ARTFunny Research Bureau Experiment status: ongoing
ARTFunny
Menu

Station 06 · Generative comedy controls

Comedy Lab

Set four creative constraints and ask the model for one original, general-audience joke. The instrument returns the attempt alongside an honest mechanism reading and possible failure point; a complete prebuilt specimen remains visible when live AI is unavailable.

Mode
Live AI + fallback
Storage
No site retention
Network
Sent to OpenAI
Generator controlsServer-side AI · no saved history
Prebuilt fallback demonstrationAI-generated comedy and analysis can be inaccurate or subjective.

Generated specimen

My houseplant survived its performance review, but management says it needs to branch out.
Comedy type
one-liner
Primary mechanism
wordplay
Secondary mechanism
personification
Attempted humor
The line treats plant care like office management, then uses “branch out” in both its career and botanical senses.
Why it may work
The office premise is familiar, and the final phrase delivers a compact double meaning.
Possible weakness
The phrase is common enough that some readers may predict the plant pun before it arrives.

Constraints change the attempt

Comedy type and tone are overlapping instructions, not sealed categories. A dry pun may still be a one-liner, while an observational premise can become absurd as surprise rises. The controls make those choices explicit so the result can be inspected as an attempted response to a brief rather than a spontaneous proof of machine wit.

Surprise level describes the requested distance between setup and turn. It does not set an objective funniness score. A restrained line may fit its subject better than a spectacularly strange one, and a model can misunderstand any control.

Why every output includes a weakness

Generation alone can hide the most useful part of the experiment: whether the mechanism is predictable, strained, generic, or overexplained. The structured readout asks the same model to identify both an intended strength and a possible weakness, giving visitors claims to evaluate rather than a verdict to accept.

Requests are validated, moderated, rate-limited, and handled on the server. The tool does not offer personalized roasting or targeted harassment. Do not enter confidential, identifying, or sensitive information because submitted text is sent to OpenAI for processing.

Continue the investigation

Read the mechanisms behind the machine.