One prompt, five runs, two temperatures.

To see the temperature dial do its work: the same prompt repeated at temperature 0 and at temperature 1, once for a question with many good answers and once for a question with one.

Somewhere the temperature can be set. Any of these is free to try: the OpenAI Playground, the Anthropic Console playground, or Google AI Studio. Each has a temperature slider. If you have none, use the fallback in step 5.

1. Set temperature to 0. Paste this and run it five times (a playground keeps no history between runs unless you add it):
   Give me a name for a small bookshop in Pune. One name only, no explanation.
   Record the five names.
2. Set temperature to 1 (or the highest the tool allows). Same prompt, five runs. Record.
3. Temperature 0, a factual prompt, five runs:
   In one sentence, what does a context window mean in an AI model?
   Record whether the five sentences are identical, nearly identical, or different.
4. Temperature 1, same factual prompt, five runs. Record the same.
5. Fallback without a playground: in a normal chat tool, ask the bookshop question in five new chats and record the names. Then in five more new chats add "Give the single most likely name, the one you would give if you had to answer the same way every time." It is a weaker version of the same thing; note the difference.

Model and tool used:

At temperature 0, five identical outputs, usually to the character. At temperature 1, the bookshop names should all differ, and the definition should keep its meaning while a word or two changes. If your temperature-0 runs were not identical, write that down: it happens, and the note's "one layer deeper" says why. If your temperature-1 definitions differed in meaning, not just wording, that is worth a margin note too.

one line: which question did the dial actually change?