In this talk, Dr. Fei-Fei Li, often called the "godmother of AI," discusses her pioneering work, the evolution of Artificial Intelligence, and her new venture, World Labs. She recounts the genesis of ImageNet in 2009, a foundational dataset that spurred data-driven AI and, coupled with breakthroughs like AlexNet in 2012, catalyzed the deep learning revolution. Dr. Li traces the progression of computer vision from object recognition to complex scene understanding and generative models. She then introduces her current "delusional" pursuit: spatial intelligence. She argues that comprehending, generating, and reasoning about the 3D world is crucial for achieving Artificial General Intelligence (AGI), highlighting its inherent complexity compared to 1D language models and its deep roots in biological evolution. Drawing on her remarkable entrepreneurial journey, from running a laundromat to founding the Human-Centered AI Institute, Dr. Li emphasizes the importance of intellectual fearlessness and a "hunker down and build" mindset. She also offers advice for aspiring AI researchers and discusses the challenges and approaches to acquiring spatial data and the role of open source in the AI ecosystem.
Summarizing Dr. Fei-Fei Li's Journey and the Future of AI:
0:03 Embracing Hard Problems: Dr. Fei-Fei Li expresses her career-long pursuit of "delusional" problems and her comfort in the entrepreneurial zone of "hunkering down and building."
0:54 The Genesis of ImageNet: As a first-year assistant professor at Princeton in the late 2000s, Dr. Li observed the lack of data and the poor performance of computer vision algorithms. Driven by a dream to make machines "see" and understand the world, she conceived ImageNet in 2009 as a bold bet on data-driven AI, involving downloading a billion images and creating a visual taxonomy.
3:31 Seeding the AI Community: Between 2009 and 2012, ImageNet was open-sourced, and an annual challenge was created to encourage global research. The breakthrough came in 2012 with AlexNet (SuperVision team), which used convolutional neural networks and the first application of two GPUs for deep learning, marking a pivotal moment for AI.
8:30 From Objects to Scenes and Storytelling: ImageNet initially focused on object recognition. Dr. Li's lifelong dream was for machines to "tell the story of a scene." By 2015, with her students like Andrej Karpathy, they achieved image captioning, bridging vision and natural language.
11:33 The Rise of Generative AI: What started as a joke—taking a sentence and generating an image—has become a reality with generative AI, showcasing the rapid advancements in the field.
12:50 The Next Frontier: Spatial Intelligence: Inspired by the progress in LLMs, Dr. Li is now focusing on "spatial intelligence" through her new company, World Labs. She argues that AGI will be incomplete without machines that can understand, generate, reason about, and interact with the 3D world.
14:20 Evolution's Lesson: Vision's Primacy: Dr. Li highlights that the development of vision took evolution 540 million years, vastly longer than language (less than 500,000 years), underscoring vision's fundamental role in intelligence and evolutionary arms races.
17:17 The "Crack Team" at World Labs: Dr. Li is co-founding World Labs with world-class technologists: Justin Johnson, Ben Mildenhall (author of the Nerf paper), and Kristof Lassner (creator of Pulsar), aiming to solve the hardest problem in AI.
18:15 Why Spatial AI is Harder than LLMs: Spatial intelligence is inherently more complex due to the 3D (and 4D with time) nature of the world, the mathematical ill-posed problem of projecting 3D to 2D (like in cameras/eyes), and the need to fluidly blend generation and reconstruction, rather than being purely generative like language. Data for spatial intelligence is also less readily available.
24:18 Applications of Spatial Intelligence: Potential applications span creation (designers, architects, artists, game developers) and interaction (robotics, metaverse), with Dr. Li expressing particular excitement for the future of the metaverse as hardware and software converge.
25:50 A Lifelong Entrepreneurial Journey: Dr. Li shares her personal history, including immigrating to the US, learning English in her teens, and even running a laundromat at 19 to support her family and fund her college education, demonstrating her ingrained entrepreneurial spirit.
27:31 Fearlessness in Career Choices: Throughout her career, she chose paths less traveled, such as being the first computer vision professor in certain departments and founding the Human-Centered AI Institute at Stanford, always embracing challenges.
30:38 Advice for Aspiring AI Leaders: Dr. Li credits her legendary students for making her a better researcher and identifies "intellectual fearlessness" as the unifying trait among successful individuals and a key hiring criterion for World Labs.
32:23 Advice for PhD Students: Given that industry now commands more AI resources, Dr. Li advises PhD students to pursue interdisciplinary AI for scientific discovery, theoretical problems (e.g., explainability, causality), small data challenges, and representational problems that industry cannot easily outpace.
34:50 Nuance on AGI Definition: Dr. Li expresses her struggle with the term "AGI," viewing it as largely synonymous with the original founding goal of AI – creating machines that can think and act intelligently. She avoids defining it monolithically, comparing it to the specialized functionalities within the human brain.
37:54 The Essence of Graduate School: She believes graduate school is best suited for individuals with "burning curiosity" to solve problems and ask questions, a drive that transcends external pressures like startup commercial goals.
39:41 Open Source in LLMs: Dr. Li supports a healthy ecosystem with diverse approaches to open-sourcing, acknowledging that business strategies dictate decisions. However, she stresses the importance of protecting open-source efforts in both public and private sectors for entrepreneurial and public good.
41:15 Solving Spatial Data Challenges: For World Labs, she confirms a "hybrid approach" to data collection, emphasizing the need for both large quantities and high quality of spatial data to avoid "garbage in, garbage out."
42:13 Overcoming Minority Status: When asked about her experiences as a minority woman in STEM, Dr. Li advises focusing on the core purpose—learning, doing, creating—rather than "over-indexing" on one's minority status, a perspective she developed from her own immigrant journey.