Negating Negatives: Alignment without Human Positive Samples via Distributional Dispreference Optimization

Open in new window