Toward Faithful Retrieval-Augmented Generation with Sparse Autoencoders
Paper • 2512.08892 • Published • 1
RAGLens
A lightweight hallucination detector that uses sparse autoencoders to identify, explain, and mitigate unfaithful RAG outputs
A lightweight hallucination detector that uses sparse autoencoders to identify, explain, and mitigate unfaithful RAG outputs