Why Google’s smarter weather might be the most useful AI in your life (www.techradar.com)

Google announced WeatherNext 2, an AI-powered forecasting model from Google DeepMind and Google Research that generates higher-resolution, more accurate, probabilistic weather forecasts by simulating hundreds of coherent future scenarios in under a minute. Built on a Functional Generative Network (FGN), the model is trained on standalone variables—temperature, wind, humidity—and learns how those interact to form “joints” like storm fronts or heat waves. Google says WeatherNext 2 outperforms its prior models across 99.9% of variables up to 15 days out, runs up to eight times faster than traditional physics-based systems, and is being rolled into Google Search, Gemini, Pixel Weather, Maps, and the Maps Platform Weather API. The shift from single-point predictions to rapid ensembles matters because it surfaces uncertainty as actionable information: users and services can see multiple plausible outcomes and timing, not just a “40% chance of rain” label. That enables finer-grained personal planning, more reliable grid and renewable output forecasts, and better-targeted emergency responses. Faster runtimes also allow more frequent, higher-resolution updates. While not a cure for climate-driven extremes, WeatherNext 2 positions AI-driven probabilistic forecasting as essential infrastructure for decision-making across everyday life and critical systems.