Chapter 183

Musk meets with Dhaval Shroff to discuss Tesla's neural network path planner project, which trains cars to drive by learning from human behavior rather than following coded rules. After initially considering recruiting Shroff to Twitter, Musk becomes convinced of the project's importance to Tesla's future as an AI company and commits significant resources to it.

On December 2, 2022, Musk schedules a meeting with Shroff at Twitter headquarters but is delayed by a Cybertruck unveiling in Nevada and a meeting with President Macron in New Orleans. When they finally meet late that evening, Shroff pitches the neural network path planner—a system that trains autonomous vehicles by analyzing millions of video frames of human drivers rather than relying on hundreds of thousands of manually coded rules. Musk initially questioned whether this approach was necessary, but Shroff demonstrates how the neural network handles unstructured environments with debris and obstacles better than rules-based systems. Musk becomes energized, recognizing Tesla's evolution into an AI company operating in both virtual and physical domains, alongside projects like Optimus and Dojo. He implements a gamification metric—miles per intervention—displayed on massive monitors, with a gong to celebrate solved problems. By April 2023, Musk takes a 25-minute test drive with Shroff and teammates Bauch and Payne through Palo Alto using the new neural network software. The car handles cyclists, pedestrians, complex turns, and traffic flawlessly, requiring minimal human intervention. Musk is impressed enough to declare the project viable and redirect resources toward it. He notes Tesla's unique advantage: nearly two million vehicles worldwide generating billions of video frames daily—far more training data than competitors possess. The chapter concludes with Musk mentioning a secret new AI project he has just begun and regretting his earlier decision to gut Twitter's Sacramento data center.