AI World Simulators (OpenAI Sora)
Generative AI learns the physics of the real world.
Explore this event on the interactive timeline →Before 2024, AI video generation was a chaotic, hallucination-filled mess. OpenAI's Sora changed the paradigm by treating video generation not just as pixel painting, but as a simulated physics engine.
Key Numbers
- Architecture
- Diffusion Transformer
- Emergent Property
- Implicit 3D Physics
Verified Facts
- Sora utilizes a diffusion transformer architecture. By training on massive amounts of video data at scale, the model implicitly learned how 3D geometry, lighting, and object permanence work in the real world.
- It proved that scaling up compute for generative models eventually results in emergent physics simulation, allowing the AI to generate highly realistic, complex scenes with multiple characters and accurate camera motion.
- In late 2025, OpenAI released Sora 2 as a consumer application in North America, firmly establishing AI-generated reality as a mainstream medium.