Context

Waymo co-CEO Dmitri Dolgov discusses the challenges and lessons learned in developing Waymo's autonomous driving technology at Startup School 2026. He emphasizes the complexity of transitioning from a demo to a fully operational product and the importance of safety and reliability in AI systems operating in the physical world.

Key points

• Dmitri Dolgov introduces the concept of AI that operates in the physical world, specifically through Waymo's autonomous vehicles. 0:21
• Waymo's autonomous vehicles have achieved a superhuman safety record, driving over 4 million miles weekly across 15 cities. 2:30
• The transition from a demo to a product is a massive undertaking, with a demo representing only 1% of the work needed for a real product. 8:18
• Physical AI requires a different approach than digital AI, focusing on safety and reliability from the start. 3:25
• Waymo uses a multimodal approach with cameras, lidar, and radar to achieve superior sensing capabilities. 16:51
• The importance of closed-loop simulation for training and evaluating AI agents in the physical world is emphasized. 40:52
• Waymo's strategy includes building an ecosystem with an agent, simulator, and critic to accelerate progress. 42:23
• Trust is critical in deploying AI in the physical world, and Waymo earns it through rigorous evaluation and public transparency. 44:58
• Waymo's autonomous vehicles are significantly safer than human drivers, preventing serious injuries regularly. 46:53

Quotes

More verdicts

• WATCHPatrick Collison: "What If You Succeed?"4.6/5 · 31 min
• WATCHMulti-GPU Kernels, Intelligence per Watt, Heterogeneous Inference, and More | YC Paper Club4.6/5 · 1h 16 min
• WATCHJeff Dean: The 1% Rule for Building in AI4.6/5 · 57 min
• WATCHAlexandr Wang: This is a Once-in-a-Civilization Opportunity 4.7/5 · 32 min