Google’s New AI GenCast predict the weather with 99.7% accuracy ##google##ai

Google recently published a paper in Nature, where they announced a new AI weather prediction model called GenCast, which is an absolute breakthrough. When tasked to create a 15-day forecast against traditional models, GenCast was 97.2% more accurate. And that's not even the crazy part, because whereas traditional models took days to come up with their prediction, GenCast did it in only 8 minutes. Traditional weather models use what's known as numerical weather forecasting. They take the initial weather conditions, they feed them into supercomputers, and they let them run for days to create a single deterministic prediction. GenCast works fundamentally different by taking a probabilistic approach. So Google took 40 years of historic weather data with over 80 variables like temperature, humidity, and then they fed it into a diffusion model. GenCast was not only able to beat traditional numeric weather prediction models, but it actually beat them 99.8% in 36 hours and beyond.