PianoAl
Active Member
I've noticed that when you get the pay attention to the road message, wiggling the wheel dismisses it only if you clearly look at the road.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
FSD v12.x (end to end AI)
- Thread starter Buckminster
- Start date
-
- Tags
- #fixphantombraking autopilot chill is not chill elon is an ass fix fsd extreme late deceleration fsd does not commit to go fsd does not maintain proper speed fsd hard and late braking fsd jerky driving fsd late deceleration fsd stops too early too quickly fsd stops too far behind line fsd tailgating in city streets
V12 doesn't seem to be very predictable. It's making independent decisions Everytime. In the long term that's the correct way forward but right now, at least for me, it's more stressful to use than V11 because I can't figure out where it will fail.
There are times it's been great ok an open road, going 5 over. Doing the same drive 10 minutes later and seemingly under the same conditions it's going 20 in a 35.
Right, a few weeks free trial is nothing new. My son bought a new MY last September and the free trial ran out in January, about 3 months if not exactly. I suspect that more extended trial was not only to encourage a possible purchase or subscription, but to obtain some needed FSD performance results from HW4 cars.
FSDtester#1
Craves Electrolytes
It did, did you bump your head?
We were chatting in September 2021 about FSD.
No wonder you want fsd to work so bad.
If it didn't, you would get lost in your driveway!!
For me biggest difference between V11 and V12 is I didn't expect zero disengagement drives using V11 but with V12 I do. 90% of my drives now are zero disengagement. A lot of course depends on the roads you drive. I have to admit though that I give V12 a lot of leeway to complete the drive.
I dont understand why the speed is so broken with V12. Even 12.3.1. Are they hand picking drivers who go under the speed limit? It's just odd. Instead of 5 under...I was constantly going 1-2 under while my speed was set to 8 over.
Lane changing seems worse than 12.3. It seems they need a trip planner. Left turn in 200 feet, sure...get in the far right lane.
Lane changing seems worse than 12.3. It seems they need a trip planner. Left turn in 200 feet, sure...get in the far right lane.
FSDtester#1
Craves Electrolytes
I am driving to my nearest service center and connecting to the wifi.
Last edited:
FSD badly needs HD map data. This is a major weak point right now...poor data into the NN just means bad decisions.
Definitely less predictable. I'm noticing the "random" in-lane offsetting a lot.
Lower predictability is the cost of moving to an AI based planner. In traditional code you can know, precisely, what conditions will make the car do something. Neural nets are more probabilistic in their behaviors ("The car will mostly stay in it's lane under normal driving conditions").
(I'm not exaggerating, I literally shared a video earlier in this thread of the car going on top of a double yellow line for no good reason)
They’re training it on Amish people that stole a Tesla. The behavior is about right.
To, you know, get back at Elon for sneaking into the farms and taking the ketamine meant for the horse.
fasteddie7
Active Member
It also doesn’t respect when I press the turn signal button every time. I’ve tried pressing the turn signal to get into the necessary lane to turn ahead of time and the car will often ignore it and turn the signal off. Crazy.
FSDtester#1
Craves Electrolytes
I am getting the weird green message about updating when the message turns off.
However this time, 60 miles from home and at other random times.
This used to happen at the start of drives. Now happening whenever.
I think the resolution of the map data is fine because it works the vast majority of the time. I'd guess that the problems lie in inaccuracies in the map data combined with issues in the remaining heuristic code - like the route planner.
My greatest concern with the move from 12.3 to 12.3.1 is that Tesla didn't knock the ball out of the park. There was some tuning to be done, and they didn't get it right. It's odd that their training process produced behavior that was so atypical of human drivers, and they weren't able to clean it up in a point release. Are they still learning about how to train the control network, or are they hacking around with heuristics in an effort to tune the network outputs?
FSDtester#1
Craves Electrolytes
Anyone notice how teslafi.com just dropped 12.3 like a bad habit from their main page. Only showing main branches, V11 and V12.3.1.
V12.3 got Munsoned
V12.3 got Munsoned
Last edited:
Last thing that happens frequently is wrong speed limit (shown a upper right of current speed). Maybe autospeed is implemented to somehow mitigate that... I have that switched off btw.
Usual scenario is that you take a right or left turn, and the speed limit stays the same whether it is higher or lower on the new road, until you pass a speed sign. As an example it stays at 25 mph for a few blocks on a 45 or 50mph road. Manual override with the right scroll wheel helps in the sense that it does speed up.
Since this was way better in v11 (except for some places where it would always be wrong, I assume map data error), is FSD simply ignoring the map data until it sees a speed limit sign?
Usual scenario is that you take a right or left turn, and the speed limit stays the same whether it is higher or lower on the new road, until you pass a speed sign. As an example it stays at 25 mph for a few blocks on a 45 or 50mph road. Manual override with the right scroll wheel helps in the sense that it does speed up.
Since this was way better in v11 (except for some places where it would always be wrong, I assume map data error), is FSD simply ignoring the map data until it sees a speed limit sign?
powertoold
Active Member
Curious decision to make 12.3.1 the ChatGPT moment. If there will be significant improvements within the next 1-2 months, why not wait until after those?
Seems like a more reasonable approach from discussions among the teams of how to achieve the intent of getting people to experience 12.3.1. Hopefully this means it'll be quickly merged to 2024.x supporting more hardware/software configurations and widely deployed as most US vehicles currently can't get 2023.44.30.30. Giving a free month trial of 11.x could have the opposite result.
Do people think this is based on metrics/evidence that 12.3.1 is significantly better and safer? Or is this in preparation/expectation of future improvements (before the trial ends)?
Similar threads
- Article
