Jim Fan of Nvidia’s embodied AI group gave a fascinating talk at Sequoia Capital’s AI Ascent get together this week, sharing numerous examples of AI-powered and human-like motion. Fan’s “Physical Turing Test” is a challenging vision for embodied AI: coming home to an immaculate living room and candlelit dinner, with no way to tell if a human or machine had cleaned up and prepared a gourmet meal. Fan describes this as “deceptively simple, insanely hard” and the “next North Star of AI”, a dream that keeps him working late nights in the lab. There’s no rest for his robots either, who work tirelessly inside the Nvidia’s digital twin, compressing a decade of learning into every few hours.
Takeaways: AI’s learning in simulated environments is another powerful way to make training faster and more effective. Such virtual training grounds allow developers to stress-test agents under tough conditions, ensuring that when deployed to tackle high criticality uses in healthcare or financial services, they are battle-hardened and ready to go.