Weather influences our daily lives and decisions, becoming even more critical with climate change causing more extreme weather events. Perfect forecasts are impossible, so scientists use probabilistic ensemble forecasts to predict a range of possible weather scenarios.
In a paper published in Nature, Google DeepMind has introduced GenCast, a new high-performance artificial intelligence (AI) model that outperforms the European Centre for Medium-Range Weather Forecasts’ (ECMWF) system up to 15 days ahead.
Unlike its predecessor, which gave one best-guess forecast, GenCast offers multiple predictions, each showing a possible weather outcome.
GenCast uses a diffusion model, similar to those used in generative AI for creative outputs like images, but it's tailored for Earth's spherical shape and weather data.
It was trained on 40 years of historical weather data from ECMWF’s ERA5 archive.
When tested against data from 2019, GenCast proved superior to ECMWF’s own weather forecasts in 97.2% of the 1320 forecast scenarios, especially for forecasts beyond 36 hours.
Google DeepMind intends to make public the model's code, weights, and forecasts to aid the weather forecasting community.
Next: Socratic learning
In other company news, Google DeepMind researcher Tom Schaul has published a paper titled "Boundless Socratic Learning With Language Games" on arXiv. Schaul argues that an AI with a recursive self-improvement method, dubbed Socratic learning, "can boost performance vastly beyond what is present in its initial data or knowledge."
This shows "how to extend learning and adaptability beyond the initial training phase," notes a review published in HackerNoon. "DeepMind outlines a future where AI models can generate their own data, design their own tasks, and evaluate their performance without external input... the introduction of this framework represents a step toward the long-term goal of open-ended intelligence, where AI is not just a tool but a partner in discovery."
"Boundless Socratic Learning just unlocked AI's ability to self-improve forever," says Peter Diamandis. "This is history in real time."