1. With differential privacy, the analyst needs to learn some useful information from the released statistic without learning too much information. In contrast, when sending a message, the recipient of a message should be able to learn the exact original message (correctness), but the adversary should learn nothing about the message (semantic security). 

2. Differential privacy is defined in terms of neighboring databases that differ by a single row in the database (privacy is at a “row level”). However, Alice may have clicked on many ads for running shoes. Epsilon differential privacy hides that she clicked on a particular ad, but might not adequately hide her general interest in running shoe ads.