Elon Musk has claimed that it will take somewhere around 6 billion miles driven autonomously before there was enough data to give the regulators what they needed. The goal is to prove that autonomous driving is orders of magnitude safer than humans.
How accurate is this number? That is up for debate. Is it 6 billion? 10 billion? 15 billion?
Whatever the number, Tesla will soon be approaching it.
There was a lot of discussion about Tesla offering a free trial of the FSD software to all vehicles in the US and Canada (that have the ability). Was Tesla trying to generate sales? Are they laying the groundwork for robotaxis? Is the company desperate?
The answer might lie in the data.
We cannot be sure if data alone will be the barometer that solves autonomous driving. However, we were told that Tesla is no longer compute constraint, which means it can process all that it collects.
Here is where the trial of FSD comes into the picture.
Fortunately, we have a visual along with some metrics which might help us gain some insight.
Here is the path of autonomous miles driven on FSD.
@taskmaster4450le/re-leothreads-fjsdexrf
Notice how the last part of the chart is now going hyperbolic. In fact, it is almost straight up. The path from 900 million to 1 billion happened in a very short period of time. We also have to note the elbow in the curve, around September of last year. Hence we are about 7 months from the ~125 million level.
We know the past doesn't equal the future. If this continues, it could mean that the quest for autonomy is going to come upon.
How fast are things moving? Let us look at what this trial could provide.
Tesla could add another billion miles on FSD over the next 30-60 days.
The trial involved roughly 2 million miles according to the lead AI manager at Tesla.
How does this look if we apply real world numbers to it?
We know the trail is for 30 days so there will be a drop off after that. How much data is captured during that period (in miles) is going to be interesting. We already see how the chart was going straight up before this.
If this forecast is correct, and another billion miles are added how would this chart look?
Here is a mock up of it.
This is what happens with hyperbolic curves. As latter iterations are added, the earlier ones of growth flatten out. A move from roughly 125 million to a billion in 7 months was impressive. The chart will look much different if a billion miles in 2 months is added.
Here is the multi-trillion dollar question: is this the path to autonomous driving?
So far, much of our AI revolves around the idea of ever growing data combined with massive amounts of compute. One thing regarding these charts that we have to keep in mind, the growth period through March was during a period where Tesla, by their own admission, was compute constrained.
If this is the path, Tesla will win this hands down.
As a comparison, according to Databricks:
As of December 2023, Waymo, the autonomous driving technology company, had driven over 20 million miles autonomously on public roads. This is the most recent data available, as your knowledge base was last updated in December 2023.
20 million versus 1 billion?
It is no contest.
As a side note, whatever the path to autonomous driving, this is a fantastic lesson in exponentiality. This is something that is no native to human thinking yet it is standard in the world of technology. As we delve deeper into the digital world, this is going to be an important part of the understanding of how things will unfold.
If data is the golden nugget for cars driving themselves, hand the ribbon to Tesla. The exponentials are all in their favor.
