Google DeepMind is setting up a new AI research team dedicated to developing world models for real-world simulations in gaming and robot training.

Google DeepMind's latest initiative aims to advance artificial intelligence by establishing a dedicated team focused on constructing 'world models.' These models are designed to simulate physical environments and support the development of video games, movies, and realistic training scenarios for robots. Led by Tim Brooks, formerly of OpenAI, this endeavor is critical in Google’s quest for an artificial general intelligence system (AGI), positioning it competitively in the fast-evolving AI landscape.

In a recent announcement, Brooks detailed that the new team will focus on creating expansive generative models capable of simulating real-world scenarios. They will tackle challenges related to training at scale, curating training data, and integrating with multimodal language models. DeepMind believes that scaling pretraining on video and multimodal data is pivotal to achieving AGI, with potential applications in visual reasoning, simulation, and real-time interactive entertainment. As the race for AGI intensifies, Google's foray into world modeling demonstrates its commitment to maintaining a lead in the AI race, especially against formidable players like Nvidia and emerging startups.