New 'Humanity's Last Exam' AI Benchmark Aims to Measure Progress Toward Human-Level Intelligence

Researchers unveil an extremely difficult AI exam. Top models still score below 50%, far behind human experts, in this new AGI benchmark.