Lenny's PodcastDr. Fei-Fei Li: Why world models come next, not bigger LLMs
Through ImageNet, AlexNet, and GPUs the modern AI recipe was set; today's LLMs cannot reliably count chairs in a video, and world models are how that changes.
At a glance
WHAT IT’S REALLY ABOUT
Godmother of AI Fei-Fei Li bets big on world models
- Dr. Fei-Fei Li traces the evolution of AI from early machine learning and the "AI winter" through ImageNet and deep learning to today's large language models, emphasizing how data, neural networks, and GPUs enabled the current boom.
- She argues that AI is a civilizational technology and a double‑edged sword whose impact on jobs, dignity, and society is ultimately determined by human choices, not technological inevitability.
- Li explains why current systems are still far from human‑level intelligence, why AGI is mostly a marketing term, and why new breakthroughs in spatial understanding and "world models" are needed—especially for robotics and embodied intelligence.
- She introduces Marble, the first large world model from her company World Labs, which can generate fully explorable 3D worlds from prompts, unlocking applications in film, games, robotics simulation, design, science, and even psychology research.
IDEAS WORTH REMEMBERING
5 ideasAI’s trajectory is shaped by people, not inevitability.
Li stresses that AI is designed, deployed, and governed by humans; whether it augments dignity and work or harms society depends on individual, corporate, and policy choices at every stage.
Big data, neural networks, and GPUs formed the ‘golden recipe’ of modern AI.
ImageNet’s millions of labeled images enabled deep learning breakthroughs like AlexNet, which, combined with GPUs, established the basic paradigm that still underpins systems like ChatGPT.
Current models are powerful but still far from general human intelligence.
Despite impressive language and coding abilities, today’s AI cannot perform basic spatial reasoning (like robustly counting chairs in videos) or creative scientific leaps (like deriving Newton’s laws), highlighting clear ceilings of the current approach.
AGI is more a marketing label than a scientific concept.
Li sees no clear scientific boundary between ‘AI’ and ‘AGI’ and prefers to focus on the long‑standing north star—building systems that can think and act in human‑like ways—rather than on hype terms.
World models and spatial intelligence are essential to move beyond chatbots.
To power robotics, immersive environments, and many real‑world tasks, AI must understand 3D space, objects, dynamics, and interactions, not just sequences of tokens, which is what world models aim to capture.
WORDS WORTH SAVING
5 quotesThere’s nothing artificial about AI. It’s inspired by people, it’s created by people, and most importantly, it impacts people.
— Fei-Fei Li
I’m a humanist. I believe that whatever AI does, currently or in the future, is up to us.
— Fei-Fei Li
AGI, I feel, is more a marketing term than a scientific term.
— Fei-Fei Li
We operate on about 20 watts… and yet we can do so much. The more I work in AI, the more I respect humans.
— Fei-Fei Li
No technology should take away human dignity.
— Fei-Fei Li
High quality AI-generated summary created from speaker-labeled transcript.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome