14 AI models tested on 8,000 startups: none beats simply ranking by hiring growth
HeyAmit_ · x · 2026-10-31
SAB-Bench findings: 14 leading AI models scored 97/100 on 960 real analyst tasks, but across 101 rounds of picking winners among 8,000 startups, not one beat the simple heuristic of ranking companies by hiring growth. AI is a capable analyst yet no better than random at predicting which startups will grow — a caution for funds plugging AI into their pipelines.
Related event: AI Aces Analyst Tasks but Fails at Picking Winners, Benchmark Shows(2 posts)→
More from AGI Musings
- "AI shouldn't turn everyone into programmers — it should boost existing jobs" — AryHHAry · 2026-10-31
- AI skeptic predictions are correlated bets that never cost anything, argues 79K-follower researcher — teortaxesTex · 2026-10-31
- teortaxesTex pushes back on 'LLMs can't disassemble engines' cope in embodied AI debate — teortaxesTex · 2026-10-31
- Pedro Domingos: Probability of Zero Fraud in the AI Boom Is Zero — pmddomingos · 2026-10-31
- tszzl: AI debunking won't change academia's thermostat dynamics on star researchers — tszzl · 2026-10-31
- Sound is representationally richer than vision, and computers hear in the dark ages — JosephJacks_ · 2026-10-31