Testing GPT-6 Astra's weak spot: mostly bland jokes, one likely original
paraschopra · x · 2026-09-05
Amid a flood of impressive GPT-6 Astra demos, @paraschopra tested its weak spot: asking the model to write original programming jokes that can't be found online.
Result: the first joke made him chuckle and appears genuinely unique after a web search; the rest were bland. His takeaway: AI progress is jaggy — in unverifiable domains like humor, where quality can't be objectively checked, full automation remains far off.
More from Fun
- Claim that GPT-6 Astra solved Hadwiger–Nelson with chromatic number 7 meets skepticism — MikePFrank · 2026-09-05
- Qwen3.8-27B beats the Wikipedia game in 6 clicks using Playwright — swagonflyyyy · 2026-09-05
- AI Agent Plays Rimworld, Writing Its Own Mods and Taking Notes as It Learns — jxnlco · 2026-09-05
- Astra confidently wrong in real-world test, changes caused performance regressions — cnakazawa · 2026-09-05
- Twin primes is not a Millennium Problem, mathematician corrects viral claim — littmath · 2026-09-05
- First weekend with AGI: planned to go for a run, ended up watching it run instead — andrew_n_carr · 2026-09-05