Why LLMs Aren't Funny: Three Reasons From Training to Incentives
paulabartabajo_ · x · 2026-10-21
A video breaks down three reasons LLMs fail at comedy and what AI labs—and users—could do about it.
- Next-token prediction inherently kills the unexpected, which is the core of comedy
- Humans fundamentally disagree on what's funny, making it hard to optimize
- From an incentive standpoint, funny models don't pay off, so labs have little motivation
It closes with a proposal for addressing these issues.
Related event: Why LLMs Aren't Funny: Video Breaks Down Three Key Reasons(2 posts)→
More from Fun
- Australia's cyber agency alert page goes 503 amid critical PHP vulnerability saga — cyb3rops · 2026-10-21
- Developer Turns LLM Structured Output Into a Playable Token-Picking Game — nicodotdev · 2026-10-21
- GPT 6.1 mocked for another silly fail in viral screenshot — sven_ai · 2026-10-21
- Security meme: best exploit mitigation is 3 attack surface reductions in a trenchcoat — dyn___ · 2026-10-21
- Opus 5.5 rebuilt a weekend-long photography simulator in one day, 60fps on iPhone — andyleenz · 2026-10-21
- 30 Apple products that don't exist (yet) — Delahuntagram · 2026-10-21