FSD v12.x (end to end AI)

[Mardak]

Apparently unnecessary lane changes has gone worse, otherwise it is better

https://twitter.com/x/status/1799332647241662853

Decision wobbling / lane indecisiveness seems difficult to solve

The 6 lane changes seemed to be pretty confident in wanting to make the lane change as opposed to 12.3.x behavior of starting a lane change and second guessing if it should complete. This 12.4.x behavior would be consistent with additional training to make it more decisive in completing maneuvers, but in this case it seems to be confused by conflicting map data with numerous turn-only lanes both left and right that it was probably "seeing" both visually and from map data. The final lane change was necessary to get out of a right-turn-only lane, and potentially the closeness of these intersections resulted in mixed signals.

Presumably this comes from focused training on 12.3.x examples of where people had to correct the lane selection, and with enough examples, it's learning to rely more on input signals of upcoming mapped turn lane data. There probably is not as much explicit training on "don't make lane change" / "just stay straight" examples as that's the common case anyway, so this could result in the unnecessary lane changes that now needs additional training to figure out when lane changes aren't actually required. Unclear if this will require more examples from a wider deployment of 12.4.x to get this data as active usage of FSD should help find its current mistakes to fix in the next version.

 

[spev18]

Well it's morning and TeslaFi shows nota and Elon indicating there is a "hold in the countdown". So even this weekend is starting to look iffy. Hope we see 12.4.2 polished up and sent to influencers by tomorrow (Tuesday) so we can have some hope of getting 12.4.2 this weekend.

I see S5OD has the team working hard and they even have the new eye tracking glasses to calibrate the eye monitoring system.

Now that we're on yet another and new FSD 12.4.x delay, curious to see what happens with us 2024.3.25 folks. There's no reason to hold us back any longer on the spring update since the latest FSD is on the 2024.15.x branch, yet as of this morning we still sit at 2024.3.5 with zero updates.

On one hand I take it as a good sign we're not being moved up to 2024.14.9. meaning they expect to have the next revision of FSD out very shortly for testing so a few days longer isn't a big deal. If we do start seeing 2024.3.25 upgrade to 2024.14.9, that would be a bad sign in that we will likely be waiting weeks more for an FSD update that will get into the hands of us common folk so they're at least getting us on the spring update in the interim. That said, there is some nice stuff in the spring update I wouldn't mind playing with while we continue to wait.

Based on the influencer videos over the weekend all pretty much being unanimous that 12.4.1 wasn't ready, a bit of a head scratcher why they even released it in the first place. You'd expect those issues must have turned up during internal/employee testing.

It is what it is I guess.

 

[rlsd]

Now that we're on yet another and new FSD 12.4.x delay, curious to see what happens with us 2024.3.25 folks. There's no reason to hold us back any longer on the spring update since the latest FSD is on the 2024.15.x branch, yet as of this morning we still sit at 2024.3.5 with zero updates.

On one hand I take it as a good sign we're not being moved up to 2024.14.9. meaning they expect to have the next revision of FSD out very shortly for testing so a few days longer isn't a big deal. If we do start seeing 2024.3.25 upgrade to 2024.14.9, that would be a bad sign in that we will likely be waiting weeks more for an FSD update that will get into the hands of us common folk so they're at least getting us on the spring update in the interim. That said, there is some nice stuff in the spring update I wouldn't mind playing with while we continue to wait.

Based on the influencer videos over the weekend all pretty much being unanimous that 12.4.1 wasn't ready, a bit of a head scratcher why they even released it in the first place. You'd expect those issues must have turned up during internal/employee testing.

It is what it is I guess.

"There's no reason to hold us back any longer on the spring update since the latest FSD is on the 2024.15.x branch, yet as of this morning we still sit at 2024.3.5 with zero updates."

Probably because it creates more work and requires more testing to combine 20.15.x with FSD 12.3.6.

 

[SSDTester#1]

"There's no reason to hold us back any longer on the spring update since the latest FSD is on the 2024.15.x branch, yet as of this morning we still sit at 2024.3.5 with zero updates."

Probably because it creates more work and requires more testing to combine 20.15.x with FSD 12.3.6.

I would be interested to see how the software development works at Tesla. Of course we’ll never know anything because it’s all top secret when it comes to customers finding out about why, how, or when anything happens at Tesla.

 

[zoomer0056]

Anticipation is making me late, is keeping me waiting...

Yup, Carly Simon says these are the good old days. This thread sometimes reminds me of all the anticipation as people wait for the next golden hinges. "When I was thinkin' about how right tonight might be". The very interesting banter between some members is like one line in the song, "And how right your arms feel around me".

 

[SSDTester#1]

Yup, Carly Simon says these are the good old days. This thread sometimes reminds me of all the anticipation as people wait for the next golden hinges. "When I was thinkin' about how right tonight might be". The very interesting banter between some members is like one line in the song, "And how right your arms feel around me".

I used to be super excited for new releases, after 2 years of false hype and mostly disappointments, now I don’t really give a damn. I would like for them to figure out how to give us the latest regular features instead on this junk branch.

Since it’s no longer beta, and is now FSD sUpErViSeD, why are we still on some different branch anyway?

 

[LSDTester#1]

I used to be super excited for new releases, after 2 years of false hype and mostly disappointments, now I don’t really give a damn. I would like for them to figure out how to give us the latest regular features instead on this junk branch.

Since it’s no longer beta, and is now FSD sUpErViSeD, why are we still on some different branch anyway?

Looks like all that ends with the next update. And for all of us who are anxious to get the spring update, I think technically Elon has till the 20th to call it a spring update?

 

[gsmith123]

Since it’s no longer beta, and is now FSD sUpErViSeD, why are we still on some different branch anyway?

Agreed! I mean it doesn’t matter what we think is reasonable, but Tesla chose to graduate FSD out of beta.

I don’t think I should have to make compromises if I’m not a beta tester any more.

On the flip side while it was in beta IMO it was reasonable to expect willing testers to accept bugs and slow access to other features. That’s beta testing. The bar is higher now

 

[spev18]

"There's no reason to hold us back any longer on the spring update since the latest FSD is on the 2024.15.x branch, yet as of this morning we still sit at 2024.3.5 with zero updates."

Probably because it creates more work and requires more testing to combine 20.15.x with FSD 12.3.6.

The spring update 2024.14.9 with FSD 12.3.6 already went wide, so I'm saying they in theory could just push us on 2024.3.5 to that spring update branch with zero additional development work. It's the same exact version of FSD we're on now just with the spring update included.

Since 2024.14.9 is still lower than 2024.15.x that FSD 12.4.x is on, we won't be at a disadvantage and still be able to upgrade when it finally becomes available instead of being too far ahead on the main branch to be eligible.

 

[PianoAl]

Anticipation is making me late, is keeping me waiting...

Tesla, the Heinz ketchup of cars.

 

[rlsd]

The spring update 2024.14.9 with FSD 12.3.6 already went wide, so I'm saying they in theory could just push us on 2024.3.5 to that spring update branch with zero additional development work. It's the same exact version of FSD we're on now just with the spring update included.

Since 2024.14.9 is still lower than 2024.15.x that FSD 12.4.x is on, we won't be at a disadvantage and still be able to upgrade when it finally becomes available instead of being too far ahead on the main branch to be eligible.

2024.3.25/12.3.6 is only 3 percent of 2024.14.9/12.3.6. That small percentage could be for release testing purposes. If Tesla update all cars with 3.25/12.3.6 to 2024.14.9 then there is no guarantee that the update is successful. And if the update is not very good then they have to release 2024.14.10 again and again. The main goal of 2024.14.x is to let who didn't have fsd 12.3.6 to have it. These people have been crying.

 

[JB47394]

It's the same exact version of FSD we're on now just with the spring update included.

Unless it's not. There may be instrumentation collection and reporting in the stuff the testers get that the broader public doesn't see. Tesla may be racking up the miles on FSD, but they also may not want full data from two million cars. And, yes, I know all two million aren't using FSD. There may be other differences that go to the testing cohort that just isn't desired for the broader public.

 

[AlanSubie4Life]

My intervention rate I'm sure is now far lower than yours since I'm fine with the ride in almost all situations

Yes, makes sense. As you can tell here, my experience is somewhat more typical - based on typical feedback here, most users seem to have the same main complaints as I do, but it is likely dependent on the type of situations typically encountered. Suburban driving with lots of traffic and reasonable complexity will lead to many interventions of the types mentioned. Hence that is why my experience is similar to most - it’s a very typical Tesla use case.

In the end I suspect that Tesla will focus on lower intervention rates.

 

[kabin]

Agreed! I mean it doesn’t matter what we think is reasonable, but Tesla chose to graduate FSD out of beta.

I don’t think I should have to make compromises if I’m not a beta tester any more.

On the flip side while it was in beta IMO it was reasonable to expect willing testers to accept bugs and slow access to other features. That’s beta testing. The bar is higher now

As many have said before it would've been more effective to do the traditional system design and engineering upfront versus this seemingly never ending low credibility, slow, piece-meal, hype parade.

 

[spev18]

2024.3.25/12.3.6 is only 3 percent of 2024.14.9/12.3.6. That small percentage could be for release testing purposes. If Tesla update all cars with 3.25/12.3.6 to 2024.14.9 then there is no guarantee that the update is successful. And if the update is not very good then they have to release 2024.14.10 again and again. The main goal of 2024.14.x is to let who didn't have fsd 12.3.6 to have it. These people have been crying.

So what you're saying is that it also matters what branch you upgrade from and not just the branch you are upgrade to? Per TeslaFi, 39.4% of the entire fleet is running 2024.14.9 currently so are you saying that branch would potentially not be successful for those upgrading from 2024.3.25 because so few have done it?

Based on the wide distribution it seems successful for so many others who upgraded from a different branch, but the branch you come from influences success rate? I'm still new so trying to better understand how this all works. If Tesla has to manage different success rates based on each separate previous build a car came from, yikes that gets complicated quick.

 

[jiten87]

Does anyone know what the deal with 3.25 people STILL not on the spring update?

There was the force check update bug that wasnt working last week that seems to have gotten fixed over the weekend, hopefully its coming very soon?

 

[rlsd]

So what you're saying is that it also matters what branch you upgrade from and not just the branch you are upgrade to? Per TeslaFi, 39.4% of the entire fleet is running 2024.14.9 currently so are you saying that branch would potentially not be successful for those upgrading from 2024.3.25 because so few have done it?

Based on the wide distribution it seems successful for so many others who upgraded from a different branch, but the branch you come from influences success rate? I'm still new so trying to better understand how this all works. If Tesla has to manage different success rates based on each separate previous build a car came from, yikes that gets complicated quick.

2024.14.x is the main branch that have all types of hardware: No AP, AP1, AP2, AP2.5, AP3, and AP4. 2024.3.25 branch is mainly for new FSD version testing and it only has AP3 and AP4. Some changes/features that support all hardware types may not work with AP3 and AP4. I think Tesla just wants people who didn't have v12 to have it. That's the main goal. People who already have v12 can wait. It's a strategy to reduce risk, to avoid customer complaints, and to control negative news. This is my speculation and only Tesla knows the truth.

 

[MP3Mike]

Does anyone know what the deal with 3.25 people STILL not on the spring update?

As has been said many times before, this is the "early FSD tester" group, that gets new version of FSD sooner than the rest of the fleet. (But gets the mainstream updates later than the rest of the fleet.)

There will likely always be at least these two different update paths. (We know there are many FSD groups; Internal testers, employees, OG testers/influencers, and the FSD tester group most of us are in.) It would be nice if there was an easy way for people to opt out of the FSD testers path, but I haven't seen anyone find one yet. (I have suggested that someone could try moving their Tesla to a different account, as if they had sold it, and see if that puts it back in the mainstream update path or not, but I haven't seen reports of anyone actually trying it. If it does, it would be interesting to know if transferring it back to the original account moved it back to the FSD testers path or left it on the mainstream path. Note: if you have anything that doesn't transfer to new owners, like FUSC, you obviously wouldn't want to do that.)

It would be nice if they made the standard upgrade setting the mainstream path, and the advanced upgrade setting the FSD tester path. (But it is possible that they don't want a lot more people to be able to join the FSD tester path.)

 

[JulienW]

HELP I think I may have taken too much LSDtester#1.

On my drive this morning I kept noticing nuance differences. It stopped at stop signs much quicker and then almost immediately creep out and then committed much faster. Plus some other things. Also on a roundabout was coming to my exit and the street had just been closed off for construction. The sign was about ¾ up and had a person signaling until they could install the Construction Ahead signs. So this happen in a near instant. I got ready to take over since I expected FSD to head towards and slow down. Instead FSD just continued past with out ANY hesitation and exited the next right. This happened even before the Nav rerouted.

THEN the BIG difference happened in AUTOPARK and others need to check this. Before Autopark ALWAYS took about 2 seconds to switch between R-D. Now it is doing it instantly. I even pulled out and parked in a couple of different spots to try.

Would others please try Autopark and confirm that it is working faster OR LSDtester1's hit they gave me is working.

Also not having any issues with check Software Updates. I checked agin this morning 24 hours latter and it checked again.

 

[OldManCan]

San Diego peeps... Here's something 12.3.6 does consistently trying to kill me. I'm wondering if this is due to map issue or FSD issue or both. Please let me know if anyone had the same experience with this.

Route planned - Genessee ave to I15. FSD is supposed to join I5 from Genesee and then stay on the right to the Local Bypass to eventually take the 56 exit. The issue happens on the Sorrento exit where the nav says there's still 0.2 miles to the exit but arrows show the car following rightmost 2 lanes out of the current road. FSD goes straight initially (as it should) but at the last minute tries to take a right exit.

Apologies for the poor video quality. If you pause it at 23 second mark you can see the exit being wrongly attempted as well as the map / nav data arrows etc.

map overview - Black rectangle is the exit FSD should be taking but it jumps the gun on the exit marked with yellow rectangle.