Runway Co-founder: Video Generation Is Just the Start, the Goal Is General World Simulation
Latent Space · youtube · 2026-09-25
Latent Space podcast interview with Runway co-founder Anastasis Germanidis, tracing the company's evolution from creative tools to frontier video models (Gen-1 through Gen-4.5) and laying out its bigger thesis: video generation is a path toward general-purpose world simulation.
Key points:
- The early bet on a 1,000-A100 cluster kickstarted video research; Gen-2 came from a weekend hack; the Sora release was an existential moment, after which the team scaled Gen-3 roughly 10× in a three-month push.
- Core arguments: scaling video prediction alone may be enough to learn physics; real-time generation is inevitable; learning directly from the physical world may unlock capabilities language cannot capture.
- Interface World Models: neural networks render software pixels directly with no HTML/CSS/React — the endgame being a fully neural operating system.
- Robotics: world models as robot simulators; third-person internet video could be the largest training source for robotic intelligence; World Action Models turn video models into robot policies.
- Proposes the "Lucid Dream Test" for when world models are good enough; argues orchestration eventually moves inside the model; also covers open-source world models, Chinese video models, and the global generative video race.
More from Embodied
- NYU's Steertuning Lets a Robot Learn a Chore From 5 Minutes of Human Demo — CyberRobooo · 2026-11-11
- Boston Dynamics Atlas demoed with Blender 5.3 — TomLikesRobots · 2026-11-11
- Lamport Robotics posts robot 'wake-up test' demo video — R4nsack · 2026-11-11
- Embodied AI Faces Two-Order-of-Magnitude Data Gap as Industry Heads to 10M Hours — GaryMarcus · 2026-11-11
- NVIDIA hosts CoRL 2026 meetup on scaling robot training with Isaac Lab — m_wulfmeier · 2026-11-11
- Lighthouse extends Obsidian welding foundation model to run on any robotic platform — bennash · 2026-11-11