Google DeepMind is forming a new team of specialists to develop artificial intelligence (AI) systems capable of imitating real physical environments. These types of algorithms are known as world models or models of the world and represent a significant advance towards general AI (AGI).
The initiative is led by Tim Brooks, former co-director of OpenAI’s Sora project, who has shared the company’s intentions through a post on social networks. “DeepMind has ambitious plans to create massive generative models that simulate the world,” he says. The statement was accompanied by two job openings targeting engineers and researchers with experience in large-scale training of generative models, training data curation, and multimodal language systems.
The job descriptions highlight that “expand pre-training with data Multiformat is a fundamental step towards AGI. “World models will advance numerous domains such as visual reasoning and simulation, planning for embodied agents, and real-time interactive entertainment.”
What are the world models from Google?
The business development agency The Black Box Lab indicates that world models are algorithms that generate internal representations of an environment, allowing AI solutions to simulate events and predict future outcomes based on those emulations. The approach seeks to replicate the reasoning and thinking processes of humans to provide machines with a deeper understanding of the physical context.
An article from TechCrunch explains that a generative model “trained on years of video” can accurately recognize that a basketball bounces, but is unable to interpret why. In contrast, a model of the world has “a basic understanding” of the causes of rebound, which allows it to represent the phenomenon and anticipate future movements with greater precision.
Yann LeCun, chief scientist of artificial intelligence at Meta, has highlighted the potential of this technology to carry out complex forecasts and planning in digital and physical fields. “We need machines that understand the world, that can remember things, that have intuition and common sense. [Se trata de algoritmos] that they can reason at the same level as humans. “Current AI systems are not capable of any of this,” LeCun explained last year at a conference. Various technology organizations are investing more and more resources to accelerate the development of these types of models.
An example is World Labs, a startup co-founded by technologists Fei-Fei Li, Justin Johnson, Christoph Lassner and Ben Mildenhall and defines itself as “a spatial intelligence company that builds large world models to perceive, generate and interact with the world in 3D.” Johnson explains that it is currently possible to produce interactive virtual scenarios, but the cost is still high and the process is time-consuming. “[Los modelos de mundo] will allow not only to obtain an image or a clipsbut a completely simulated, vibrant and interactive three-dimensional world made to measure.”
For its part, Nvidia this week unveiled its Cosmos platform, a platform comprising next-generation generative global-based models designed to advance the development of physical AI systems, autonomous vehicles and robots.
Experts suggest that this technology is the previous step to achieving general AI. Systems of this type would be capable of executing any intellectual task that humans perform with equal or greater accuracy and efficiency.
#Google #creating #team #develop #capable #simulating #real #world
