December 12th, the technology media, Ars Technica, published an article today, December 12, reporting Runway PUBLISH GWM-1 "world model"Sequence, claiming to be able to maintain a consistent picture for several minutesIn addition, the company ' s strategic focus is expanding from film production to robotics, physics and life sciences。
Note: The World Models, an AI system that understands and simulates physical world operations, not only generates images, but also predicts causality such as “If I throw an apple, it falls on the ground”, similar to human cognitive models of the environment。
In terms of structure, GWM-1 is not a single model, but consists of three different post-training models. In response, Runway explained that this was a long-term plan and that the current pattern was an important step towards that ultimate goal, as the company was working to integrate different areas and action spaces into a single underlying world model。
In terms of development strategies, Runway is trying to break the Hollywood-style video applications. Although the world modelcinematicThere is a wide range of theoretical applications in the development of television, advertising and games, but Runway has made it clear on the air that companies have turned their eyes to new areas of robotics, physics and life science。
These new areas require a higher level of understanding of the physical law of the model than the creative industries of the past, and also mean that Runway needs to prove its technical value on a completely new track。
The media noted that, unlike in the video-generation area, the differential advantage of Runway here is unclear. Prior to that, Runway had successfully established itself in the film and advertising industry, thanks to the creative industry background of its founders and the tools designed for its creators。
However, in the field of robotics and scientific research, many technological giants are well established and well funded. As a start-up company, Runway ' s ability to take food from a giant without pre-emptive advantages and resource barriers remains to be tested in the market。
In the future of the layout, Runway did not ease the upgrading of core business. The company synchronized the launch of Gen 4.5 video generation model, adding primary audio generation, audio editing and multi-photographic video clips, and further refining the video production workflow。
In addition, to support the growing computing needs, Runway announced cooperation with the cloud computing company CoreWeave, which focuses on AI. According to the agreement, Runway will use the British GB 300 NVL72 hangar on the cloud infrastructure of CoreWeave to train and reason future models。