It's hard to train deep learning algorithms when most of the positive feedback they get is sarcastic.