Personalized Weather Services: Using Large Language Models to Provide Insightful and Actionable Weather Summaries

Recent developments in large language models and AI weather models have created new possibilities for communicating weather to the public. Google is at a unique intersection of these technologies, with an opportunity to reach a global audience through products like Google Search and Maps.

This talk presents studies and a prototype of a personalized AI weather service powered by a large language model, specifically Google's Gemini. This talk will summarize the user experience research that informed the prototype, shaping an experience that prioritizes concise and actionable summaries that help people be prepared for the day ahead. This prototype can generate personalized and context-aware weather summaries that go beyond the standard user experience of a weather app. This prototype uses the multimodal capabilities and large context window to interpret multiple sources of information, including raw forecast data, to surface the most important forecast details. Probabilistic outcomes, regional analysis, and severe events are made understandable and accessible to different audiences. . This approach can also scale effective weather communications to parts of the world where meteorology is less resourced.