Why Are LLMs Still So Bad at Jokes? Great at Math, Stuck on Humor
michaelsoft__binbows · reddit · 2026-11-14
A Reddit post highlights a striking contrast: LLMs have made a dent in mathematics yet remain unable to write genuinely funny content. The author used humor as a general-capability eval two years ago, fearing cheap slop could scale into a Wall-E future — but that scenario never materialized because the humor never got good.
Testing different reasoning tiers (Astra Light/Medium/High) with the same joke setup produced structurally competent but unfunny examples — from a stale parrot joke to a mildly amusing AI-intern bit; the High tier even spent 24s thinking. The author asks why no model, including open-weight abliterated ones, produces compelling humor.
More from Fun
- Only one plane over Miami at 6:30pm Friday as FAA ground stop hits — MicahBerkley · 2026-11-14
- OpenAI ships staff a ModRetro Chromatic handheld with a mystery 'Codex Cartridge' — OpenAIDevs · 2026-11-14
- Developer threatens to quit posting over flood of AI "reply slop" on X — gem_ray · 2026-11-14
- Reward hacking 101: tell AI to minimize delayed flights and it grounds them all — KordingLab · 2026-11-14
- Fan-made One Piece cover casts Jensen Huang as captain with CEOs hunting GPU treasure — HenryEXV · 2026-11-14
- Denny's Reportedly Removes Plaque Honoring the Booth Where Jensen Huang Founded Nvidia — Polymarket · 2026-11-14