Last year, I drove one of GM’s cars, a Cadillac Escalade, from San Francisco to Park City. The 766-mile drive was the perfect laboratory for Super Cruise. It not only made the drive less taxing, but shaved roughly a couple of hours off the total drive time by keeping a more consistent average speed.

I mentioned then that allowing drivers to take their eyes off the road would be a game-changer. Assuming it works, it is a major win for GM.

Even though I would rather drive GM’s new Escalade on a long road trip, it’s an unavoidable truth that Tesla’s “Full Self Driving” system is more advanced and may in the long run pull ahead of GM, and possibly even Waymo.

Early on in Musk’s robotaxi efforts, he claimed the company would have 1 million of them by 2020, a wildly ambitious timeline that is now much ridiculed. The path to Level 5 autonomous driving has turned out to be a much longer and bumpy one than the optimists once hoped.

Ironically, the stretched-out timeline could end up being an advantage for Tesla. Musk’s strategy from the beginning was to build an autonomous driving system powered by a neural network and cameras.

Waymo and other self-driving startups used a combination of cameras, lidar, and other sensors and built a complex network of fail-safes, detailed mapping, and human oversight.

The result of Waymo’s efforts is a robotaxi service that is a technological marvel, but one that requires human labor to expand it to new locations and maintain operations. The cars sometimes require remote operation when they encounter a situation that stumps the onboard computers.

Musk hasn’t achieved his goal of building a Level 5 car that can operate virtually anywhere, in any condition, without human intervention. But with recent advances in AI, it appears that he might have chosen the best path to that goal, even if his timeline was off.

It’s looking increasingly likely that lidar and other sensors are not necessary and that cameras may be the best sensors. And all those cameras on Teslas have amassed one of the largest, if not the largest, datasets for robotics today.

A Tesla equipped with the latest FSD is already safer than the average human driver. But that isn’t good enough. A single fatality would be politically unacceptable, and it’s unclear if or when we’ll get to that level of reliability.

The hope is that we are a handful of breakthroughs away from AI models that can leverage a massive dataset like Tesla’s while eliminating the unpredictability that makes today’s models impossible to deploy without human oversight.

Musk is betting his company on reaching that goal, which could enable Level 5 driverless cars, or robotaxis, and possibly even humanoid robots. Most CEOs would never take such a big risk because they would look foolish if they failed.

That obviously is not a concern for Musk. In some ways, the question has always been about the patience of the company’s investors. Will they stick around if there still isn’t a Level 5 car in five years? How long does the company really have to make good on its humanoid robot promise?

That’s an unanswerable question at the moment, but it is certainly a lot longer than Musk’s harshest critics.