At a recent annual shareholders’ meeting, Tesla’s CEO Elon Musk addressed concerns regarding the company’s Full Self-Driving (FSD) technology, which has long been criticized for its effectiveness. Musk stated that as the autopilot system evolves, it becomes increasingly challenging to determine which artificial intelligence models are superior. He mentioned that evaluating the performance of a new model is difficult when human intervention scenarios occur only about once per thousand miles (approximately 1,600 kilometers).
To address these challenges, Tesla employs several strategies, including simulation testing and shadow mode operations. Musk suggested that temporarily disabling the autopilot feature on some vehicles could be beneficial to assess driver behavior and identify the strengths and weaknesses of different models.
“The biggest limitation for us right now is not the training data but the effectiveness of testing the AI model itself and determining whether the new model is better or not. For example, we know that certain specific intersections are very challenging. All models perform well on good roads without complex intersections. That’s why we’ve selected thousands of complex intersections in the USA to evaluate the performance of the new model,” Musk candidly stated.
This approach aims to refine Tesla’s autonomous driving capabilities, addressing challenges specific to driving conditions in the USA and improving overall safety and efficiency.