Learning to summarize from human feedback

Nisan Stiennon, Long Ouyang, Jeff Wu +6

As language models become more powerful, training and evaluation are increasingly bottlenecked by the data and metrics used for a particular task. For example, summarization models…