According to a report by the Financial Association on January 7th (Editor: Zhou Ziyi), $Alphabet-A (GOOGL.US)$ / $Alphabet-C (GOOG.US)$ a new team is being formed, focusing on developing AI models that can simulate the physical world.

The head of this team is Tim Brooks, who was one of the co-leaders of OpenAI's video generator Sora and left OpenAI in October 2024 to join Google's AI research lab, Google DeepMind.

According to Brooks’ latest Post on social media X on Monday (January 6), the new team he leads belongs to DeepMind and will focus on developing AI models that can simulate the physical world.

Brooks wrote, "DeepMind has ambitious plans to create large-scale generative models to simulate the world. I am recruiting a new team to accomplish this task."

According to the description, Brooks' new team will focus on developing "real-time interactive generation" tools and researching how to integrate their models with existing multimodal models (such as Gemini).

The new team will also collaborate with Google's existing teams including Gemini, Veo, and Genie, expanding the models to the "highest level of computation" based on the work of the latter team.

It is understood that Gemini is Google's flagship AI model series used for tasks such as image analysis and text generation; Veo is Google's video generation model; and Genie is Google's foundational world model that provides users with an unprecedented interactive experience, allowing them to create playable interactive environments and controllable virtual worlds through text, synthesized images, photos, and even sketches.

In December last year, Google also launched an upgraded version of Genie 2, capable of simulating a controllable, highly playable realistic 3D virtual world.

World model.

The job description released by Bruker Corp states: "We believe that expanding AI training on video and multimodal data is the key way to achieve Artificial General Intelligence (AGI, referring to AI that can perform any task that humans can complete). World models will power many areas, such as visual reasoning and simulation, planning for instantiated agents, and real-time interactive Entertainment."

"World models" are currently one of the most challenging in the technical field, characterized by enabling machines to have a comprehensive and accurate understanding of the real world, similar to humans.

Currently, many startups and large Technology companies are pursuing world models, including influential AI researcher Fei-Fei Li's World Labs, the rising Israeli company Decart, and the startup Odyssey. They believe that world models could someday be used to create Interactive Media, such as video games and movies, and run realistic simulations, such as robot training environments.

However, the development of world models seems unfriendly to some creatives.

A recent survey by the renowned magazine Wired found that game studios like Activision Blizzard have laid off a large number of employees while utilizing AI to increase productivity and compensate for personnel losses. A study commissioned by the USA animation association in 2024 estimates that by 2026, over 100,000 jobs in film, television, and animation in the USA will be replaced by AI.

Compared to being replaced, collaboration seems to be a friendlier approach. Some emerging startups in the world model field, like Odyssey, have already committed to collaborating with creative professionals. Whether this will become Google's approach remains to be seen.

Additionally, regarding world models, copyright issues remain unresolved. Some world models are trained using clips from video games, which may expose the companies developing these models to lawsuits if the videos are used without permission.

Google, which owns YouTube, claims that according to the platform's terms of service, it has the right to train its models based on YouTube videos. However, the company has not disclosed which specific videos will be procured for training.

编辑/jayden