What is the relationship between the sample size $n$ and the standard error of the mean (SEM) as $n$ increases, assuming the population standard deviation $\sigma$ is constant?