Frontier AI Is Faceplanting at Real-World Workplace Tasks

UC Berkeley researchers built a benchmark of 1,500+ tasks across 55 occupations and found frontier AI still struggles with real workplace work.

The top model passed only 24% overall and scored zero on the hardest tasks, underscoring major reliability gaps.

The results challenge claims that current AI can replace many jobs soon, even as routine tasks may still face disruption.

The study, from the Center for Responsible, Decentralized Intelligence and related researchers, adds a reality check to job-readiness hype.