Reproduce “RAPID: Efficient Retrieval-Augmented Long Text Generation with Writing Planning and Information Discovery” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Hongchao Gu, Dexun Li, Kuicai Dong, Hao Zhang, Hang Lv, Hao Wang, Defu Lian, Yong Liu and 1 more

RAPID introduces an efficient framework for generating comprehensive, factual, and coherent long texts, leveraging human-like writing planning and parallelized information discovery. The framework significantly improves factual accuracy, information diversity, and article coherence while reducing API calls and generation time compared to existing retrieval-augmented and multi-agent systems.

Key insights

  • Mimicking human writing processes, which involve structured planning and information discovery before content generation, substantially improves the factual accuracy and long-range coherence of LLM-generated long texts.
  • Parallelized and attribute-constrained information retrieval significantly enhances efficiency and reduces latency compared to sequential or multi-agent dialogue-based information gathering methods.
  • The introduction of a 'writing plan' as a topological graph of section dependencies is critical for maintaining logical flow and preventing content deviation across complex, lengthy articles.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.