Ming Min Cressy from the temple of Au Fei
Quantum Position | 公众号 QbitAI
AI godmother Li Feifei has started a business!
In the latest news, Stanford University professor Li Feifei is building an AI company and has completed a seed round of financing.
The company's direction is "Spatial Intelligence" – the aim is to enable AI to perform advanced inference on visual information in the same way that humans do. Sources say it will be a leap forward for the technology.
投资方包括硅谷风投a16z和Radical Ventures。
As the most influential woman and Chinese in the field of AI, Li Feifei has made significant contributions to academia and industry for a long time. She held tenured faculty positions at Stanford, served as the Chief Scientist of Google Cloud AI, promoted the establishment of the Google AI China Center, and coordinated the work of Google Cloud AI, Google Brain, and local teams in China for a long time.
But this is the first time I have started a business.
At present, Li Feifei and relevant investors have not responded to this new news.
But Li Feifei has a new resume on his LinkedIn page.
The title is "newbie", and the introduction only writes "something new", and it is confirmed that it is full-time, starting in January 2024, and it has been going on for 5 months now.
Plan "Enabling AI for Advanced Inference"
汇总各方信息，目前比较有迹可循的线索是参投的加拿大风投Radical Ventures。
Last year, Li Feifei joined the fund as a scientific partner last year.
More clues can be seen from Li Feifei's recent public activities.
Last month, Li Feifei gave a keynote speech at the TED conference in Vancouver, which was related to spatial intelligence.
In his speech, Li Feifei described "spatial intelligence" as the ability to derive predictions and insights into the relationship between objects.
Advances in AI's understanding of spatial intelligence are catalyzing robot learning and bringing us closer to the goal of enabling AI to interact with the world, she said.
To explain "spatial intelligence," she showed a photo of a cat stretching out its paws to push a glass toward the edge of a table.
In this instant, she says, the human brain can assess the geometry of the glass, its position in the 3D world, its relationship to tables, cats and other things, and can predict what will happen next and take action to stop it.
Nature creates this virtuous cycle of vision and behavior driven by spatial intelligence.
At the same time, she said that her team is also trying to teach computers how to move in the 3D world in the lab.
For example, using a large language model, a robotic arm can perform tasks, open a door, make a sandwich, and respond to human verbal instructions.
The work mentioned here is probably related to the embodied intelligence work VoxPoser released last year.
This project would allow humans to give commands to robots, such as "Open the drawer above, watch out for the vase!"
Then the large language model + visual language model can analyze the goals and obstacles that need to be bypassed from the 3D space to help the robot plan its actions.
And in the real world, robots can perform this task directly without being "trained".
In addition, Li Feifei has recently participated in the publication of a series of academic achievements, including computer vision, robotics, intelligent bodies, etc.
For example, some portable action data collection systems.
Some are related to agents.
For example, Li Feifei participated in an embodied intelligence benchmark released last month involving more than 1,000 human daily activities.
In terms of 3D space, Li Feifei's team also released an aerial 3D modeling model at the beginning of this year, which can model the movements of characters through occlusions.
In addition, she co-published a position paper on Agent AI with 15 researchers from Microsoft, Stanford, UCLA, and others.
The basic concept of Agent AI is clarified in this paper, and the basic model of Agent AI is proposed, that is, a general agent system built by using multimodal pre-trained models.
The document states that these models can process interactive multimodal data from robotics, games, and more.
In addition, Li Feifei's Stanford homepage shows that she will be on "partial leave" from the beginning of this year to the end of 2025.
This start time coincides with the start time of Li Feifei's "newbie" resume on his LinkedIn page.
AI godmother Li Feifei
Today, Li Feifei has become one of the most influential women and Chinese in the field of AI.
Corresponding to the title of AI godfather, everyone calls her AI godmother.
Her legendary experience has always been talked about -
At the age of 33, he became a tenured professor in the Department of Computer Science at Stanford, and at the age of 44, he became a member of the National Academy of Engineering, and is currently the president of the Stanford Institute for Human-Centered Artificial Intelligence (HAI).
She is a leading figure in the field of computer vision, and the founder of ImageNet, which has become a benchmark for advancing the field of computer vision recognition.
He has many high-ranking disciples, such as Andrej Karpathy, who has worked in OpenAI, Tesla, and Jim Fan, who is currently at NVIDIA, etc., are also influential figures in the field of AI.
Later, Li Feifei briefly entered the industrial world as vice president of Google, that is, chief scientist of Google Cloud AI. She single-handedly promoted the official establishment of the Google AI Center in China, which is Google's first AI research center in Asia. He also led Google Cloud to launch a series of influential products, including AutoML, Contact Center AI, Dialogflow Enterprise, etc.
In the latest trend of large models, Li Feifei's team focuses on embodied intelligence, integrating large models into robots, and transforming complex instructions into specific action plans without additional data and training, opening up an important direction in the new wave.
Li Feifei has always been a vane in the field of AI.
Recently, Li Feifei's personal memoir "The World I See" was officially published. In this book, Li Feifei describes how he gradually found his mission in the AI revolution from Beijing to Chengdu to the United States, and shares his views on the development of AI.
As a result, a more three-dimensional Li Feifei was outlined.
She said that she is not an open-source or a destroyer, and that she hopes that humans can live in peace with AI.
She shared that Google Cloud's experience has made her realize that leading the development of AI technology requires corresponding responsibilities.
She also said that if you're looking for a trendy algorithm, it's certainly not the best science.
After all, she had worked on unpopular projects in the past — annotating 14 million images for training ImageNet over the course of several years, despite the advice of her mentors.
……
Li Feifei admitted that he was very shy and did not dare to tell his story.
Writing memoirs in my forties? I'm not Einstein.
And 40 years after writing his first 40 years, Li Feifei opened a new charter for himself.
As she said, civilization is like a big ship, and we sail in the dark.
Scientist Li Feifei now has a new story.
Reference Links:
[1]https://www.reuters.com/technology/stanford-ai-leader-fei-fei-li-building-spatial-intelligence-startup-2024-05-03/
[2]https://www.ft.com/content/d5f91c27-3be8-454a-bea5-bb8ff2a85488
[3]https://blog.ted.com/mind-expanders-notes-on-session-2-of-ted2024/
— END —
量子位 QbitAI 头条号签约
Follow us and be the first to know about cutting-edge technology trends