Comment by nextaccountic

Ok but, if their moat is in cost effective inference, why are they training models?

If anything it looks like the first place is training their competition, through distillation, while not capturing much value in return