It probably shouldn't continue to surprise me how often people's "AI benchmarks" exclude a significant fraction of actual, living, humans from being "human-grade".