I tested Tesla’s latest ‘mind blowing’ Full Self-Driving v14

I’ve put over 200 km (125 miles) on Tesla’s latest ‘Full Self-Driving Supervised’ (FSD) v14 update, and I’ve gathered my thoughts in this article.

In short, Tesla FSD v14 is an incremental improvement to the automaker’s advanced driver-assist system (ADAS) and the most impressive Level 2 system available in a consumer vehicle today.

However, it is still far from what Tesla sold to car buyers: unsupervised self-driving.

I briefly tested FSD v14 in a friend’s car when it launched two months ago, but now that I’ve picked up a new Model 3 with HW4, I’ve spent much more time with it and can share more thoughts.

First off, for context, I was already a somewhat heavy user of FSD on my Model 3 with HW3, which is stuck on FSD v12 because the computer has reached its limits. I’ve been using it for years.

With v12, FSD was involved in roughly 80% of my driving – mainly on the highway. I’m a responsible FSD user because I always pay attention to the road and am ready to take control at all times. I do appreciate FSD taking over most of the driving tasks so I can focus on scanning the road.

It tried to kill me a couple of times, but it is something you sign up for when you are basically a beta tester.

I haven’t had a significant update to FSD on my HW3 Model 3 for a year now, as Tesla has focused almost all its efforts on updates to vehicles equipped with the newer HW4 computer and its Robotaxi pilot program in Austin, Texas.

There’s now virtually no hope that Tesla vehicles with HW3 computers (most cars up to 2024) will ever deliver on Musk’s promise of unsupervised self-driving. He has been talking about a computer retrofit to help, but that was almost a year ago, and Tesla still hasn’t offered anything to HW3 owners beyond transferring FSD to a new car with HW4, which is what I reluctantly did.

The latest update, available only to cars with the HW4 (AI4) computer, is called FSD v14, and CEO Elon Musk has been hyping it as “sentient” and “mind-blowing.”

After driving more than 200 km (125 miles) with the FSD v14 update (v14.2.1.25 to be exact), here are my thoughts:

Tesla FSD V14 Pros

We already reported on the new point-to-point features of FSD v14, which now enable you to go from parking to parking only on FSD under the driver’s supervision.

That’s a noticeable improvement that removed a lot of disengagements.

However, that’s more a direct capability added to the system rather than an improvement in actual driving performance, which I’m focused on here.

Coming from v12, I can say with certitude that v14 feels more confident. It drives more like a human. It feels less robotic in its actions. Acceleration and deceleration are smoother.

Obviously, that’s going to depend heavily on the driving mode you choose. Much like v12, I find that the ‘Hurry’ mode is the best suited for me, as I generally like to drive above the speed limit, but not high enough to get a speeding ticket, which the new Mad Max mode would rack up quickly

FSD v14 finally tries to go back into the right lane after passing to the left – although emphasis on the word ‘tries’ as there’s a significant con to that in the next section.

Tesla FSD V14 Cons

The obvious one is: it’s not what Tesla sold to customers, unsupervised self-driving, and I don’t see it becoming that any time soon.

I had plenty of interventions and a couple of disengagements over those ~200 km. Fortunately, none of them were safety-related.

Two disengagements were due to FSD not entering the right lane at the right time. I gave it a chance until the last second, but it kept making me miss my exits or have a car cut me off at the last second, which is not acceptable to me.

From v12, I miss the ability to directly adjust the max speed live with the scroll depending on the situation. I hate it when automakers remove capabilities instead of just adding options.

As I mentioned in the ‘pros’ section, FSD v14 now tries to go back in the right lane after passing someone, something that it would do only half the time for me in v12. However, it only “tries” to do it. Half the time, it flashes its lights and moves to the right before returning to the middle of the left lane for seemingly no reason, even though it appears safe to return to the right lane.

It often does this little dance once or twice before either giving up or successfully moving to the right lane.

Electrek’s Take

My take on FSD has always been that if it were developed by Tesla in a vacuum, it would mostly be celebrated as a great driver-assistance system.

However, the fact that Tesla sold this to customers as something that “has all the hardware to support full autonomy” and will increase in price as the software will improve, until you can eventually go to sleep in the car and wake up at your destination, has changed everything.

Now, not only was Tesla wrong about the hardware and the price, but we also have to compare it to what was promised: unsupervised self-driving.

As long as it is not that, it is a failure. There’s no way around it.

I have a theory on people who are impressed by Tesla FSD v14. Putting aside biased people who are invested in Tesla’s stock, I think people who are impressed fall into one of two categories:

1. People who, for whatever reason, are disconnected from Tesla selling FSD as a level 4, unsupervised self-driving system. And doing so since 2016.
2. Bad drivers.

There are also people who have never experienced FSD before. For those, it is obviously impressive, but personally, I expected a bigger jump from v12 to v14.

Many people are bad drivers, and they are OK with FSD doing strange things that might confuse or upset other road users. Personally, I can’t accept that. As Tesla makes abundantly clear (in court, less so in its marketing), I’m the one responsible for the car. Therefore, its driving reflects me, and I refused to let it negatively affect other road users.

As for the claim that “no other consumer vehicle can do what Tesla FSD can” and that it is “the best level 2 ADAS,” to me, those points are moot compared to: Does it actually deliver what Tesla sold to customers?

The answer to that is undoubtedly a big fat no.

Can it get there? Of course. On HW4? I have serious doubts.

FSD v14 is “feature complete”. It can perform all the tasks related to driving, but it can’t reliably cover millions of miles without human supervision. It also lacks the redundancy that you need in a system that doesn’t require human supervision.

Maybe it can do it in a small geo-fenced environment with remote monitoring, but that’s not scalable to the customer fleet.