Google DeepMind unveils Genie 4, a world model with sound and more realistic people
Prepared by Sintelligent with the help of SI. Published in SI: . Updated: .
A neural network creates playable worlds with sound and people on the fly, and such worlds could become a training ground for SI and robots.
Genie 4 builds an interactive world from a text description frame by frame in real time, without pre-rendered scenes. Compared with Genie 3 it adds sound (ambient noise, music, speech), more plausible physics and more realistic human movement. DeepMind calls world models an important element on the path to AGI and a source of endless environments for training robots. Access goes first to vetted testers, Google AI Ultra subscribers and early-access partners of the Gemini API; others can join a waitlist.
Sources 2