Do AI-R&D benchmarks measure research judgment or merely scoreable engineering?
Posts here