Log as generalized length — LessWrong