Generate rather than Retrieve: Large Language Models are Strong Context Generators

TL;DR

This paper introduces GenRead, a novel approach that replaces document retrieval with large language model generation for knowledge-intensive tasks, achieving superior results without external document retrieval.

Contribution

The paper proposes a generate-then-read framework and a clustering-based prompting method, significantly improving performance on knowledge tasks over traditional retrieve-then-read pipelines.

Findings

GenRead achieves 71.6 and 54.4 exact match scores on TriviaQA and WebQ.

GenRead outperforms the state-of-the-art retrieve-then-read pipeline DPR-FiD.

Combining retrieval and generation further enhances model performance.

Abstract

Knowledge-intensive tasks, such as open-domain question answering (QA), require access to a large amount of world or domain knowledge. A common approach for knowledge-intensive tasks is to employ a retrieve-then-read pipeline that first retrieves a handful of relevant contextual documents from an external corpus such as Wikipedia and then predicts an answer conditioned on the retrieved documents. In this paper, we present a novel perspective for solving knowledge-intensive tasks by replacing document retrievers with large language model generators. We call our method generate-then-read (GenRead), which first prompts a large language model to generate contextutal documents based on a given question, and then reads the generated documents to produce the final answer. Furthermore, we propose a novel clustering-based prompting method that selects distinct prompts, resulting in the generated documents that cover different perspectives, leading to better recall over acceptable answers. We conduct extensive experiments on three different knowledge-intensive tasks, including open-domain QA, fact checking, and dialogue system. Notably, GenRead achieves 71.6 and 54.4 exact match scores on TriviaQA and WebQ, significantly outperforming the state-of-the-art retrieve-then-read pipeline DPR-FiD by +4.0 and +3.9, without retrieving any documents from any external knowledge source. Lastly, we demonstrate the model performance can be further improved by combining retrieval and generation. Our code and generated documents can be found at https://github.com/wyu97/GenRead.