There was a new version of Tesla's FSD that is being released.
It was put out to employees a couple weeks ago and is now being pushed out to customers. Over the next week, those with the latest hardware will have the new version.
The early reports that a few items were fixed. One big one is the allows the software to pull the care out of the garage, something the previous version lacked.
One of the promises is a 5-6 improvements between interventions. Obviously, it will take a few weeks of people using the software to know if that level was attained.
FSD 13 Being Released
This is not an incremental change like FSD 12.4 to 12.5. Instead, we are looking at a full upgrade. There is a reason Tesla segregates its versions in such a manner.
The quest is for autonomous driving. Depending upon the results of this version, we could see a major step forward.
We are seeing an acceleration of the pace of updates. This is likely due to the fact that Tesla's compute levels are increasing. The end of October saw 50K GPUS on their cluster being online. The entire 100K should be live by the end of the year.
It will be one of the largest clusters in existence.
Tesla is also receiving a great deal more data. We will likely see a new addition to the chart in upcoming quarterly stacks.
Here is the secret sauce.
FSD 12 was released early in the year, with some miles added during the first quarter. We should expect something similar when the chart is released on the earnings call in January for the 4th quarter.
The total miles, at the end of September, was around 2.1 billion. The key is almost 500 million were driven during the 3rd quarter. We are likely to see an increase in that number again this quarter.
This data is being fed into the larger GPU cluster, likely accelerating things in the future. What we are seeing NOW is the result of training done a few months ago.
Supervised Driving
Tesla is still operating in a supervised manner. This is not unsupervised self driving. However, it does provide something else that others do not have.
The software will work anywhere, even if the car was never in that area. If a human can drive there, so can the software. Of course, it might not be perfect, which is why this is still supervised.
The goal for Tesla is to keep increasing the time between interventions along with getting more data. With more miles being driven, through a combination of more vehicles on the road along with people driving more using FSD, this is going to accelerate through 2025.
Robotaxi Scaling
Robotaxis are getting a lot of attention
Waymo and Cruise have operations in a few cities. The problem for them is they are geofenced. That means there is only a certain location they are trained, via mapping, to operate. They cannot extend outside that area.
This is not Tesla's approach. The goal is to have the ability to drive in any area, once autonomy is achieved. This is being done without Lidar, something that was laughed at years ago. Now, some of the other players in the field are realizing that Lidar is not necessary.
Of course, we are back to the original problem: data.
Here is where Tesla has a huge advantage. We could see the quarterly miles driven on FSD top 750 million in the 4th quarter. Look at the chart again...it keeps growing larger.
Elon Musk predicted it would take around 6 billion miles driven on FSD before regulators would have the data needed to feel comfortable authorizing the technology on the whole. There is a good chance we will see 4 billion miles driven in 2024.
This would put the total at 6.5 billion.
We will see if that is enough to satisfy the regulators.