GenCast uses machine learning and decades worth of reanalysis data for its predictions, and is more accurate than the ECMWF system 97.2% of the time. It will be made open-source on GitHub and integrated with Google Earth. Genie 2, an evolution of DeepMind's Generative Interactive Environments tool, allows users to create 3D game worlds using a simple text prompt or a prompt image. However, the model currently suffers from stability issues and DeepMind has not disclosed its plans for its release.
Key takeaways:
- Google DeepMind has released two powerful AI models, GenCast and Genie 2. GenCast is focused on weather forecasting and outperforms the most accurate system currently in use, while Genie 2 is designed to create elaborate, visually striking virtual worlds for video games.
- GenCast can generate 15-day weather forecasts in just eight minutes, compared to hours for the supercomputers running the current most accurate system, and was more accurate 97.2% of the time in tests.
- Genie 2 allows developers to create expansive and playable 3D game worlds using a simple text prompt. It can simulate a virtual world along with animations, physics, and interactions between elements.
- DeepMind plans to make GenCast open-source on GitHub and integrate it with Google Earth for public use. The future of Genie 2 is less clear, with no release date announced and ongoing issues with stability.