I wonder if this will hold as other models with different biases achieve parity.