Tesla's AI Drives Itself With Neural Network Learned from Millions of Human Drivers

Tesla's new self-driving system, FSD 12, uses a neural network that learns from analyzing millions of examples of human driving, similar to systems like ChatGPT.

The neural network approach allows the car to handle complex unstructured environments better than rigid rules-based systems.

The metric guiding the neural network is to maximize miles driven without human intervention. Engineers track this on TV monitors like a video game score.

In tests, FSD 12 impressed Musk by smoothly navigating obstacles and making smart unprompted moves.

With a fleet collecting driving data, Tesla has an advantage in training neural networks over other companies. But human drivers often fudge rules, which the system replicates.