The goal should be learning, not necessarily performance on a timed and written test. Grades really shouldn't be competitive either. We need to establish mechanisms of evaluation that determine a student's mastery of a topic without high pressure, high stakes tests. Class discussions, (hand) written essays, individual and group projects. Basically, AI is a reality now, just like we don't make calculus students do long division, autoregressive generative text is a tool you can't prevent. So, ask the student interesting questions the AI would be incapable of answering correctly. And, perhaps that's the real problem. If the AI is more knowledgeable, discerning and insightful than the instructor such that the instructor can't ask questions that challenge it, then why are they the instructor?