Firehose

This paper proposes a way to align language models with human values by learning to prioritize different moral values in different situations, and shows that this can lead to more accurate and fair decision-making. Practitioners might care because they want to use language models in applications where values and morality are important, such as in healthcare, law, or education.