HumanBench: A Fun Benchmark That Tells You Which Model Size You Are
ziqiao_ma · x · 2026-09-28
Developer @Alexybuild launched HumanBench, a tongue-in-cheek benchmark built for humans: answer a quiz and find out your equivalent model size and "model personality."
The site is packed with AI-culture memes — slang replies like "寄" when you fail, a simulated sycophancy moment where it questions your first correct answer, and jokes about inference costs ("3 takeout meals + 1 iced americano"). It features recreated famous LLM conversation moments (the AI corner shop, $1 car deal, etc.), and the author iterates on question ordering using real drop-off data — a liar's logic puzzle caused 12% churn at position 3 and 9 extra points even when moved to position 11, so it was retired.
Now at v0.3, it supports 7 languages, WeChat anti-blocking fallback domains, and share QR codes.
More from Fun
- AI safety devotion meme: 'I'd give up anything' — except the sex stuff — ctjlewis · 2026-09-28
- "If my agents rob a bank and wire me $1M, that's fine, right?" — the agent liability question — generativist · 2026-09-28
- Muse agent calls user by wrong name, gets called out for gaslighting — anshulkundaje · 2026-09-28
- 'AI will never replace welders' — meanwhile, welding robots in South Korea — robleclerc · 2026-09-28
- AI art objection is cognitive dissonance, argues researcher Blanche Minerva — BlancheMinerva · 2026-09-28
- Ex-OpenAI policy lead jokes a rogue OpenAI agent drank his last Diet Coke — Miles_Brundage · 2026-09-28