AI beats experts on GDPval 83% of the time — so why do teams still matter?
danshipper · x · 2026-11-11
Every's Dan Shipper argues teams still matter as AI scales. On OpenAI's GDPval, AI beat human experts 12.4% of the time in 2024 vs 83% in 2026, and he now does 'five-people, $200k' work solo. But when Everyone at Every got the same survey and models, each person surfaced different insights. A Slack experiment showed Claude matched his style 93% of the time when generating, but only ranked his answers in its top 10% 60% of the time. Generating possibilities is cheap; knowing which matter requires perspective and responsibility. The measure of AI should shift from aggregate benchmarks to 'how good is it at my work, from my perspective.'
More from AGI Musings
- Buck Shlegeris: ~1/3 chance of superintelligence within a year, likely AI takeover — jammastergirish · 2026-11-12
- Kai-Fu Lee: Chinese labs open-source to survive DeepSeek competition, not government directive — Scobleizer · 2026-11-12
- Skydance's Ellison fires James Gunn, earnings call reveals AI plan to reshape Hollywood — adariostrange · 2026-11-12
- Creator says AI video editing is already good enough to stop hiring human editors — eptwts · 2026-11-12
- Prediction: hospitals will keep their own tested open-model formularies in two years — MaziyarPanahi · 2026-11-12
- Packy revives Qwest CEO's 1998 'no bandwidth glut' quote for the AI oversupply debate — prestonpesek · 2026-11-12