Seems like the author is surprised the AI can be mean but not surprised it can be nice. All responses still align with the fact that it was trained from human responses and interactions esp on Reddit.