GM Trains Self-Driving AI 50,000x Faster Than Real Time

What Happened General Motors has revealed its approach to training autonomous driving AI at unprecedented speeds, using simulation environments that operate 50,000 times faster than real-time driving conditions. The automaker is combining large-scale simulation, reinforcement learning, and foundation-model-based reasoning to develop what it calls “scalable driving AI.” The system specifically targets what engineers call the “long tail” problem in autonomous driving—the rare, ambiguous, and unexpected events that occur infrequently but pose the greatest safety challenges.