1 paper · 1 filter
Yihang Chen, Pin Qian, Su Wang +4
Retrieval-Augmented Generation (RAG) is usually evaluated by whether the final answer is correct. Under knowledge conflict, this hides a key question: did the model follow retrieve…