There are algorithms to add variables and observations to an estimated linear regression model, so scaling is not an advantage of gradient descent. You should mention in the post that there are better ways to estimate this model, even if you're just presenting it as an example.
