TECH STARTUPS
Stanford University has unveiled a new AI tool called STORM, designed to help anyone create detailed, Wikipedia-style reports on any topic. Open to the public for free, STORM is a research prototype that supports interactive knowledge curation.
The tool allows users to “Get a Wikipedia-like report on your topic with AI” by simply entering a topic. It then scours hundreds of websites to produce an article summarizing key findings.
What Is STORM?
Developed by Stanford’s OVAL lab, STORM—short for Synthesis of Topic Outlines through Retrieval and Multi-perspective Question Asking—leverages large language models (LLMs) to automate the research and writing process. Its goal is to help users generate comprehensive, well-organized, and reliable content. The tool searches across hundreds of websites to compile major findings and presents them in an organized report.
How STORM Works
STORM operates in two main stages:
- Pre-writing
The tool gathers references by conducting online research and generates an outline for the topic. It incorporates diverse perspectives by simulating conversations between AI agents and using targeted questions to enrich its findings. - Writing
Once the outline and references are ready, STORM creates a full-length article, complete with citations. This ensures the content is structured and supported by credible sources.
Who Benefits from STORM?
STORM is particularly useful for students, researchers, and content creators who need detailed and well-referenced material. Its flexible and modular approach allows users to customize the tool to fit their specific requirements, including different retrieval and language models.
Open-Source Availability
STORM is open-source and hosted on GitHub, where users can access its code and documentation. A live research preview is also available through Stanford’s website, offering an opportunity to test its capabilities firsthand.
Research Insights
Stanford researchers focused on applying LLMs to produce grounded and organized long-form content with the same breadth and depth as Wikipedia articles. They tackled challenges such as researching and outlining topics before writing. STORM addresses these through:
- Discovering multiple perspectives during the research phase.
- Simulating dialogues among AI agents with varied viewpoints.
- Curating the collected information into cohesive outlines.
To evaluate STORM, the team curated a dataset called FreshWiki, featuring recent high-quality Wikipedia articles. Feedback from experienced Wikipedia editors showed that STORM-generated articles were better organized and had broader coverage compared to baseline models. However, challenges like source bias and fact misassociation were identified for future improvement.
Limitations and Caution
As a research prototype, STORM has its limitations. There is potential for it to generate offensive or inaccurate content. Users are encouraged to verify the outputs for appropriateness and reliability.
STORM showcases promising steps toward making AI-assisted writing more accessible and reliable. For those interested in trying it, the tool offers an innovative way to produce well-researched and structured content.
Connect with us on our socials: