In AI, the term Temperature refers to the prediction tolerance permitted.
If set to 0, then predictions are tight and only the most defined, logical path is selected.
At this setting, asking the same question multiple times will reliably give you the same, predictable, safe, boring answer.
If set to 1, wild predictions and inconsistencies may appear.
Answers become less likely realities, but normally sit within the realm of possibilities. Asking the same question repeatedly will give wildly different answers that may bear little resemblance to reality.
Typical temperature for practical use is usually set around 0.8
If you want to have an AI on acid, then, if allowed, set the temperature to 1.5 and watch the show 😂
Temperature is another one of those wildly misleading labels, so much so, my private LLM had to warn me it had nothing to do with the weather multiple times.
I have therefore renamed "Temperature" to "Wiggle Room" for my internal AI discussions.
Below is the screen shot of one of my local LLMs warning me of its improper association 😂
