LEVEL 01 · LESSON 03
Temperature
Temperature changes variation in generation, not whether the answer is true.
What it is
Temperature is a generation setting that changes how strongly a model favors likely next tokens. Lower settings generally concentrate the choices; higher settings allow more variation.
Why it matters
A brainstorming task can benefit from variety, while a routine classification usually needs consistency. This is a behavior choice, not a quality score.
A business example
A marketing team asks for five campaign headlines. More variation can help exploration. The same team’s process for assigning an approved campaign category benefits from constrained choices and validation instead.
How it actually works
At each generation step, a model assigns scores to possible next tokens. Temperature changes the distribution used to select among them. Other sampling settings also affect the result, and some models do not expose temperature. A low setting can repeat a wrong answer consistently. For business rules, combine explicit criteria, structured outputs, and tests; do not rely on sampling settings to enforce policy.
A common misconception
Temperature is not a truthfulness dial. A setting of zero does not guarantee factual accuracy or identical results across infrastructure and model changes.
ONE SENTENCE TO REMEMBER
Temperature changes variation in generation, not whether the answer is true.
CHECK YOUR UNDERSTANDING
An invoice-classification workflow sometimes assigns the wrong category. A colleague proposes setting temperature to zero to fix it. What is the best response?
Show the answer
Answer: C.
Temperature changes how much generation varies, not whether an answer is true. A low setting can make the same error repeat consistently. Enforce business rules with explicit criteria, structured outputs, and tests.
Take it into your next meeting
Does your task need diverse options or a repeatable decision format?
One more concept, understood.
Mark this lesson complete to keep your place.