Ofir Press calls for new benchmark design as human baseline era ends

Ofir Press of AI2 and Princeton argues that many human tasks are now trivially solved by AI, so benchmarks can no longer be derived from them directly. He proposes three principles for a new era of benchmark design, including tasks that would take human teams years to complete.

2026-10-13 ~ 2026-10-13 · 2 related posts

• Ofir Press: the human baseline era is over — new guidelines for building benchmarks — OfirPress · 2026-10-13
• Ofir Press lays out 3 rules for post-human-level benchmarks: tasks humans need years or decades to solve — OfirPress · 2026-10-13