Combining document retrieval and large language models improves open-domain question answering accuracy.
