About 25% of the fleet is on 2024.8.7, so I expect Tesla to release an update for them soon. That will provide an upgrade path for just about everyone.
-
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
fasteddie7
Active Member
I’d imagine they are working diligently to get it to everyone and are probably anxious for people to try it. This is the first version they can legitimately be proud of.
mgs333
Member
If we believe what Tesla says on the in car screen where we can choose the profile, those profiles (chill, average, assertive) only affect the following distance and how often speed related lane changes happen. Not how fast it accelerates or brakes, this matches my experience.
Not victimized, and really don’t care, haven’t even driven my “FSD” car in like 3 days lmao. Just pointing out that this is another process that Tesla somehow manages to half ass.
I have FSD 12.3. I know many like the fact it accelerates fast from a stop, but I would like it to accelerate from a stop a little less aggressivly. . I am more comfortable with that.
fasteddie7
Active Member
For me, proper speed on secondary roads, better lane choices, and a solid first attempt at vision only auto park are what I’ve seen. It’s definitely focusing more on the eyes and head than hands on the wheel now.
Musskiah
DisGruntled
@Discoducky I (and others) would be grateful, if you feel so obliged, when you have a moment to please further expound on your robotaxi thoughts you began on the Perpetual Investor Forum...
mgs333
Member
Please keep the investment stuff in the investment thread. Thanks.
If you're looking at TeslaFi numbers, most of those on 2024.8.7 are outside of North America. 2024.3.x already has more vehicles on it than US/Canada on 2024.8.x. Yes, Tesla will need to prepare a new version of 12.x that these vehicles can update to, but they're closer to 5% of TeslaFi fleet (which has a US bias). It seems like those who could get 12.x but were already on 2024.8.4 ended up getting 2024.8.7 for the security fix, so Tesla might be preparing a 2024.8.10 for these vehicles.
Tesla does have other vehicles to get 12.x to as well such as new Model 3s and Cybertrucks as well as portrait S/X including those with MCU1. And this is on top of working towards updating Smart Summon and Banish for parking lots as well as preparing for China, Europe, etc. and maybe also other features like skipping camera calibration. And there's probably now a bunch of cleanup to combine the various differences in all these 12.3.x.
AlanSubie4Life
Efficiency Obsessed Member
The issue is the time scale. It’s been a month or two since the initial 12.x releases to the public (OG testers)?
With NN training and massive NVDA capability, it was widely stated we’d have rapid improvement of outstanding issues.
But instead nothing has happened over that very long timeframe, and the OG testers claimed it got worse.
So I have to conclude we’re going to be stuck with slow iteration with minimal resolution of issues from release to release.
We may even see things fixed then come back, which is even worse.
We don't even know yet since we haven't seen a single iteration of attempted improvement yet. (And if we have seen an attempted improvement as Elon claims (I generally don't think his claims of what is changing from release to release are reliable, so best to ignore) ...yikes.)
It seems like it was able to see this car just fine.
There's going to be no change. Tesla is laser focused on maximizing revenue for Q1. They don't care if there's no revenue left in your vehicle - they'll worry about that next week. So this is some way to enable more Tesla revenue (not sure what exact permutation it is).
They're currently laser focused on rolling out versions which take 40 seconds to park the vehicle in trivial situations, since that brings in the big money.
Maybe one should only expect actual changes when we go to 12.4.x. Perhaps that will be our first datapoint. Maybe with a one-month cadence if we're lucky? Everything else is just nothing.
Last edited:
Have we seen 2 week improvement cycle with 12.x so far? It sounds like you're experienced with this type of Elon Musk mandate of getting more customers to experience 12.3.1 from less than a week ago, and it did result in a lot of activity of various 12.3.x released this past week. I would think that sudden reprioritization would in the near term delay 12.x improvements, but presumably with more customers actively using FSD, it should allow for even better data collection going forwards.
Assistant to the ASSistant regional manager
To take liability I think it needs to be safer in every situation and that anything less will be problematic.
Overall mileage will be an interesting metric but I’ve often thought they would end up breaking it down into more granularity by road type, maneuver, and who knows what else. It won’t work if the system is significantly safer on one road type or certain situations but much less so in specific situations, and overall mileage wouldn’t provide that detail.
Like large complex intersections or intersections with odd geometry, that were mentioned back in 2020. To this day, I’ve seen clips of v12.3 still incorrectly assigning lights to its lane and making the wrong moves because it thinks a red or (more commonly) green light is controlling its lane but it’s actually for an adjacent lane — we saw this in Elon’s demo of v12 as well. Liability wont be taken if the system can handle 100% of straightforward perpendicular intersections with lights all conforming to what the system is trained on but then it incorrectly hits the accelerator in a large complex intersection or intersections where the lights are hung in less ideal locations relative to the lane positions.
I'm on 2024.3.6 with FSD Beta v12.3.2.1
I see 2024.3.10 out there with FSD (Supervised) v12.3.3 which I have not received yet
Is there any information on what has changed between 12.3.2.1 to 12.3.3 ?
Thanks
I see 2024.3.10 out there with FSD (Supervised) v12.3.3 which I have not received yet
Is there any information on what has changed between 12.3.2.1 to 12.3.3 ?
Thanks
data point
just got software upgrade and 30 day (supervised) FSD!
just got software upgrade and 30 day (supervised) FSD!
Good work! How did you locate that intersection?It's not a dedicated right turn lane (it merges with into another lane and it's also not a right turn only lane). Admittedly it does seem like there should be a green right turn arrow there.
Google Maps
Find local businesses, view maps and get driving directions in Google Maps.www.google.com
I was a bit surprised that FSD would know that dedicated lane carve out in Calif law. I don't expect it is allowed in all states.
It looks like you are correct, far easier to see from clear daylight satellite images than the rainy night dash-cam. Being a T intersection, there is no through traffic from the left, but opposite direction, left turn traffic could pose a conflict during a dedicated through and left turn phase of the light.
It is a good thing Tesla hasn't released rob taxi yet, or the yet to be announced rob-cop capability, or else the car could have issued itself a robot-ticket for not stopping. ;-)
Similar threads
- Article
