Study: LLMs share 3/4 of humor taste, prefer absurd and roast jokes, Gemini most consistent
Gold-Bat-3225 · reddit · 2026-09-16
A report from laugh.so tests LLMs' humor taste:
- Models overlap on humor judgments by roughly 3/4, but individual agreement ranges from 54.9% to 92%
- Gemini is the most consistent judge, though all models are less consistent than humans
- Nearly all models show a strong preference for absurd humor and roast-style joke formats
Full breakdown in the report.
More from Fun
- User jokes OpenAI lacks the compute to grant 10x usage limits — CtrlAltDwayne · 2026-09-16
- 'That e is friskier than the others': tokenization humor makes the rounds — AI_Andrew · 2026-09-16
- e/acc Shirt Meets LessWrong Hoodie: The Funniest Contradiction in AI Twitter — eigenron · 2026-09-16
- One User Wants an Autonomous EV RV Based on Tesla's Adaptable Robovan — JoeJustice · 2026-09-16
- AI community braces for AGI-level OpenAI DevDay announcement, dubbed 'D-Day' — scaling01 · 2026-09-16
- Recreating the iPhone Duo app on an Android foldable with AI, in native Material Design — op7418 · 2026-09-16