Google AI can create 3D worlds from text prompts
Google DeepMind just introduced Genie 3, an AI that builds detailed 3D worlds from simple text prompts—think ski slopes or warehouses you can actually explore.
It's designed to help train robots and self-driving cars by letting them practice in these virtual spaces before trying things out in real life.
Genie 3 can interact with the world
Genie 3 stands out because it lets AI agents interact with these worlds and even remembers changes—like if you add a herd of deer to a snowy scene, they'll stick around.
This makes training way more realistic and flexible than before, even though the tech isn't public yet.
Genie 3 could help language AIs understand the real world
If world models like Genie 3 catch on, we could see smarter robots and safer autonomous vehicles sooner.
Plus, experts think it might help language AIs get better at understanding the real world by giving them a sense of "physical" context during training.