Hit-RAG: Learning to Reason with Long Contexts via Preference Alignment

ArXi:2603.07023v1 Announce Type: new Despite the promise of Retrieval-Augmented Generation in grounding Multimodal Large Language Models with external knowledge, the transition to extensive contexts often leads to significant attention dilution and reasoning hallucinations. The surge in information density causes critical evidence to be submerged by voluminous noise, which complicates the discernment of relevant fragments within a dense input.