Statistical models & the irrelevance of rare exceptions — LessWrong