Researcher pushes back on Jensen Huang's AGI claim: benchmark scores aren't general intelligence

ValerioCapraro · x · 2026-09-07

Behavioral scientist Valerio Capraro disputes Jensen Huang's claim that AGI has been achieved. He concedes GPT-6 performs impressively on benchmarks including ARC-AGI-3 and can solve math problems no human can.

His core argument: benchmark performance is not general intelligence. True general intelligence isn't just crushing closed-ended problems with fixed goals—it also means solving simple but unfamiliar problems every ten-year-old can handle, and navigating an open-ended world. His full essay elaborates, and a new paper on adaptive intelligence is due in about a month.

Original post →

More from AGI Musings

AGI Musings channel →