Tutorials · Chapter C (3/4) · ~6 min
Temperature — safe vs creative
Play → read → next
Temperature controls how safely vs wildly the model picks the next token.
Playground
Temperature: safe vs creative
Same prompt: “Name a friendly pet.” Slide Safe → Wild, then match dials to jobs.
Sample outputs (balanced)
Buddy · Nova · Pepper · Ziggy
Recap
What you just did
You turned temperature from safer to wilder and watched next-token choices change. Low = stick to likely words. High = more variety (and more weirdness).
Teach
How it works
At every step the model ranks possible next tokens. Temperature reshapes those odds:
- Safer — strong favorites win more often (good for facts, formats, code)
- Wilder — underdogs get a bigger chance (good for brainstorms, flavors, stories)
Wild ≠ wiser. It just means “less stuck on the most common path.”
Use it
When you'd use this
- Low/safe: summarize a policy, write a checklist, draft formal email
- Higher: name ideas, metaphor brainstorms, playful rewrites
Watch out
Watch out
Do not crank temperature hoping for more accurate science or legal advice. Creativity and correctness are different jobs.
Try next
Try this next
Ask for five product names twice — once careful, once wild. Notice which list you’d show a client vs a brainstorm board.