Benchmark AI by high‑value domains

Frontier AI progress should be measured by performance in the specific economic and social domains where agentic models will actually be used (education, coding, drug discovery, logistics), not by headline GDP share or broad occupational taxonomies that dilute concentrated impacts. A domain‑focused benchmark (as advocated by Dylan Patel’s GDPx idea) would let researchers, investors and policymakers track meaningful capability changes and target regulation and investment where frontier models matter most. — Shifting from aggregate metrics to domain‑specific benchmarks changes how policymakers assess risk, design regulation, allocate infrastructure, and evaluate bets about AI‑driven growth.