous amounts of text to predict what words come next.
- TokenA small chunk of text that a model reads and writes, usually part of a word.
- PromptThe instruction and information you give a model.
- Context windowThe maximum amount of text a model can consider at one time.