In the Rao-Blackwell Theorem, if $\hat{\theta}$ is an unbiased estimator of $\theta$ and $T$ is a sufficient statistic, how does the Mean Squared Error (MSE) of the improved estimator $\tilde{\theta} = E[\hat{\theta}|T]$ compare to the original estimator?