But otoh if they beat the test, is that a bad thing? It means the test is weaker, but the models have picked up some capability, even if it's just "saying what the test wants to hear"
If the test measures something good, having the models fake it till they make it has some (maybe a little) value