World Labs, a ‘spatial intelligence’ startup led by Stanford University professor Feifei Li, has reportedly succeeded in attracting investment of 230 million dollars (about 300 billion won). The corporate announced that it is going to develop a ‘Large World Model (LWM)’ beyond the Large Language Model (LLM).
Reuters and Bloomberg reported on the thirteenth (local time) that WorldLabs had emerged from stealth mode and declared its official launch.
WorldLabs was first reported by Reuters in May as having established the corporate, and in July, The Verge reported that it had raised $100 million (roughly KRW 130 billion) from Andreessen Horowitz and Radical Ventures.
Nonetheless, there was no official announcement on the time. Also, the investment attraction content announced that day was expanded from the previous one.
A variety of big names have joined as investors, including NVIDIA and Google DeepMind chief scientist Jeff Dean, University of Toronto professor Geoffrey Hinton, actor and enterprise capitalist Ashton Kutcher, Salesforce CEO Marc Benioff, LinkedIn co-founder Reid Hoffman, and former Google CEO Eric Schmidt.
The corporate didn’t disclose its corporate value, nevertheless it was reported that it had already surpassed $1 billion when it attracted $100 million in investment in July.
Professor Lee disclosed specific information in regards to the company on at the present time. He explained that the goal is to develop spatial intelligence that understands and judges the actual world and to construct LWM.
“The present LLM generates text and pictures, while the LWM focuses on the power to reason about how the 3D world works,” said Professor Lee. This spatial intelligence “may be utilized in augmented reality (AR) and virtual reality (AR), in addition to robotics.”
“Images and videos from generative AI models which have emerged thus far don’t adequately convey a way of how the 3D world is constructed,” he points out. But spatial intelligence could enable broader reasoning abilities, which could help avoid hallucinations like counting fingers incorrectly, he explains.
AI that understands the physical world can also be a key research area for major AI firms. Overcoming the restrictions of LLM, which only learns in regards to the world through text, and acquiring knowledge of space through vision like humans is known as one solution to achieve artificial general intelligence (AGI).
Because of this, firms aiming for AGI, including Meta, Google, OpenAI, and xAI, are all specializing in this field.
Professor Lee said that the model can be built using images and artificial data, and the identical transformer-based architecture because the LLM. Nonetheless, the transformer is just not all the pieces, and other elements can be integrated, he said.
“If we wish to advance AI beyond its current capabilities, we want greater than AI that may see and talk. We would like AI that may do things directly,” he said, emphasizing that “spatial intelligence can be the following standard that can change the direction of AI.”
Meanwhile, Worldlab has been joined by AI veterans Justin Johnson, Christopher Lessner, and Ben Mildenhall as co-founders, and currently has 20 employees.
Professor Lee, who led the birth of vision AI by constructing ‘ImageNet’ in 2010, is named the ‘godmother of AI’. While conducting research at WorldLabs, he can even work at Stanford University’s Human-Centered AI Research Institute.
Reporter Im Dae-jun ydj@aitimes.com
