Delving into the Reversal Curse: How Far Can Large Language Models Generalize?