Self-RAG (Figure right) is a new framework to train an arbitrary LM to learn to retrieve, generate, and critique to enhance the factuality and quality of generations, without hurting the versatility ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results