[usef-]

Distillation of this kind isn't the only data that is fed into models, it's still a small minority. It might be weighed higher in training to learn the thinking patterns/etc but you need a lot more tokens.