Over the past year, world models have transitioned from a research topic to the foundation of new commercial projects and significant funding rounds. These models aim to simulate the physical world or at least a useful approximation of it, enabling AI systems to interact with environments in ways that go beyond traditional language processing. This shift has sparked interest among researchers and investors alike, with several companies making notable announcements in the space. Source: arstechnica
In August, Google DeepMind unveiled Genie 3, a model that builds real-time interactivity on top of a video generation model. In November, World Labs introduced Marble, a toolset that allows users to generate immersive environments based on text, images, or video inputs. Runway followed with the announcement of GWM-1, a trio of specialized world models built on its prior work with video models. These developments highlight the growing interest in world models as a practical approach to simulating and interacting with physical environments. Source: arstechnica
Vincent Sitzmann, an MIT researcher, described a world model as any system that takes in an interaction and enables simulation of what would happen next in an environment. Runway defined it as an AI system that builds an internal representation of an environment and uses it to simulate future events. Ben Mildenhall, co-founder of World Labs, emphasized that world models differ from large language models by allowing continuous, real-time interaction with spatial environments. Source: arstechnica
