Anastasis Germanidis on Runway’s Evolution from Video Generation to World Models and Robotics
Runway co-founder Anastasis Germanidis joins Latent Space to examine the company's progression from early generative video tools to frontier world models and simulation. He discusses the engineering milestones from Gen-1 through Gen-4.5, including an early 1,000-A100 cluster bet, Runway's role in Stable Diffusion, and how the team scaled Gen-3 tenfold following OpenAI's Sora launch. Germanidis explains how scaling video prediction allows models to learn physics and serve as simulators and World Action Models for robotics. Additionally, he explores Interface World Models that generate software pixels directly without code, the potential for a neural operating system, and the convergence of video agents and omni models.
Original sources
YouTube Latent Space
Video:Runway’s Bet Beyond Video: World Models, Robotics, and the Neural OS — Anastasis Germanidis ↗
AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.