A new study from Google researchers introduces "sufficient context," a novel perspective for understanding and improving retrieval augmented generation (RAG) systems in large language models (LLMs).