Skip to main content

Tesla FSD - Not Stopping At Better Than Average Human Driving - It's Going For Superhuman Driving: What This Means

Tesla's FSD is not going to stop at being better than the average human driver. It's going to be superhuman one day. This is what that means.

Tesla FSD - Going For Superhuman

Tesla's FSD software is not going to stop at being better than the average human driver, as stated by Elon Musk on X recently. To understand what this means, let's take some things said by Andrej Karpathy, who used to work on Tesla's AI team for FSD and Autopilot. All of this will tie into FSD and how it becomes superhuman.

There are three key things here to look at for AI:

  • Imitation Learning: Learning by studying humans
  • Reinforcement Learning: Learning by training itself
  • Superhuman FSD: What behaviors superhuman FSD will exhibit

Imitation Learning

Andrej Karpathy talked about AI learning and training and that AI models aren't good enough yet and that people haven't seen what is possible in the AI space yet.

He says that we have done "step 1 of Alpha Go". This is imitation learning. There is "step 2 of Alpha Go" and this is a huge step forward. Alpha Go is a board game that AI eventually beat the best players in the world at.

"Step 2 of Alpha Go" uses "RL", which is reinforcement learning and this means to optimize model behavior through rewards and punishments, while a model is doing trial and error. In this sense, the AI is training itself.

Reinforcement Learning

Data collection isn't great right now. For something like ChatGPT, humans are filling out data and polluting models with data.

What you need is for the model to practice on its own in order to solve problems. The model needs to learn for itself and right now, this isn't working well.

People need to look for better ways to train models and data. Reinforcement learning is the way.

There needs to be a type of learning that is not tied to humans. Imitation learning imitates humans. This can get you to human level and even above average human level by using video data from above average human driving. But, you are capped with this type of learning.

You May Also Like: Tesla To Unveil Robotaxi on 8/8 - According To Elon Musk Directly Posting On X

What Superhuman Tesla FSD Means

What does all of this have to do with Tesla's FSD? Tesla's FSD is going to drive better than the average human by using video clips from human beings doing successful. But, there will eventually be a cap or ceiling to this. Let's call it 10 times better than the average human driver.

To me, superhuman driving is 100 times or better than the average human driver. The driver must do things that nobody thought to do. It must cause people to say wow and be surprised positively. It must see what we cannot see and do what we cannot do, all in the name of safety and efficiency.

FSD, up to this point, is likely using imitation learning solely. This data is curated and pulled in to allow the car to drive itself. The neural network imitates good human drivers, just like the original Alpha Go AI - it imitated human players to learn how to play.

Tesla FSD can get as good as the best human drivers using this approach. It also won't get distracted, won't get drunk, and will have 8 "eyes" always paying attention. This is wonderful, but how does it get SUPERB and superhuman?

Tesla FSD needs to have a way to do reinforcement learning. Tesla might be starting to do this internally. For a car to be 100 times better than a human driver, you have to start with a synthetic data landscape and start driving in a new virtual world with the ability of the very best human drivers.

You reconstruct this virtual world with all the attributes of a real world and the best human drivers. You create all the scenarios possible in the real world and allow these drivers to compete and drive. Certain things that would cause a driver to be the best would be:

  • Complete Safety - no collisions with anything
  • Speed - how fast a destination was reached
  • Energy Saving - how much energy was conserved

These are only 3 attributes of success, and there are likely more. These "virtual drivers" would compete against each other in the synthetic world to see who does the best.

This is what allows Tesla's FSD to do things that aren't thought of yet. All drivers will follow existing laws in this scenario, but I can imagine a few scenarios of superhuman driving to be like so:

  • A driver is able to change lanes at just the right time in the most efficient way
  • A driver drives slightly on a curb to avoid an oncoming vehicle
  • Turns are taken perfectly every time
  • Acceleration from stop to the speed limit is done in the most fast and efficient use of energy
  • A car weaves in the road perfectly based on animals and people that may be crossing

Until we see a car driving at a superhuman level, we can only speculate with our human brain what it might do. When Tesla tests out the best human drivers in a synthetic world with the criteria of success and takes the winners, we'll see the car doing things nobody thought possible - and it will do it safely, quickly, and with an efficient use of battery energy.

For Further Reading: Blatant Manipulation of Tesla Happens With False News Report That Tesla Is Scrapping $25K Compact Car - Elon Confirms It Is NOT True

What do you think about Tesla FSD being superhuman? Will this happen soon?

Share this article with friends and family and on social media - or leave a comment below. You can view my most recent articles here for further reading. I am also on X/Twitter where I post more than just articles daily, as well as LinkedIn! Thank you so much for your support!


Hi! I'm Jeremy Noel Johnson, and I am a Tesla investor and supporter and own a 2022 Model 3 RWD EV and I don't have range anxiety :). I enjoy bringing you breaking Tesla news as well as anything about Tesla or other EV companies I can find, like Aptera. Other interests of mine are AI, Tesla Energy and the Tesla Bot! You can follow me on X.COM or LinkedIn to stay in touch and follow my Tesla and EV news coverage.

Image Credit: Tesla, Screenshot

Article Reference: Tesla