Why World Models Could Change Robotics, 3D, and Creativity

The gist

1. Atlas can create bullet time video effects with just 3 smartphone cameras instead of hundreds of studio cameras, cutting production complexity by 50 to 100 times.
2. The model takes 3 to 25 sparse images and fills in missing views through generation, removing the need for exhaustive manual capture that previously took hours per room.
3. Robotics teams can use Atlas to convert real environments into simulations from sparse images, then generate synthetic variations to train robot policies without extensive manual scene setup.
4. The co-founders argue next-view prediction can frame any spatial reasoning task the way next-token prediction frames language tasks in large language models.