Elon Musk misrepresents data that shows Tesla is still years away from unsupervised self-driving

Elon Musk is praising data that he claims shows Tesla is on the verge of achieving unsupervised Full Self-Driving, when in fact, it shows it is still years away and he is misrepresenting it.

It’s hard to take Musk seriously when it comes to self-driving timelines because he has been so consistently wrong for years.

Some argue that you can’t hold that against him, even though he uses his claims to sell cars and sell “Full Self-Driving” packages for up to $15,000, because it is such a difficult and important thing to achieve.

Even if you agree with this argument, there are clear problems with Musk’s claims regarding Tesla’s progress and timelines toward unsupervised self-driving.

The biggest one is data.

Tesla has consistently refused to share any data regarding its self-driving progress. That’s despite more recently starting to use “miles between necessary disengagement”, sometimes called “miles between critical disengagement”, as a metric to track progress and claiming x multiplicators in miles between critical disengagement in recent updates without any actual data to back it up.

A recent example was Musk hyping Tesla’s Full Self-Driving (FSD) software updates 12.4 and 12.5 by claiming they will be able to drive “5 to 10x more miles per intervention“.

Again, Tesla never released any data to back this up, but we have some crowdsourced data that pointed to FSD 12.5 achieving 183 miles (all versions combined excluding testers with fewer than 50 miles) between critical disengagement. Musk never specified the “5 to 10x” improvement was compared to what version, but if we compared it against the last update, FSD 12.3, miles between critical disengagement went down from 228 miles.

There are no prior versions of Tesla FSD over the last 3 years that would add up to a 3x improvement in miles between critical disengagement. We can forget about “5 to 10x.”

Now, Tesla has upgraded to FSD v13 and Musk again claims that it will “blow people’s mind.”

The automaker claimed that v13 would bring “5 to 6x improved miles between necessary interventions” compared to v12.5.

That means that Tesla anticipated FSD v13 to achieve between 915 and 1,098 miles between critical disengagement based on the crowdsourced data.

Currently, after over 8,000 miles of crowdsourced data, FSD v13 is at 493 miles between critical disengagement:

That’s a 2.7x improvement. It’s significant, but it is also a significant miss compared to what Tesla predicted.

Now, Elon and Tesla fans like to claim that this crowdsourced data is flawed and that Tesla FSD is actually performing better.

The data is indeed limited, but it is the best we have by far since Tesla refuses to share its own data. I have often fought against this accusation both because it is undoubtedly the best data available and because Elon Musk himself referred to this specific crowdsourced data in the past.

Now, he has done it again and he did it to claim “exponential improvement” in Tesla’s FSD performance, but it is grossly misleading:

This data only refers to highway miles and Tesla has been operating the same highway stack for years. The city driving software stack is different and based on “end-to-end neural nets”. The automaker kept promising to update it, but it barely ever did – leading to the stagnation you see in this chart.

Tesla worked on this update for years, but it actually wasn’t released in v13. It came in v12.5.6.1. If we take all the v12 updates after this one, the average on highway was already 393 miles:

This is no indication of “exponential improvement”. It is merely Tesla finally releasing a long overdue update to its highway software stack after working on the city software stack for the past 2 years.

Furthermore, if we can take this acknowledgment from Musk that this data is representative of Tesla FSD performance, which should be the case otherwise it would be greatly misleading for him to share it, it shows that Tesla is still years away from achieving unsupervised self-driving despite Musk saying it will happen in “q2 2025”, which is months away.

Ashok Elluswamy, the head of FSD at Tesla, has previously stated that for Tesla to enable unsupervised self-driving, Tesla needs to achieve the average in miles per critical intervention “equivalent of human miles between collision,” which stands at 670,000 miles, according to NHTSA.

Therefore, based on this data shared by Musk, Tesla needs to go from 493 miles between disengegament to 670,000 miles between disengagement within the next 5 months.

Electrek’s Take

I’m no hater. I’m a realist. Without patting myself too much on the back, you have to give me some credit for predicting this with v13. After Tesla’s AI team released the planned improvements coming with v13, I reported that I could see it achieve close to “690 to 828 miles between critical disengagement.”

But then I reported that v13 would result in improvements but come short of that goal after v13 was delayed by a few months and then released with a somewhat dumb-down version.

Now, it ends up at 493 miles between disengagement. It makes sense. It is an impressive improvement, but it is also far short of what Tesla said would happen and still hundreds of thousands of miles away from what Tesla itself said it needs to be to achieve unsupervised self-driving.

Not only that, but Elon is now misrepresenting the data to claim Tesla has achieved exponential growth without no evidence whatsoever.

He is purposely only looking at highway data, which is misleading because the stack was barely updated for years.

I think it’s clear that Elon either lies about self-driving or he has no idea what he is talking about, which somehow doesn’t stop him from confidently making statements that happen to help Tesla sell cars. It’s not suspicious at all.

Again, I liked to point out that I believe that if Tesla was developing FSD in a vacuum without Elon Musk making claims about Tesla achieving unsupervised self-driving on x timeline, making “Tesla vehicles appreciating assets”, and then using this to sell cars and $15,000 self-driving packages, I think that Tesla’s FSd development would be celebrated.

Instead, it is vastly seen as a fraud by many people. That’s Elon Musk’s fault.