I never understood this criticism. Any good data set contains copyrighted data so any "open source" model will be crappy, most of the price is in training these

I never understood this criticism. Any good data set contains copyrighted data so any "open source" model will be crappy, most of the price is in training these models (if you have the compute, why would an "open source" model be helpful anyways), and you can already modify a model from just it's weights. Why should we want an "open source" model. Just give me the weights.