Tesla FSD 12.3.2 is now releasing wide in the US and Canada.
Elon Musk tweeted:
V12.4 is another big jump in capabilities. Our constraint in training compute is much improved.
Three significant improvements to FSD will roll out roughly every two weeks. Should be really shining bright by late April or early May.
In this video, we’re predicting that Tesla FSD 12.4 will be out in April during the free one month trial FSD for all Tesla’s in the USA. We reviewed the crowdsourced self-reported Tesla FSD 12.3 data from teslafsdtracker.com and predicting what are the upcoming capabilities for FSD 12.4.
V12.4 is another big jump in capabilities.
Our constraint in training compute is much improved.
— Elon Musk (@elonmusk) March 17, 2024
Most people still have no idea how crushingly good Tesla FSD will get.
It will be superhuman to such a degree that it will seem strange in the future that humans drove cars, even while exhausted and drunk!
Cars will take you where you want automatically, just like getting in an…
— Elon Musk (@elonmusk) March 29, 2024
Brian Wang is a Futurist Thought Leader and a popular Science blogger with 1 million readers per month. His blog Nextbigfuture.com is ranked #1 Science News Blog. It covers many disruptive technology and trends including Space, Robotics, Artificial Intelligence, Medicine, Anti-aging Biotechnology, and Nanotechnology.
Known for identifying cutting edge technologies, he is currently a Co-Founder of a startup and fundraiser for high potential early-stage companies. He is the Head of Research for Allocations for deep technology investments and an Angel Investor at Space Angels.
A frequent speaker at corporations, he has been a TEDx speaker, a Singularity University speaker and guest at numerous interviews for radio and podcasts. He is open to public speaking and advising engagements.
I have v12.3 and it is a lot better. But it stalled at a blinking red light that it should have easily handled as the equivalent of a stop sign.
Lets talk about disengagements. I just got the 12.3.2.1 FSD and am trying it. Without setting a destination it does not know where I am going, so I disengage so I can finish the drive. I am building trust in the system, so in scary situations I disengage some times. I disengage for dips in the road that it wants to traverse too fast (for me). Do these all count towards too many? On my first day of testing 12.3.2.1 there was only one disengagement for safety where the car seemed like would crawl into a trailer blocking part of a left hand turn lane. At this point I am very impressed with FSD 12.3.x. It is NOT FSD yet but is very impressive.
Clearly most of the comments are from people who know little about FSD’s capabilities and have yet to drive a Tesla with FSD beta enabled.
Think of FSD like teaching someone how to drive. They are going to make mistakes or be indecisive. But unlike sitting next to the newbie driver with no ability to correct their steering speed it braking, a Tesla driver can cause FSD to disengage by touching the brake (similar to most cruise control) or turning the steering wheel in whichever direction you want the car to go. I have had many FSD drives 12.3.2.1 where my model y took me from the house to the store or work without issue. And really, FSD is just going to get better and better, as the beta label implies:)!
Still level 2 self-driving. Reality is hard to accept for Musk fanatics.
I still don’t understand what happens if the car gets disconnected from the mothership compute. Does it forget how to drive? Even a second disconnect could be fatal.
The driving itself is done locally on the car’s FSD chip.
The mothership (Cloud compute) is used for new versions along with real-time road congestion, supercharger status… FSD would be massively unsafe if it required Cloud connectivity at all times.
If FSD sees a condition it doesn’t understand, it should ask for the driver to take over, then pull over and/or stop safely as possible.
FSD doesn’t, and has never, used remote computing.
If you look at the cause of a lot of critical disengagements many of them do not sound amenable to improvements in a neural net that has the intelligence of approximately a mouse (Tesla ~100 Tflops vs ~100 pflops for human brain) – needing more fundamental comprehension of the world and humans to make the required choices:
https://www.teslafsdtracker.com/
distance between critical disengagements (ie nearly accidents) with 12.3 is about 500km, so they need to prevent about 999 out of every current 1000 disengagement to get to even 1 per 500,000km accident rate typical of humans, which is probably still too low for regulatory acceptance.
While I would be surprised to see a one order of magnitude improvement in distance to critical disengagements for FSD 12.x over next year (eliminating 9 out of 10) I think three orders of magnitude improvement is probably not on the cards for current ‘mouse’ level hardware, rat or cat (10-20x bigger brain) is probably a minimum.