Temperature and Sampling: The Setting That Controls How "Creative" an AI Is
Many AI tools expose a "temperature" setting without explaining what it does. It controls how the model picks its next word at each step — and that single number has an outsized effect on how predictable or varied the output feels.
- Temperature controls how predictable vs. varied a model's word choices are.
- Low temperature suits factual questions, code, and anything with a "correct" answer.
- Higher temperature suits brainstorming and creative writing, where variety is the goal.
What it actually controls
At each step, a model assigns probabilities to many possible next words. Low temperature makes it almost always pick the highest-probability word, producing focused, predictable text. Higher temperature lets lower-probability words get picked more often, producing more varied — and more error-prone — output.
When to go low
Factual questions, code, data extraction, anything with a "correct" answer — low temperature reduces the chance of the model wandering off into a plausible-sounding but wrong tangent.
When to go high
Brainstorming, creative writing, generating varied options — higher temperature is genuinely useful here because "predictable" is the opposite of what you want.
Frequently Asked Questions
What does the temperature setting do?
It controls how likely a model is to pick lower-probability words at each step — low temperature produces focused, predictable text; high temperature produces more varied, less predictable text.
What temperature should I use?
Low for factual/technical tasks where accuracy matters most, high for brainstorming and creative work where variety is the goal.
Keep Learning
Continue Exploring