In the midst of its continued success, some investors are betting that the company's seemingly unstoppable rise is about to come to an end.$Tesla (TSLA.US)$During the 2024 annual shareholder meeting, CEO Elon Musk discussed a major issue plaguing Tesla's Full Self-Driving (FSD) system. This issue confirms speculation from industry experts two years ago and could potentially affect Tesla's goal of achieving self-driving taxis.

Musk stated that as the FSD system continues to improve, it becomes increasingly difficult to determine which AI model is better. This is because, after thousands of miles of driving without any intervention needed, it becomes a challenge to quickly evaluate the performance of a new model.

"These different AI models won't necessarily solve all the problems perfectly," Musk admitted. "One model may solve problem A, but may introduce problem B."

Cleantechnica editor Zachary Shahan speculated two years ago that Tesla's FSD system may have this "teeter-totter problem," where the system may introduce new problems while trying to fix existing ones, resulting in overall stagnation.

So how does Tesla solve this problem? "We're addressing this in a whole bunch of ways, including simulation testing, shadow mode operation," Musk explained. "Having some percentage of cars not running FSD is actually helpful, and we can compare the driving behavior of the new models versus the user behavior through shadow mode to identify differences in the strengths and weaknesses of different models."

Tesla can use its millions of autos for testing and compare the predicted behavior of AI models with the actual driving behavior of users to determine which model performs better.

"The biggest constraint [currently] is not training data but rather [testing] the efficiency of the AI models," Musk said. "And how to cleverly determine whether a new model is better. For instance, we know some intersections are very tricky...all models do well on nice roads without complex intersections. So, we picked several thousand complex intersections in the US specifically to test the new model's performance."

Musk also mentioned that the full potential of the Hardware 4 self-driving hardware has not been fully utilized and plans to upgrade it later this year.

In summary, resolving the teeter-totter problem and rapidly iterating FSD models will be important challenges for Tesla to achieve full self-driving and self-driving taxis.

Edited by Jeffrey