Google wants to simulate the world using artificial intelligence

Clearly not satisfied with its grip on the world, Google is in the process of doing so Employment Its DeepMind research lab builds generative models capable of simulating the physical world. The project — which will be headed by Tim Brooks, one of the leaders who helped build OpenAI’s video generator, Sora — will be an important part of the company’s attempt to achieve artificial general intelligence, according to the OpenAI website. Job listings Related to the new team.

Brooks, who later joined DeepMind Escape from OpenAI Last October, his team had “ambitious plans to make massive generative models that simulate the world.” According to the role description, efforts to build universal models will “advance several areas, such as visual reasoning and simulation, planning for embodied agents, and real-time interactive entertainment.” If you’re willing to take on one of these roles, maybe you can find out what these mysteries mean and get back to us.

A Universal modelAnd, as simply as possible, it usually seeks to mimic the way the world actually works. Generative models like Sora are able to replicate things they have seen before within their training data, and have no real understanding of why that thing happened. So he can successfully create a video of someone throwing a baseball, but he has no understanding of the physics of what’s happening. Global models aim to provide the machine with sufficient information to analyze how an action occurs and its likely outcome.

Yann LeCun, a senior AI scientist at Meta, described global models 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 a sequence of actions you might take, and your world model will allow you to predict what effect the sequence of actions will have on the world.” world.”

Global models are difficult to build for a number of reasons, including the massive amount of computing needed to run the model and the lack of sufficient training data to create an accurate model, resulting in most global models only working in limited, specific contexts.

The DeepMind team seems intent on scaling up the global model. The plan is to build “real-time interactive generation” tools on top of the models and perhaps look at how to integrate their global model into Google’s larger Gemini language model.

One potential area that DeepMind will try to address is video games. The job description for the new team states that they will collaborate with Google’s Veo and Genie teams. The genie A video generator similar to Google’s Sora, Genie is an existing world model that can simulate 3D environments in real time. The video game industry already exists We are keen to adopt artificial intelligence toolsAnd the displacement of thousands of workers. A CVL Economics The survey found that more than 86% of all gaming companies have already adopted generative AI tools and that nearly 15% of all gaming functions could be disrupted by 2026.

Improving this world is probably a better use of time than modeling it.