2026
Copy-Paste to Mitigate Large Language Model Hallucinations
ICLR 2026poster
While Retrieval-Augmented Generation (RAG) enables large language models (LLMs) to generate contextually grounded responses, contextual faithfulness remains challenging as LLMs may not consistently trust provided context, leading to hallucinations that undermine reliability. We observe an inverse co…