98 Score - Got FSD Beta?

[johnmodely]

I understand being done "playing this game", I get that but I think you're way off on timeline. The more testers, the more data and that's what they're looking for. I wouldn't give up so fast but yes, if you're at a 97, if could be a couple months for you but not "years"

Lol you must be new...Tesla has been promising full FSD since 2018...I bought my MY when it first came out since FSD on city streets (AKA the current FSD Beta) was promised by end of year and even I at the point was like well there is no way they won't have this out by end of year.

Fast forward end of 2021 and its no where near public release...if you don't think it will be years you're naive.

 

[davefink]

Sustained 98, no fsd beta received yet. Waiting.......... This sucks

Same here M3 performance with FSD and 98 since around 11/15 and no beta yet

 

[Spaceship89]

Lol you must be new...Tesla has been promising full FSD since 2018...I bought my MY when it first came out since FSD on city streets (AKA the current FSD Beta) was promised by end of year and even I at the point was like well there is no way they won't have this out by end of year.

Fast forward end of 2021 and its no where near public release...if you don't think it will be years you're naive.

Dude... I'm not talking about public release, I'm talking about access to beta for 97/98 with right hardware. I may be new to this forum but I've had my Tesla since 2018 so I've been apart of the Tesla community about 50% longer than you have.

 

[danaoredson]

Sustained 98, no fsd beta received yet. Waiting.......... This sucks

Same here 98 since inception. Purchased FSD 3+ years ago.

 

[dropgems]

Any word on 97’s?

 

[ABerry]

Still waiting at 98. Looking like 99 tonight or tomorrow based on notateslaapp calculator.

 

[MattEllison]

I am at a 99 now so hopefully soon (if they are even rolling it out anymore)

 

[yurei]

Got to 99 in the meantime - no beta yet.

 

[LionCheeser]

I also finally crossed into 99... but the wait continues...

 

[MattEllison]

I honestly think that we might be waiting until this weekend for 10.6 potentially. It seems like the 10.5 rollout has paused or stopped entierly.

 

[JonNRb]

I honestly think that we might be waiting until this weekend for 10.6 potentially. It seems like the 10.5 rollout has paused or stopped entierly.

+1

I'd thought they'd continue after the holiday, but seems like a no there. Either they decided to just wait to pull people onto 10.6 when that released or somebody forgot to press "unpause". (Or they've decided the beta program is big enough. Please don't let that be the case...)

 

[alee]

+1

I'd thought they'd continue after the holiday, but seems like a no there. Either they decided to just wait to pull people onto 10.6 when that released or somebody forgot to press "unpause". (Or they've decided the beta program is big enough. Please don't let that be the case...)

They need a lot more drivers and data to be successful. I can't see a scenario where the beta would be too big at this point ... only too big, too soon.

 

[Lisska70]

Sustained 98, no fsd beta received yet. Waiting.......... This sucks

Same here 2018 M3 LR 98 since 11/20 no beta and no news as to why

 

[JonNRb]

only too big, too soon

Lol yeah agreed, I'm thinking short-term here.

 

[johnmodely]

They need a lot more drivers and data to be successful. I can't see a scenario where the beta would be too big at this point ... only too big, too soon.

Doubt they have dojo running yet, or even if it is its probably at a point where they can't train a shitton of video in any meaningful way. They are probably up to 10K+ for beta, and the beta size for Tesla has probably never been over 1k cars.

They are for sure overwhelmed and don't need to expand beta, so they probably won't...at least not until there are significant improvements or dojo gets to the point where it can train on 100's of thousands of data coming from cars. I don't see either happening for months, which probably means beta will just a a trickle of new people going forward.

 

[Mardak]

They are probably up to 10K+ for beta, and the beta size for Tesla has probably never been over 1k cars.

I estimate the FSD Beta population to be closer to 20k than 10k, and adding 98 safety scores will definitely get closer to 30k.

At least in the last 50 days of public beta, this current 5-day gap in additions matches two previous 5-day gaps of additions although 10.4 and 10.5 were released during those times.

at least not until there are significant improvements or dojo gets to the point where it can train on 100's of thousands of data coming from cars

The release notes for FSD Beta 10.5 includes an item of 165k auto-labeled videos, so that is some significant improvement to their training pipeline. Dojo does help in the neural network training portion, but there's still data collection and labeling that they have been improving in parallel. Arguably, Tesla has been increasing the FSD Beta fleet so they have more diverse data collection with the extra benefit of distributed human processing of deciding when to send video snapshots of interesting situations.

 

[johnmodely]

I estimate the FSD Beta population to be closer to 20k than 10k, and adding 98 safety scores will definitely get closer to 30k.View attachment 738909At least in the last 50 days of public beta, this current 5-day gap in additions matches two previous 5-day gaps of additions although 10.4 and 10.5 were released during those times.


The release notes for FSD Beta 10.5 includes an item of 165k auto-labeled videos, so that is some significant improvement to their training pipeline. Dojo does help in the neural network training portion, but there's still data collection and labeling that they have been improving in parallel. Arguably, Tesla has been increasing the FSD Beta fleet so they have more diverse data collection with the extra benefit of distributed human processing of deciding when to send video snapshots of interesting situations.

Yea 165k videos would make sense from a 10-20k beta fleet. Now can they scale that to millions of videos yet? Probably not. There are also diminishing returns on more video as well...im no AI expert but know enough about datasets that quality always trumps quantity.

 

[alee]

Yea 165k videos would make sense from a 10-20k beta fleet. Now can they scale that to millions of videos yet? Probably not. There are also diminishing returns on more video as well...im no AI expert but know enough about datasets that quality always trumps quantity.

With training sets, as long as they can be classified, more is always better (at this stage). We are extremely early days here... regular AP is based on billions of miles driven. I doubt they have more than 100MM miles of FSD on the books.

For instance, they definitely don't have enough capture on unprotected left turns with center divider on a 2x2 highway. They have a lot from one particular driver, but they cannot just solve for that one intersection. They need to see many more of those, which can only happen with more drivers.

As I see it, the Safety Score game was just a means to gate the number of inductees into the program so they can scale up capacity. Not knowing anything, I'd guess 100k test cars would be a good aspirational goal - you'll have a small percentage that get booted in a cycle, a good percentage that never report anything (the car has to decide what to send home), a small group casually reporting, and a small percentage that are actively reporting.

 

[jwoyame]

I estimate the FSD Beta population to be closer to 20k than 10k, and adding 98 safety scores will definitely get closer to 30k.

Data point: When Tesla issued a recall to the NHTSA for the 10.3 beta, there were 11,704 vehicles recalled. That was Oct 23, so I think your 20k estimate is valid.

 

[MattEllison]

Downloading beta now at a 98...