On the contrary, small efficient models using recent improvements should be more sustainable than huge frontier models that require billions and years to train