Comment by generalizations

Testing models on their tokenization has always struck me as kinda odd. Like, that has nothing to do with their intelligence.