Survey identifies benefits and challenges of retrieval-augmented generation in natural language processing tasks, suggesting future improvements.
Key Points
Retrieval-augmented generation addresses hallucination problems and knowledge updates in large language models, enhancing their effectiveness.
Key techniques in retrieval-augmented generation, especially retriever and retrieval fusions, are systematically reviewed in this paper.
Assessment focuses on RAG training methods with and without datastore updates, offering practical implementation codes for researchers and developers alike. This survey highlights the necessity for ongoing improvements in retrieval-augmented generation to advance its capabilities for various applications.