A new AI benchmark reveals that top models score under 1% while humans hit 100%, raising serious questions about whether AGI is actually within reach.