Why the same prompt gives different answers — temperature, explained in 40 seconds

Ajouté : 2026-05-05

[00:00:00]You send the same prompt twice. You get two different answers. That's on purpose. Every time the model picks a word, it's choosing from a list. Each word has a probability. The highest one is the safe bet. There's a knob called temperature. Low temperature sharpens the list. The top word wins almost every time. High temperature flattens it. Now any word has a shot. So the exact same prompt rolls. Different dice every call.

[00:00:26]Same question, different answer every time.

[00:00:30]Need reproducibility? Set temperature to zero and pass a seed. You'll still see tiny drift from GPU scheduling, but the big randomness goes away.