ML Interview Q Series: Why might Mean Absolute Error (MAE) be harder to optimize than Mean Squared Error (MSE) in high-dimensional feature spaces, and how could you mitigate those difficulties?