Stanford University has once again made significant advancements in the field of artificial intelligence. Their latest development, the STORM&Co-STORM system, is now open-source. This system is capable of comprehensively integrating multi-source information by inputting simple themes to generate high-quality, lengthy articles. This innovation not only avoids information blind spots but also significantly enhances the efficiency and quality of scientific writing.
The core technology of the STORM&Co-STORM system includes support from Bing search and GPT-4o mini. The STORM component generates outlines, paragraphs, and articles iteratively through multi-angle Q&A between "LLM Experts" and "LLM Moderators." Co-STORM, on the other hand, generates interactive dynamic mind maps through dialogue among multiple intelligent agents, ensuring that no information needs that the user may have overlooked are missed.
Users only need to input English keywords to generate high-quality long-form articles integrating multi-source information, akin to Wikipedia articles. By experiencing the STORM system, users can freely choose between STORM and Co-STORM modes. Given a topic, STORM can form a structured high-quality long-form article within 3 minutes.
Additionally, users can click "See BrainSTORMing Process" to view the brainstorming process of different LLM roles. In the "Discover" section, users can refer to articles and chat examples generated by other scholars, while their own generated articles and chat records can be found in the sidebar "My Library."
The automated writing process of the STORM system is divided into three stages: multi-perspective question generation, outline generation and refinement, and full text generation. The system reviews relevant Wikipedia articles to determine various perspectives covering the theme and then simulates a dialogue between a Wikipedia contributor and an expert based on reliable online sources. Based on the LLM's inherent knowledge, dialogue content collected from different perspectives is ultimately carefully arranged into a writing outline.
Despite STORM identifying different perspectives when researching a given topic, the collected information may still be biased towards mainstream sources on the internet and may contain promotional content. Another limitation of the study is that although researchers focused on generating text similar to Wikipedia articles from scratch, they only considered generating free-form text. High-quality Wikipedia articles typically include structured data and multi-modal information.
Co-STORM aims to improve the issue of information omission in information collection and integration, thereby greatly promoting learning efficiency. It helps users understand and participate in information organization through multi-agent collaboration dialogue, dynamic mind maps, and report generation modules. Researchers conducted a human evaluation of 20 volunteers, comparing the performance of Co-STORM with traditional search engines and RAG Chatbot. The results show that Co-STORM significantly improves the depth and breadth of information, with 70% of users preferring Co-STORM, as it significantly reduces cognitive burden.
Currently, the STORM&Co-STORM system only supports English interaction, and it may be expanded to multilingual interaction capabilities in the future. The open-source nature of this system signifies that we are living in a remarkable era where the way we access information can be fully customized to individual levels, making it possible to learn anything.
暂无评论