Open-Source STORM Evolves: AI Agents Engage in Roundtable Discussions Like Humans
Discover STORM, Stanford's powerful tool for writing Wikipedia-like articles using AI. Generate in-depth content in minutes and explore collaborative features with Co-STORM!
Not only is there a host, but you can also receive answers from different AI experts.
In April of this year, Stanford University launched a powerful tool that utilizes large language models (LLMs) to assist in writing Wikipedia-like articles.
This tool is called STORM, and it can convert your input topic into a long article or research paper in about three minutes, with the option to download it directly as a PDF.
Specifically, with the assistance of LLMs, STORM generates a writing outline by retrieving information, asking questions from multiple angles, and simulating expert dialogues, ultimately producing a detailed, in-depth, and accurate content report.
STORM excels particularly in writing tasks that require extensive research and citations. Remarkably, users can experience STORM for free directly on its website.
Since then, STORM has continuously introduced new features and services, garnering over 12k stars on GitHub.
Recently, the team launched a new feature — Co-STORM.
The difference from STORM is that it introduces a collaborative dialogue mechanism and employs round management strategies to achieve smooth collaborative AI academic research. The features include:
Co-STORM LLM Expert: This type of agent generates answers based on external knowledge sources and can pose follow-up questions based on dialogue history.
Moderator: This agent generates thought-provoking questions based on information discovered by the retriever but not directly used in previous rounds. Of course, question generation can also be based on facts.
Human User: Human users actively observe the dialogue to gain deeper insights into the topic, or guide the focus of discussion by injecting dialogue, and actively participating in the conversation.
The interface of Co-STORM looks like this.