AI Summary
→ WHAT IT COVERS Dr. Fei-Fei Li, the "godmother of AI," discusses how ImageNet sparked modern AI, her new world model company World Labs, and why spatial intelligence will unlock robotics and human augmentation. → KEY INSIGHTS - **ImageNet breakthrough:** Li created 15 million labeled images across 22,000 concepts in 2007, providing the big data foundation that enabled 2012's AlexNet breakthrough using just two NVIDIA GPUs to solve object recognition. - **AI adoption timeline:** Tech companies avoided calling themselves "AI companies" as late as 2015-2016, fearing it was a "dirty word," with widespread AI branding only beginning around 2017 - less than a decade ago. - **World models vs language models:** Spatial intelligence requires understanding 3D worlds for interaction and reasoning, not just passive video generation - essential for robotics, scientific discovery, and human augmentation beyond conversational AI. - **Robotics data challenge:** Unlike language models where training data matches output format, robotics lacks sufficient action data in 3D worlds, requiring teleoperation data, synthetic environments, and world models to bridge this gap. - **Marble production impact:** World Labs' new world model tool reduced virtual production time by 40x for Sony collaborations, enabling creators to generate navigable 3D environments from text prompts for films, games, and simulations. → NOTABLE MOMENT Li reveals that modern AI still cannot perform tasks a toddler can do, like counting chairs in office room videos, demonstrating how far current systems remain from human-level spatial reasoning capabilities. 💼 SPONSORS [{"name": "Figma", "url": "figma.com/lenny"}, {"name": "Justworks", "url": "justworks.com"}, {"name": "Synch", "url": "sinch.com/lenny"}] 🏷️ AI History, World Models, Spatial Intelligence, Computer Vision, Robotics