Creating 'world models' for robots + An AI math shakeup
Summary
Tech journalist Joanna Stern discusses the challenges of training AI for physical tasks, noting that humanoid robots are not yet ready for homes. She shares her experience filming daily chores to train models, highlighting the need for better data and safety. Later, mathematician Dr. Emily Reel explains how AI is solving long-standing conjectures, acting as a tool to find counterexamples while humans still interpret the results.
Topics discussed
Sponsors: Brooklyn Solar Works and Odoo
Intro: AI World Models vs. Text-Based Models
Why Physical World AI Needs Different Data
Training Robots via Teleoperation and Video Data
Current State of Laundry Folding Robots
Privacy Concerns and Timeline for Humanoid Robots
Outro and Sponsor Messages
Intro: AI Disproving Math Conjectures
Surprising Pace of AI Progress in Mathematics
OpenAI's 10 Solved Problems and Verification
AI Strengths: Finding Needles in Haystacks
Future of AI: Exploring the Mathematical Universe
Credits and Final Sponsor Message
Listen ad-free on Castria