Is the Gaussian approximation to P(w | D, α) too heavy-tailed or too light-tailed, or both? It may help to consider P(w | D, α) as a function of one parameter wi and to think of the two distributions on a logarithmic scale. Discuss the conditions under which the Gaussian approximation is most accurate.