
Limitless: An AI Podcast World Models: What are They, and is Google's Genie 3 AGI?
13 snips
Feb 10, 2026 A deep dive into Google’s Genie 3 world model and how it simulates reality in real time. They compare frame-by-frame physical simulation to text-based models and show interactive demos. Discussions cover impacts on gaming, robotics training, autonomous driving simulations, and the challenges of fidelity and compute costs.
AI Snips
Chapters
Transcript
Episode notes
Real-Time Simulation Beats Predictive Text
- Genie 3 simulates reality in real time at 720p and 24 fps, producing interactive worlds rather than fixed video.
- Waymo adopted it to create advanced driving simulators that generate rare scenarios like tornadoes or elephants on highways.
World Models Are Embodied Understanding
- World models are 'lived and experienced' AIs that understand physics and how environments change.
- They predict the next frame of reality rather than the next word, enabling dynamic interaction.
Dynamic, Persistent Virtual Worlds
- World models generate dynamic, persistent changes (e.g., paint dripping and staying) versus predetermined video.
- They can scan personal objects and spaces to create customizable, explorable simulations.
