Yeah, that's a good point. On the other hand, the model is out there, irrevocably so, and you can use knowledge distillation techniques to train other weaker models. So yeah, we're not fully independent, but the situation is significantly improved.