If you've ever asked the same question twice and got two different answers, you've seen AI randomness at work. Settings like temperature control how adventurous the model is when choosing words.
What temperature does
At each step, the model picks the next word from many possibilities. Low temperature makes it choose the most likely words, giving predictable, focused answers. High temperature lets it pick less likely words, giving more varied and creative output, but also more mistakes.
When to use low vs high
- Low: data extraction, code, factual answers, classification, consistent formats.
- Medium: everyday writing, emails, explanations.
- High: brainstorming, poetry, fiction, naming ideas.
In chat apps
Most consumer chat apps don't show a temperature setting. You can still influence it with your prompt: "Give me the most standard answer" vs "Give me unusual, surprising ideas".
In APIs and image tools
Developers can set temperature directly. Image and music generators have similar controls, like "stylize", "chaos", "guidance" or seed values, which affect how literal or creative results are.
For consistency
Use clear instructions, examples of the desired output, low temperature and, in image tools, a fixed seed.