That may be true. But an even more interesting point may be that you don’t have to train a huge model ever again? Or at least not to train a new slightly improved model because now we have open weights of an excellent large model and a way to train smaller ones.