News
DeepMind Unveils GenCast and Genie 2: Two Groundbreaking AI Models
- By John K. Waters
- 12/06/2024
Google’s DeepMind AI research organization has introduced two advanced large language models designed to tackle vastly different challenges: precision weather forecasting and immersive video game world creation. The announcements, made today, underline the diverse potential of AI in both practical and creative applications.
Announced in a paper published in Nature, GenCast was designed to predict weather and the risks of extreme conditions with state-of-the-art accuracy. DeepMind researchers say it delivers faster and more accurate forecasts up to 15 days ahead.
Leveraging nearly 40 years of high-quality weather data, the GenCast model surpasses existing forecasting systems, including the industry-standard European Center for Medium-Range Weather Forecasts (ECMWF) model. DeepMind claims that GenCast produces 15-day forecasts with higher accuracy 97.2% of the time, achieving this in just eight minutes on a single processing unit—compared to hours on ECMWF’s supercomputer.
Genie 2 is a "foundation world" model the researchers say pushes the boundaries of virtual world generation for video games, enabling developers to create complex, interactive 3D environments from simple text prompts. Building on its predecessor, Genie 1, which was limited to 2D worlds, Genie 2 incorporates animations, physics, and dynamic interactions, paving the way for AI-generated gaming experiences. Genie 2 isn’t a traditional game engine but a diffusion model that generates visuals as players—either humans or AI agents—navigate the simulated environment.
The model creates frames dynamically, allowing it to infer elements of the environment, such as water, smoke, and physics effects. While some interactions can appear rudimentary, Genie 2 supports diverse perspectives, including first-person, third-person, and isometric views.
To generate a virtual world, Genie 2 requires just one image prompt, which can be sourced from Google’s Imagen 3 model or a photograph from the real world. This approach enables a wide range of creative possibilities for interactive experiences.
Although still in development, the model demonstrates significant potential for transforming game design and AI training.
About the Author
John K. Waters is the editor in chief of a number of Converge360.com sites, with a focus on high-end development, AI and future tech. He's been writing about cutting-edge technologies and culture of Silicon Valley for more than two decades, and he's written more than a dozen books. He also co-scripted the documentary film Silicon Valley: A 100 Year Renaissance, which aired on PBS. He can be reached at jwaters@converge360.com.