Researcher at Unknown
Work cited showing most jobs consist of multiple tasks where AI can only match human performance on some
How media typically covers Brynjolfsson
Research or work cited
OpenAI's GDPval benchmark shows Claude Opus 4.1 achieves expert-level performance on 47.6% of real-world professional tasks across 44 different professions, while performance varies significantly across models with GPT-4o performing worst at only 10%.
“Work cited showing most jobs consist of multiple tasks where AI can only match human performance on some”