@odysseyml
Simple, causal prediction objectives have led to surprisingly general intelligence. We’re now seeing this approach extend beyond language to world models, and the emergence of world simulators. A world simulator is a model trained to predict how the world evolves over time, frame-by-frame, using large amounts of video and interaction data. These models output continuous, interactive simulations that can power new consumer, enterprise, and intelligence applications. Many challenges remain and these models are still nascent, but after two years of research, we feel we’re at the dawn of a world simulator—something fundamentally new. Today, we’re excited to share the research directions we’re pursuing, and what we feel will mature and scale these early models into general world simulators. Let us know what you think!