Clearly not content with taking over the world, Google is in the process of Staffing increased DeepMind Research Lab to create generative models that can simulate the physical world The project, which will be led by Tim Brooks, one of the leaders who helped build Sora, OpenAI's video creation engine, will be a key part of the company's efforts to achieve artificial general intelligence. Work list related to the new team
Brooks, who later joined DeepMind Escape from OpenAI Back in October and his team has “An ambitious plan to create a large-scale generative model that simulates the Earth,” according to the role description. Efforts to model the world will “It drives many different domains, such as reasoning and visualization. Planning for embodied agents and real-time interactive entertainment” if you are willing to take on one of these roles. You may understand the meaning of those vague things and come back to us.
g world modelSimply put, it tries to simulate how the real world works. Next-generation models like Sora can replicate what has been seen before within training data. But there is no real understanding of why that happens. Thus, a video of a baseball pitcher can be successfully created. But there was no understanding of the physics of what was happening. World models aim to equip machines with enough information to parse how actions occur and their possible outcomes.
Yann LeCun, Meta's chief AI scientist, described the world model this way during Speech at the Hudson Forum Earlier this year: “A world model is your mental model of how the world behaves…you can imagine the sequence of actions you might take. And your world model helps you predict what the impact of a sequence of actions will be on Earth.”
World models are difficult to construct for a number of reasons. This includes the enormous amount of computation required to run the model. and the lack of sufficient training data to build accurate models. As a result, most world models work only in limited and specific contexts.
It seems that the team at DeepMind intends to make the world model more expansive. The plan was to create a tool. “Real-time interactive creation” on top of the model And they might consider how they can integrate the world model with Google Gemini's large language model.
One likely area that DeepMind will attempt to tackle is video games. The new team's job description states that they will work with the Veo and Genie teams at Google. genius It is a video creation program similar to Google's Sora and Genie. It is a built-in world model that can simulate 3D environments in real time. The video game industry already exists. Eager to adopt AI toolsreplacing thousands of workers CVL Economics The survey found that more than 86% of all gaming companies have already adopted AI tools, and nearly 15% of all gaming work could be disrupted by 2026.
Perhaps improving this world would be better spent building a model.