Observed arrival · 2026-11-06
Humor Data is building an AI test for comic timing
A benchmark and training-data project for testing when AI should joke, how it should deliver a line, and how it should recover when it misses.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Teams building conversational AI with a distinct personality
- Worth noticing
- Each situation gets a safety score from 1 to 5 and a specified fact that would flip the appropriate response.
Field notes
The central test uses minimal pairs: a robot drops a mug, and whether it breaks changes the appropriate response from joking to cleaning up. Each situation carries a 1–5 safety score and a specified flip fact. The page says lines, explanations, and delivery scripts are written and rated by people, with rights cleared for training; it also labels benchmark version 0.1 as in the works.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
$PaidCommerce or pricing visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
◎NicheUnusually specific use
One card from the complete issue