How to come up with verbal probabilities — LessWrong