Chat Model Behavior Sandbox

See how chat model parameters change tone, length, creativity, and safety behavior.

Prompt & Parameters

Select a prompt above via one of the provided buttons, then click Generate Comparison to see how model behavior shifts.

Parameter Controls

These are simplified versions of the knobs you’d configure on a real chat model.

0.7

Higher values → more varied and creative responses; lower → more focused and predictable.

256

Rough limit on response length. Higher values allow longer, more detailed answers.

Top-p style

Conceptual sampling mode: toggle between narrower vs. wider token exploration.

Standard

Simulates stricter guardrails. When enabled, unsafe prompts get more cautious responses.

Behavior Comparison

No runs yet. Adjust the parameters and generate a comparison.
Response A Uses your chosen parameters.

Waiting for first run.

Response B Automatically adjusts parameters to highlight contrast.

Waiting for first run.

Behavior Notes

As you generate comparisons, this section will translate parameter shifts into plain-language observations you can use in interviews (e.g., “Higher temperature creates more varied phrasing, but also more rambling.”).