The evolution of artificial intelligence has led us, among others, to the ability to create content from basic instructions. So many stories as articles similar to those published on Wikipedia. But the logical question is whether they are reliable. Until now most, if not all, lacked this ability. Now Stanford University has created STORM (Structured Task-Oriented Research Machine), a large language model (LLM) system that exceeds the qualities of well-known chatbots.
Basically, STORM employs a multi-agent system that simulates a team of experts collaborating on a research project. This is not an ordinary LLM, but rather a “team” of AI agentseach of which plays a crucial role in the content creation process.
In the first phase, STORM begins searching the digital archives of Wikipedia and other reliable sources. Its creators describe it as a fleet of highly efficient librarians, each tasked with collecting relevant information on the topic at hand. But STORM doesn’t just copy and paste: it analyzes, extracts and synthesizes this information into a coherent structure.
With this “library” of information, STORM prepare a detailed outline that serves as a skeleton for the final article. This is not a simple list, but a compendium that guarantees comprehensive coverage of the topic.
This is where things get really interesting. STORM doesn’t just write: it talks to itself. Or rather, it simulates conversations between multiple AI agents, each representing different perspectives on the topic. It’s like eavesdropping on a roundtable discussion of experts, except they’re all artificial intelligence.
And here’s the key to STORM: this roundtable gets you questioning your own assumptions, exploring different angles of the topic, and improving the scheme. The result is a less schematic and more complete text that benefits from multiple points of view, although these are not human.
But it doesn’t all end here. The next step is the final writing and for this STORM goes over each paragraph to ensure that there are no contradictions or inconsistencies while maintaining academic rigor. And in this section it would seem that those responsible have emphasized the appropriate citation. Every statement and every figure or data that is mentioned in an article generated by STORM is supported by a referenceallowing readers to verify the information for themselves. And this has been evaluated to reach an accuracy rate close to 90%.
Undoubtedly, it is a tool with the ability to revolutionize our way of relating to artificial intelligence and to obtain reliable content… as long as we trust its developers and Wikipedia.