AI Is Still Surprisingly Bad at Writing and Judging Comedy, Says Dev
ivan_bezdomny · x · 2026-11-11
ivanbezdomny argues models remain extremely poor at writing and judging comedy — beyond personal taste, training on great humor likely yields little beyond obvious basics, with occasional good output mostly due to randomness. enginoid proposes a mass-annotation site with pairwise preferences to train a reward model for humor tasks; ivanbezdomny says it would work, but progress would be far slower than model capabilities elsewhere suggest.
More from Models
- Google staffer teases Gemini 4 Carbon: capable, but a stronger internal model is coming — jason_mayes · 2026-11-11
- Greg Isenberg Tests Fable 5.2, Calls It Possibly the "Best Thinking" Model, Shares 12 Prompts — FinanceYF5 · 2026-11-11
- Gemini 4 Carbon Burns 45 Minutes and 262k Tokens, Delivers No Code Review — No-Park606 · 2026-11-11
- Anthropic's Fable 5.2 Formalizes Fermat's Last Theorem Proof in 30 Hours with ~1,000 Agents — sven_ai · 2026-11-11
- Google poised to drop Gemini 4 Hydrogen (Flash) on second straight day of releases — gaganghotra_ · 2026-11-11
- Wave of tiny TTS releases: 8M-param Paradee hits 0.69% WER on Open TTS Leaderboard — LysandreJik · 2026-11-11