Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register
This site may earn commission on affiliate links.
I care much more about the basic behaviors than all the failures and amazing moves it makes. Who cares if it can get over four lanes quickly after a turn in no traffic?
At least I'm interested in these rare maneuvers because it shows the neural network has the capacity to consider many and/or subtle things to then control appropriately. The hope is that it's then able to use that capacity to handle "easier" stuff that is much more common such as stopping, and more training should continue to improve all aspects including basic and amazing behaviors.
 
  • Like
Reactions: JHCCAZ
At least I'm interested in these rare maneuvers because it shows the neural network has the capacity to consider many and/or subtle things to then control appropriately. The hope is that it's then able to use that capacity to handle "easier" stuff that is much more common such as stopping, and more training should continue to improve all aspects including basic and amazing behaviors.

We might fall into the same AI hype trap that's going on right now.

We have these AI chatbots that output amazing things but also constantly fail at obvious simple concepts.

The major problem with the current NN AI approach is that the NN doesn't really understand the underlying principles.

Our only hope is better compute and more parameters, but HW3 seems dead on that front.
 
It looks like medians are causing some trouble. We typically drive on the right side of medians.
Another instance of median induced confusion?
Yeah, it seemed pretty clear that these "medians" caused some confusion about the expected direction of traffic flow. This was something Tesla heavily trained networks to predict back in 2020 but seems to have potentially been thrown out and relearned from scratch for end-to-end. Hopefully dedicated training on parking lots in preparation for smart summon will help the new 12.x neural networks understand these lane/median structures without requiring explicit training data labeling flows, and more importantly avoiding regressions on city streets when particular parking lot behaviors would be inappropriate.
 
  • Informative
Reactions: kabin
100x intervention reduction over in 11.4.9 is a tall order Seems impossible based on my experience with V12.2.1
To be clear, he said "V12 is going to be," so you don't think some later 12.x will eventually get there? If 12.3 was able to smooth out speed control to be consistent on empty roads and following traffic, would that change your expectations? How about if those weren't fixed until 12.4?

From what I can tell from 12.x videos, there are examples of consistent speed control, so the neural networks are able to control that way, but maybe the focusing on problematic 11.x situations has resulted in 12.x getting confused about the easy stuff.
 
  • Like
Reactions: powertoold
. No one is crazy enough to use v12 in parking lots, so who cares how it performs there, but maybe parking lots are a good training ground for the FSD team for figuring out how to avoid randomly wrong or indecisive behaviors.
Funny you say that because I've seen a lot of videos with parking lots included. Far more videos relative to the small number of testers and for the most part V12 parking lot driving is significantly better than V11 based on what I've watched and driver feedback. And yes I know all about the drama related to the parking lot accident.
 
  • Like
Reactions: zoomer0056
I can say definitively that it does not pass the wife test yet. My poor wife has to suffer FSD everytime we go anywhere and invariably the car does at least one thing that makes her jump or yell. From my viewpoint as the driver they aren't much of a problem, I know I can grab back control anytime it goes too far, but these moves jitter up passenger nerves every time. They are much less frequent than V11 but even once makes passengers distrust the car.
You'll just have to ask your wife to sit in the back where passengers don't seem to pay as much attention. (Disclaimer here- I cannot use FSD on city/streets with my wife in the car). :)
 
Yes, but I feel a lot less safe without FSD. That's what I tell my wife, but she probably doesn't believe me :)

Like right now, my cameras are stuck at 95% FSD calibration, and I get a lot more anxiety and sensory overload driving without FSD.
Absolutely. I tell my wife that even if FSD occasionally does startling or annoying things, it is still a better driver than I am. I'm sure that's the truth and I think she, reluctantly, agrees.
 
small number of testers and for the most part V12 parking lot driving is significantly better than V11 based on what I've watched and driver feedback.
I can totally understand people posting for the engagement because this was not something v11 could really do. And that is fine.

But it just is not somewhere you’d ever want to use FSD, since parking lots are one of the most dangerous places for vehicle damage (they are perhaps the most complex environment), and FSD is certainly not going to help in that regard. Useful (potentially - I have no idea) for the reasons mentioned earlier.
 
Last edited:
FSD users are being moved to 2024.2.7. So, it looks like the next V12 version will be on 2024.x.
From NotATeslaApp
1000028117.jpg
 
  • Informative
Reactions: sleepydoc