quantamagazine.org/he-wants-ai-benchmarking-to-be-a-science-not-a-dark-art-20261016 He Wants AI Benchmarking To Be a Science, Not a Dark Art
A computer scientist who calls benchmarks "the mistake that made machine learning" argues the whole system has been broken since 1959, kept alive mostly by luck. He explains how a 1987 speech-recognition test helped trigger a 25-year AI winter, and what might fix benchmarking now.
— via Quanta Magazine, John Pavlus
