Temperature
The Creative vs The Accountant
5 min read
The Creative vs The Accountant
When you need a poem, you want wild creativity. When you need tax figures, you want precision.
Turn the temperature up and the AI becomes adventurous, creative, and sometimes surprising — like a writer who's had two cups of chai. Turn it down to near zero and it becomes precise, predictable, and literal — like an accountant. Temperature controls how random vs. deterministic the model's token selection is.
In Plain English
Temperature controls how creative or predictable the AI's responses are. High temperature (0.8–1.0) = more creative and varied. Low temperature (0.0–0.2) = more focused and consistent. For factual tasks, use low temperature; for creative tasks, use higher.
The Technical Picture
Temperature scales the logits (raw model output scores) before the softmax function, affecting the probability distribution over the vocabulary. High temperature flattens the distribution (more uniform, higher entropy), increasing diversity. Low temperature sharpens it (peaked around highest-probability tokens), increasing determinism.
Real-World Examples
- Temperature 0.0 for medical diagnosis assistance — always the safest answer
- Temperature 0.9 for brainstorming 50 product name ideas
- Cursor IDE uses low temperature for code generation to ensure reliability
Temperature is the creativity dial — low for facts, high for ideas.