FSD v12.x (end to end AI)

[rlsd]

It sounds plausible, so I offer this tiny datapoint.
My 2018 3 got FSD 10.3 way back when, this year I kept the 3 and transferred FSD to a new Y.
Soon after, the 3 ended up on 2024.14.7 and the Y on 2024.3.25
About a week ago, the 3 got 2024.14.9 and yesterday the Y also got it.

2024.20.1/12.3.6 is active again now.

 

[Mardak]

2024.20.1/12.3.6 is active again now.

2024.14.6 introduced 12.3.6 for those who were still waiting with 2024.8.x / 11.4.9. A big wave of trials started for this group around May 15th, so a lot of this group now no longer has an active FSD purchase/subscription/trial and are getting updated to 2024.20.1. The previous rollout for 2024.20.1 was June 12th presumably targeting vehicles that do not have active FSD, and I would guess that same rollout check was applied today now catching most who did not continue with FSD after the 1-month trial ended.

Basically unless you have the video snapshot button for true early access updates, FSD active or not seems to determine if we're "held back" so that we can get 2024.15.x / 12.4.x when that's ready for wider release.

 

[AlanSubie4Life]

The Tesla ADAS drivers are back in Chuck's neighborhood:

https://twitter.com/x/status/1802787600438120880

Chuck first spotted them testing what might have been 12.4.1 on May 23 and got it himself on June 7. So maybe 2 weeks from today we'll see 12.4.2.

I think with prompt engineering they can probably get a clean pass for the first 9 with 12.4.2 in moderate traffic. Just needs to be faster; look for it in the overhead videos if we get them. If it isn’t rocketing across there’s going to be little hope of not impeding other traffic on occasion and getting the multiple failures 6/9 to 8/9 score (normal scoring vs. generous “a***ole” scoring) of 12.4.1.

I’m much less clear on whether they’ll be able to deal with unusual situations with heavier traffic; hard to judge without seeing its decisions. And the tricky cases may only show up in the second 9.

There’s also a very real danger of its uncertain creep behavior resulting in disengagements. Still leaves a lot to be desired. Has to be a very binary “stay” or “commit” (except if there is a surprise!) which is tough, or perhaps impossible, for NNs.

I don’t know that I would betanymore against success. But my winning streak discouraged any takers in any case. Too little confidence in the software team perhaps?

 

[AlanSubie4Life]

Umm… wow 12.3.6 hit a cone and ripped off the garnish:

View attachment 1057207

I assume this was not your vehicle! This seems like it would be super difficult to accomplish. Video?

 

[Twiglett]

2024.20.1/12.3.6 is active again now.

Yep. The 3 was part of the remaining 1% of the fleet still on 2024.14.6 before it got 14.9, but within the last hour it was offered 2024.20.1.
It got to come back onto Wi-Fi to download it though.

 

[jgleigh]

No this does not work. It’s permanent.

Yeah noticed this today. You can actually run it in full-screen all the time even when FSD isn't active. If it's full-screen, then the menu will slide it out of the way and go back to full-screen when the menu goes away.

I think the only automatic return to non-full screen is when FSD is disengaged or you manually slide it back to the left.

 

[OldManCan]

In reply to Chuck's post shared above

 

[JulienW]

View attachment 1057327

In reply to Chuck's post shared above

For Elon that sounds between pessimistic and so so. Like he said it will probably be a few releases before we get it.

 

[Yelobird]

I don’t know that I would betanymore against success. But my winning streak discouraged any takers in any case. Too little confidence in the software team perhaps?

Have you considered there are no takers not due to buying a beer but possibly due to the company they would have to share it with?? lol. As for your winning streak, as your detail of “rules” are so specific I don’t think any version could win even if all cars were stopped.

 

[sleepydoc]

I've never seen it be so off as to be a risk of collision during a merge. But I also try to avoid merging into super tight gaps too, and avoid using only the visualization because it doesn't show rearward far enough for my taste.

IME the position of cars in the blind spot is routinely off by 10-20 feet with the car actually being farther forward than depicted on the display. Plenty of error to cause an accident.

 

[swedge]

I'm driving a 2023 MYLR bought in October 2023. Originally on EAP and now subscribing to FSD on month to month basis since the free 12.x trial expired at the end of April. I had got one month subscription in December '23 on FSD 11.x and had dropped it afterwards.

With all that background above, I am also moved from 3.25 to 14.9 so I have a feeling they may be keeping the earlier testers + newer paying customers like myself on the same pool going forward. We can validate this once they release next FSD drop and see if we all get it at the same time or not.

Time will tell. But 2024.3.25 was given to FSD users as well as to testers. The testers were held till just before noon on 6/11, when all 5,000 or so started getting moved to 14.9. Prior to that, many more already were on 14.9 and thousands of those had already moved on to 2024.20.1.

It is hard to tell from the TeslaFi site who is using FSD, or who is in the tester group. It appears that some FSD users have recently been getting the same updates as testers, but since everyone is currently being moved to the same FSD version, 12.3.6 now, we all appear to be in the same boat. Even the next release may not distinguish testers from all FSD users.

For all I know, the tester group may have been dissolved and merged into the FSD branch. But if so, why are we still on 14.9? We'll see...

 

[Electroman]

In a lot of ways I think Chuck’s UPL is much easier than many UPLs I have seen in DFW area.

Here is one where the current version (12.3) struggles to cross even in moderate traffic. Waits for traffic to ease completely on both sides, but that means waiting forever most of the day.

 

[zoomer0056]

IME the position of cars in the blind spot is routinely off by 10-20 feet with the car actually being farther forward than depicted on the display. Plenty of error to cause an accident.

That error is directly related to the speed difference between the ego and the other car. At any rate, or differential rate, a vehicle in the visualization warants a direct check with eyeballs. Visualization is an aid only, not a replacement for a skilled and attentive driver.

 

[Electroman]

That error is directly related to the speed difference between the ego and the other car. At any rate, or differential rate, a vehicle in the visualization warants a direct check with eyeballs. Visualization is an aid only, not a replacement.

Agreed. That is why I would say that Tesla has very poor blind spot detection and dissemination. While other manufacturers have a light in the side mirrors that comes on when a car is in the blind spot, the Tesla onscreen visualization cannot be relied upon to make decisions on lane changing.

 

[FSDtester#1]

Time will tell. But 2024.3.25 was given to FSD users as well as to testers. The testers were held till just before noon on 6/11, when all 5,000 or so started getting moved to 14.9. Prior to that, many more already were on 14.9 and thousands of those had already moved on to 2024.20.1.

It is hard to tell from the TeslaFi site who is using FSD, or who is in the tester group. It appears that some FSD users have recently been getting the same updates as testers, but since everyone is currently being moved to the same FSD version, 12.3.6 now, we all appear to be in the same boat. Even the next release may not distinguish testers from all FSD users.

For all I know, the tester group may have been dissolved and merged into the FSD branch. But if so, why are we still on 14.9? We'll see...

I think the tester group will remain who came late from 3.25. Tesla probably doesn't want someone who just purchased a new car to be in an early release version of the software from a liability perspective. 12.3.6, while it has its faults, is pretty stable and predictable.

 

[zoomer0056]

Agreed. That is why I would say that Tesla has very poor blind spot detection and dissemination. While other manufacturers have a light in the side mirrors that comes on when a car is in the blind spot, the Tesla onscreen visualization cannot be relied upon to make decisions on lane changing.

Yup, I don't think the visualization was meant to be blindspot detection. This thought is supported by a blind spot ribbon on the video when turn signals are on and the light in the A piller speakers of the Highland. Those indicators should be used primarily instead of visualization vehicle. And, I bet the visualization color of the vehicle beside the ego synchs with the Highland A piller light and the turn signal blind spot camera ribbon.

 

[AlanSubie4Life]

I don’t think any version could win even if all cars were stopped.

Wow so little faith.

Now we know who the true believers are.

I am confident that FSD will be able to make these turns to one 9 of capability within the next couple releases.

The rules are completely reasonable. It’s basically just make the turn with traffic present and don’t cause anyone to take evasive action (and without freaking out the DIC enough to result in disengagement of course). How much simpler or logical could it be?

 

[rlsd]

Yup, I don't think the visualization was meant to be blindspot detection. This is supported by a blind spot ribbon on the video when turn signals are on and the light in the A piller speakers of the Highland. Those indicators should be used primarily instead of visualization vehicle. And, I bet the visualization color of the vehicle beside the ego synchs with the Highland A piller light and the turn signal blind spot csmera ribbon.

This is how I use it:
1. With manual driving I use the VS for secondary check. If after checking the mirrors and starting lane changing, I will move back to the original lane when seeing a red car coming up.

2. With FSD, a lot of times I want to change lanes to my liking. Before pulling the stalk I check the VS. If there are close cars I will not pull the stalk. I don't want the blinker keep blinking but the car cannot change lanes.

 

[sleepydoc]

I'm still on 3.25, so I'm one of the special 3%'ers

no, you're just 'special!'

 

[jgleigh]

IME the position of cars in the blind spot is routinely off by 10-20 feet with the car actually being farther forward than depicted on the display. Plenty of error to cause an accident.

There's also a camera blind-spot between the side cameras and the front camera. I had someone merge into me while I was turning left and I reviewed the camera footage and the car isn't anywhere to be seen until if appears in the front camera.