Shortcuts With Chained Probabilities — LessWrong