CoAuthorAI: Human-in-the-Loop System for Scientific Book Writing
A new writing system called CoAuthorAI has been developed by researchers, integrating retrieval-augmented generation, expert-crafted hierarchical outlines, and automated reference linking to facilitate coherent and precise scientific writing in book format using LLMs. In tests involving 500 chapters from various domains, CoAuthorAI achieved an impressive soft-heading recall of 98%. Additionally, a human assessment of 100 articles revealed an 82% satisfaction rate for the produced content. The book 'AI for Rock Dynamics,' created with CoAuthorAI and Kexin Technology's LUFFA AI model, has been released by Springer Nature. These findings illustrate how systematic collaboration between humans and AI can enhance the capabilities of LLMs, transitioning from articles to comprehensive books.
Key facts
- CoAuthorAI is a human-in-the-loop writing system for scientific book writing.
- It combines retrieval-augmented generation, expert-designed hierarchical outlines, and automatic reference linking.
- Evaluated on 500 multi-domain literature review chapters, achieving 98% maximum soft-heading recall.
- Human evaluation of 100 articles showed 82% satisfaction rate.
- The book 'AI for Rock Dynamics' was generated with CoAuthorAI and Kexin Technology's LUFFA AI model.
- The book has been published by Springer Nature.
- The system addresses issues of inconsistent structure and unreliable citations in LLM-generated book-length texts.
- CoAuthorAI allows experts to iteratively refine text at the sentence level.
Entities
Institutions
- Springer Nature
- Kexin Technology