
It’s cheap to copy already built models from their outputs, but likely still expensive to train new models that push the boundaries.
It’s cheap to copy already built models from their outputs, but likely still expensive to train new models that push the boundaries.