The Genie 3-powered web app lets US subscribers create, explore and remix interactive worlds from text and images, with generations currently capped at 60 seconds.
Google DeepMind CEO Demis Hassabis told the World Economic Forum that scaling LLMs alone will not produce scientific breakthroughs, put AGI by 2030 at roughly 50 percent, and said China trails by about six months.
Researchers report that small fine-tuned models such as Qwen2.5-7B predicted environment states with about 99 percent accuracy in structured text-based tasks.
Speaking at the Axios AI+ Summit, the Google DeepMind chief predicted that models uniting text, images, video, audio, 3D and robotics will define the next year of AI.
World models like Google's Genie 3 have given new life to the idea that reality might be simulated, while a 2025 UBC study argues the universe cannot be computed.
