AI Foundations Playgrounds · Beginner
Temperature Playground
Slide temperature from 0 to 2 and watch next-token odds, sampled completions and use cases change live.
Temperature slider
Live sample outputs
When to use what
Understanding Temperature Parameter
Low Temperature (0.0-0.3)
AI gives predictable, focused responses. Always picks the most likely next word or phrase.
Deterministic: Same input → Same output
Perfect for: Facts, coding, formal writing
Medium Temperature (0.4-0.7)
Balanced between accuracy and creativity. Introduces controlled randomness for varied responses.
Balanced: Reliable yet varied
Perfect for: Conversations, content creation, explanations
High Temperature (0.8-1.0)
AI becomes highly creative and unpredictable. May produce surprising or unusual responses.
Creative: Unexpected but diverse outputs
Perfect for: Brainstorming, creative writing, art
How Temperature Works
Temperature controls the randomness in AI responses by adjusting how the model selects the next word. Lower values make the model more confident in its top choices, while higher values allow it to explore less likely but potentially creative options.
Think of it like cooking: Low temperature = following the recipe exactly, High temperature = experimenting with new ingredients!
Recommended Use Cases
0.5
Next-Token Sampling Lab
An LLM scores every possible next token (logits), turns the scores into probabilities with softmax(logits / T), then samples one. Move the slider above and watch the odds change.
–chance of the top token
–entropy (bits) — higher = less predictable
Click “Sample 5 completions” to draw tokens at the current temperature.