Can Compressing Retrieved Documents Boost Language Model Performance? This AI Paper Introduces RECOMP: Improving Retrieval-Augmented LMs with Compression and Selective Augmentation