Silver's pet peeve is about how people talk about models that fit historical results well. I'm noting that people simply don't even talk about models that don't fit historical results well. His comment "prediction is easy if you know the result" is a reflection on this.
maybe he's only annoyed by word choice and really wants people to say "fitting" or something for historical results and reserve "predicting" for future events. I think the fact that he makes this distinction at all is an artifact of how people talk about statistical models.