Banerjee et al.
RAG's retrieval, not reading, is the bottleneck for long-context question answering
By requiring line-level citations, the authors separate retrieval failures from reading failures across thirteen large language models.
Top University
Debopriyo Banerjee · Owais Aijaz · Ahmer Tabassum · Zhuohan Xie · Preslav Nakov · Veselin Stoyanov
MBZUAI · Inception42
Research Digest··2 min read
The authors propose evidence-based generation (EBG), a task in which models must return an answer with line-level citations, and a joint Ans+Evid metric that scores correctness and grounding simultaneously.
§