NYC Gazette

Researchers at New York University's Courant Institute of Mathematical Sciences have developed a tool named DietNerd, aimed at providing evidence-based insights into diet and nutrition. The tool utilizes a large language model to answer questions with information sourced from peer-reviewed research.

Dennis Shasha, a professor at the Courant Institute who contributed to the creation of DietNerd, explained the motivation behind the project: “Everyone is concerned about diet, but there’s a lot of bad advice about diet. We wanted to build a system that’s based on scientific literature and that presents the results in a simple way.”

DietNerd operates by searching peer-reviewed articles in PubMed in response to user inquiries about dietary strategies, nutrition science, and diet-related health outcomes. It provides summaries of findings along with evaluations for statistical power and bias, as well as links to the original articles.

The creators have emphasized efforts to ensure the accuracy of DietNerd's outputs. Shela Wu, a graduate student involved in developing the tool, stated: “We have made a significant effort to eliminate ‘hallucinations’—the misleading or false search results that often plague AI.”

The tool also addresses complexities within diet-based information by continuously updating its database to reflect ongoing studies. Wu elaborated on this aspect: “This tool utilizes an innovative article search strategy that identifies areas of debate and pulls in relevant articles that either represent these varying viewpoints or that introduce a safety or risk consideration.” She added that once generated, "the output presents users with a comprehensive response that includes a safety analysis evaluating the pros, cons, and risks of the topics in question."

Wu concluded by highlighting their mission: “Our mission is to enrich conversations between patients and their medical providers and equip users with valuable insights into diet and nutrition.”