Are humans still smarter than machines?

AI now surpasses human experts on major reasoning benchmarks (GPQA Diamond: 94% vs 65% PhD baseline; MMLU-Pro: saturated at ~90%). Autonomous agents handle complex multi-step workflows in law, medicine, and engineering. However, reliable operation without human oversight remains elusive — agentic systems still fail unpredictably on novel tasks, and legal/ethical accountability frameworks require human decision-makers. The margin is razor-thin.