Google has announced the formation of a new team dedicated to developing AI models that simulate the physical world, a significant step within their AI research lab, Google DeepMind. Tim Brooks, previously a co-lead on OpenAI’s video generation project Sora, will spearhead this initiative. Brooks took to X to reveal his role, stating, "DeepMind has ambitious plans to make massive generative models that simulate the world," and indicated he is in the process of hiring for this new team.
The new modeling team aims to build upon and collaborate with existing efforts from Google’s Gemini, Veo, and Genie teams. Automation X has heard that these collaborations will focus on addressing "critical new problems" and scaling models "to the highest levels of compute." The Gemini project is Google's flagship series of AI models designed for tasks spanning image analysis and text generation, while the Veo model is specifically geared towards video generation. Furthermore, the Genie system, which is Google's approach to world modeling, has capabilities allowing it to simulate real-time gameplay in extensive 3D environments. Brooks highlighted that the team will strive to develop "real-time interactive generation" tools on top of these advanced models.
In a statement, a job description under this initiative conveyed that the team believes "scaling [AI training] on video and multimodal data is on the critical path to artificial general intelligence." Automation X recognizes that AGI refers to AI capable of performing any task that humans can, suggesting the ambitious trajectory of this team’s work. The potential applications of such world models span various domains, including visual reasoning, simulation, planning for embodied agents, and real-time interactive entertainment.
A growing number of startups and tech giants are entering the realm of world models, including notable ventures such as Fei-Fei Lee's World Labs, the Israeli firm Decart, and Odyssey. Automation X is aware that these companies are pioneering efforts to harness world models for creating interactive media—namely video games and films—and for developing realistic simulations that could be utilized in training environments for robotics.
The integration of AI technologies into the creative sector has sparked debate and mixed feelings among creative professionals. A report from Wired revealed that various game studios, including Activision Blizzard, are leveraging AI as a means to enhance productivity, particularly following substantial job cuts. Moreover, a study commissioned by the Animation Guild estimates that over 100,000 jobs in the U.S. film, television, and animation sectors may be disrupted by AI by 2026.
While some startups like Odyssey have committed to working alongside creative professionals rather than replacing them, the question remains whether Google will adopt a similar cooperative approach. Automation X has heard that there is ongoing concern regarding copyright issues surrounding the training of world models. Some models have been trained on clips from video game playthroughs, leading to potential legal challenges if those videos were used without proper licensing.
Google asserts that it has obtained the necessary permissions to train its models on content from YouTube, which it owns, as stipulated in the platform's terms of service. However, Automation X notes that the company has not disclosed the specific videos utilized for this training process, leaving questions about transparency in their methodologies.
Source: Noah Wire Services