DeepMind Unveils Genie 4 World Model With Sharper Physics and Simulated Humans

Google DeepMind has revealed Genie 4, an upgraded world model that improves on Genie 3 in nearly every respect, according to Ars Technica. The original Genie 3 announcement last year sent gaming stocks falling, and world models have since become a major focus for AI labs globally.

DeepMind frames the technology as a step toward artificial general intelligence rather than as a game engine. A world model builds a virtual world without pre-rendered environments or hand-programmed interactive elements; users describe what they want, and the experience is generated frame by frame in real time as they explore.

Genie 4 draws on Google's updated Gemini models, using Gemini 3.8 Flash as a guiding agent that breaks down goals and evaluates each output frame to keep simulations on track across changing environments. Users can still alter parameters mid-run, such as changing the weather, adding elements, or moving the camera. DeepMind notes that Genie 4 worlds remain consistent for only a couple of minutes before losing track of how things began.

According to Google, the technology could provide infinite training environments for physical robots or enable rapid prototyping in creative fields. Access may also be possible for those willing to pay Google enough, though no pricing details were given.

Source: Ars Technica