Li Feifei, a renowned overseas Chinese artificial intelligence scientist and professor at Stanford University, recently launched a new AI company. This is also the first time she has directly participated in a business program since leaving Google and returning to Stanford in 2018.
Li’s personal English page has also undergone corresponding innovative changes. The resume column at the top shows that she has had a new position since January 2024, with the relevant name anonymously displayed as “something new” and the rest status as “full-time”.
According to Reuters, Li Feifei’s new company mainly deals with the research and development of “space intelligence” and has achieved seed round financing, with investors including a16z and RadicalVentures. Li Feifei himself is also one of the partners of Radical Ventures, a self reliant fund specializing in artificial intelligence and related refurbishment. It was established in Canada and received investment from top AI scientists such as Jeff Hinton and Li Feifei at the beginning of its establishment.
The new company’s focus on “spatial intelligence” is also consistent with the AI application bias frequently mentioned by Li Feifei in many places recently.
According to its central report presented at the TED conference last month, space intelligence is achieved by understanding the connections between objects, losing new discoveries or speculations. This is a computer vision intelligence that is more advanced than conservative visual recognition. It is a study that decomposes the results of natural speech models, robots, and computer vision multi domain research. Machines can stop more complex visual reasoning like humans and adopt more realistic behaviors.
In her report, she presented a photo of a cat extending its paw to push a glass cup towards the edge of a table to assist the audience in understanding what spatial intelligence is: the human brain can usually evaluate several forms of glass cups, their position in the real three-dimensional world, their relationship with tables, cats, and other objects at this moment, and can also speculate on what will be cooked when handed over, adopting behavior.
And what spatial intelligence needs to do is teach computers how to behave in the real three-dimensional world.
Last November, Li Feifei’s laboratory had already released a “VoxPoser” intelligent robot. This research combines the universal speech model with various techniques such as computer vision and robotics, allowing robots to directly understand human natural speech commands by immersing themselves in the model and turning these complex commands into specific behavioral layouts. This robot can not only smoothly accomplish daily tasks similar to human behavior, such as “cleaning up trash into a dustpan”, “closing drawers and being careful not to avoid nearby vases”, but also integrate and manipulate according to special behavioral preferences such as “I am left-handed” and “you are 1 centimeter off”.
More importantly, the team claims that this robot can achieve a “zero sample”, meaning that it can be practiced without delaying specific mission exercises. Although this name is still in the laboratory trial stage, it is still enough to make the robotics industry, which has been struggling with “insufficient data” for years, extremely frustrated.
“Let’s get closer to a world where personal artificial intelligence can not only be seen and invented, but also interact with the physical world around us.” Li Feifei summarized the ultimate vision of space intelligence in TED’s evolution.
Making computers “see” like humans is also a footnote in Li Feifei’s years of artificial intelligence research. Her recently published personal autobiography also uses the action book title “The World I See”.
In 1976, Li Feifei was born and matured in Beijing, China, and grew up in Chengdu, Sichuan. In 1992, he followed his parents to the United States for immigration leave and was admitted to Princeton University and California Institute of Technology for undergraduate studies, where he became proficient in his studies.
In 2005, Li Feifei, who was under 30 years old and had just obtained a PhD in Electronic Engineering from the California Institute of Technology, co launched the ImageNet project with one of his fellow researchers. This title marks the beginning of her academic maturity and is also her most important contribution to the artificial intelligence industry to date.
ImageNet is a database of over 15 million annotated high-definition images, revealing 22000 different types of images. The vast dataset provides abundant capital for the practice of artificial intelligence deep learning models. Afterwards, she and Menshu launched the ImageNet Challenge, which has a history of 14 years since 2010 and is held once a year. It has become the most eye-catching top-level competition in the field of computer vision.
The above two pauses are also believed to have laid the foundation for the comprehensive outbreak of artificial intelligence reactionaries. In 2022, when Li Feifei was elected as a member of the American Academy of Engineering, the public wrote in the award vocabulary, “Li Feifei has made contributions to establishing a knowledge base for large-scale mechanical education and visual understanding.”
