Core Insights - Elon Musk estimates that achieving truly safe unsupervised Full Self-Driving (FSD) requires approximately 10 billion miles (about 16.09 billion kilometers) of training data [3] - Musk's statement responds to Paul Baizel, a former employee of Apple and Rivian, who highlighted the gap between technical demonstrations and actual products, asserting that Tesla's data-driven approach gives it a significant lead in the autonomous driving sector [3] - Baizel argues that relying solely on simulation and limited road testing to catch up in FSD technology is overly optimistic, emphasizing that the challenge is about scale, data, and iteration speed [3] Group 1 - As of the end of 2025, Tesla's FSD system has accumulated nearly 7 billion miles of testing mileage, with over 2.5 billion miles on urban roads, indicating that Tesla likely has the highest training data reserves for autonomous driving projects globally [4] - Musk commented on the challenges of achieving autonomous driving technology, stating that while reaching 99% of technical goals may be easy, overcoming the remaining long-tail challenges is extremely difficult [4] - Ashok Elluswamy, Tesla's Vice President of AI Software, echoed Musk's sentiments on social media, noting that the complexity of long-tail issues exceeds most people's imagination [4]
马斯克:要实现安全的无监督自动驾驶,需要100亿英里的训练数据