This comment is a perfect encapsulation of AI hype.
Recitation is not competence. It doesn't matter if Bobby can say "4" when I say "What is 2 + 2?". What matters is if Bobby can add.
If you try to actually create tests that way, you will discover that the institutions and political atmosphere around you will punish you harshly. It's "unfair" to give a test that you can't pass by memorizing things, you see. And the people who will ultimately make the decisions about what you're allowed to do with your test are rarely going to care much about actual mastery, and often wouldn't be able to recognize mastery if they did care.
In practice, standardized tests are always going to be easily gamed, so if you make people's and institutions' rewards dependent on them, you will end up diverting more time, energy, and other resources away from actually teaching mastery and into meaningless gaming.