AI Video Is Becoming a World Simulator

AI video models are beginning to simulate the physical world without calculating every interaction like a traditional game engine. By predicting the next frame from internet scale video data (think YouTube!) these models can learn an implicit approximation of how objects move, collide, deform, and persist through time. That does not mean they have mastered physics yet -- you'll still encounter oddities like swapped limbs, and disappearing objects. But when next-frame prediction runs in real time, something bigger starts to emerge. Google DeepMind’s Genie 3 and a myriad of other real-time video models like Odyssey ML and Lingbot points toward generated worlds with memory, user control, promptable events, and embodied agents. This video breaks down the architecture behind that shift, why autoregressive world models differ from diffusion video, and how the same technology could become a simulation layer for media, robotics, and physical AI. Chapters: 0:00 – AI is learning the physical world 0:30 – Physics without an engine 1:54 – When video runs in real time 2:44 – Two ways to generate reality 4:06 – The memory problem 5:05 – Steering the simulation 6:07 – From world models to agents 7:39 – The road to physical AI Related: - Google Just Turned Street View Into a Video Game: https://www.youtube.com/watch?v=bxv4IkobUPI - I Tested Google’s Genie 3… AI Game Engines Are HERE!: https://www.youtube.com/watch?v=lALGud1Ynhc References: - Google DeepMind — Genie 3: https://deepmind.google/models/genie/ - Google DeepMind — Genie 3 technical overview and limitations: https://deepmind.google/blog/genie-3-a-new-frontier-for-world-models/ - Google DeepMind — SIMA, the Scalable Instructable Multiworld Agent: https://deepmind.google/blog/sima-generalist-ai-agent-for-3d-virtual-environments/ - Apple — ARKit: https://developer.apple.com/documentation/arkit - Google — ARCore: https://developers.google.com/ar -- Subscribe for more in-depth AI & creative tech videos! 👉 ‪@bilawalsidhu‬ Join My Newsletter: https://spatialintelligence.ai Connect with me on X/Twitter here: https://x.com/bilawalsidhu Everywhere else here: https://bilawal.ai Business inquiries: [email protected] Bio: Bilawal Sidhu is a creator, engineer, and product builder obsessed with blending reality and imagination using art and science. Bilawal is the technology curator for TED Talks, and a venture scout for Andreessen Horowitz. With more than a decade of experience in the tech industry, he spent six years as a product manager at Google, where he worked on spatial computing and 3D maps. His work has been featured in major publications including Bloomberg, Forbes, BBC, CNBC, and Fortune, among others. Bilawal’s journey into computer graphics began at 11, when he fell in love with seamlessly blending 3D into real-life footage. Since then, he's captivated over 1.5M subscribers, garnering more than 500M+ views across his platforms. Driven by a mission to empower the next generation of artists and entrepreneurs, Bilawal openly shares AI-assisted workflows and industry insights on social media. When he’s not working, you can find Bilawal expanding his collection of electric guitars. TED: https://www.ted.com/speakers/bilawal_sidhu