I think the real problem is that AIs have super human performance on one off assessments like exams, but fall over when given longer term open ended tasks.

This is why we need to continue to educate humans for now and assess their knowledge without use of AI tools.