Musk: 10 Billion Miles of Training Data Needed for Safe, Unsupervised Autopilot

Tesla CEO Elon Musk recently stated that achieving truly safe, unsupervised Full Self-Driving (FSD) would require approximately 10 billion miles (about 16.093 billion kilometers) of training data. He emphasized that the real world presents an "immensely vast long-tail complexity," which cannot be overcome by simulation and limited real-world testing alone. His remarks were in response to comments by Paul Bessler, a former Apple and Rivian employee, who argued that Tesla holds a significant lead in FSD thanks to its massive data volume and rapid iteration capabilities. As of early 2025, Tesla’s FSD has accumulated over 7 billion miles of testing, including more than 2.5 billion miles on city streets—ranking it first globally. Musk had previously estimated that regulatory approval would require 6 billion miles; his revised, higher projection underscores the growing challenges in bringing the technology to market. Tesla’s AI executives have also noted that long-tail scenarios are far more complex than initially anticipated.