Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

FSD Beta 10.13

This site may earn commission on affiliate links.
As an aside, it's a very interesting story and probably one of the first examples of this kind of large-scale ML training research. Certainly there are massive data-gathering practices by governments and our beloved tech companies, of debatable wholesomeness. But this amount of highly specific user data, uploaded openly and directly from the user terminal (rather than scraped from users' other-purpose browsing and information consumption), and applied directly to a project that most people would call helpful progress, seems like a historic first. (No doubt that may provoke some cynical responses but I think it's true.)

lol what? Tesla must have made you forget about all the other smart devices we have had for over a decade.

Every time you used your voice assistant on your phone, tablet, desktop, smart speaker, etc. The audio recording of your voice was being stored on Google, Amazon servers.
 
Last edited:
And of course a system that knows what it cannot see. This is (in theory) easy for a human. Not sure how it is done with a visual perception system.
This is probably one of the biggest challenges. If the system can see something it can determine speed, direction etc and make an educated decision on how to react. If the system does not know something is there ( unable to see ) or even if something could be there ( road not visible) it has no data to act on.
 
lol what? Tesla must have made you forget about all the other smart devices we have had for over a decade.

Every time you used your voice assistant on your phone, tablet, desktop, smart speaker, etc. The audio recording of your voice was being stored on Google, Amazon servers.
Yes I expected this kind of response, as I said. Google, Amazon, Facebook have long been doing data collection as you say, and certainly one of the motivations along their path is further training of their voice-recognition and behavioral prediction algorithms. But a huge part of the motivation, and indeed the ultimate goal, is to get to know you better than you know yourself, to push advertisements, sell your profile, urge you to expand your connections to grow the social networks and so on. There's a component of pure technical progress to be sure, but it's siphoned from your current-usage intent of doing something else. Google's very annoying and time-consuming image Captcha isn't so much there for your ID security as it is to help them train visual-recognition edge cases. If you ask Alexa or Google how the Greeks make their olive oil, you'll be offered all kinds of olive oil and related products next time you go to Amazon, or via helpful shopping-cart suggestions from Alexa herself. Same for the others.

I see the FSD beta driving-clip uploads as a much more intensive, directed-purpose collection that Tesla is quite clear about. The amount of user-terminal data being pulled is staggeringly high as we've discussed. Sure, perhaps someday they could branch out into the profiling and selling of your destinations and habits. But at least at this point, it's pretty clear that they need and are using the data directly to improve the product you're using, and both parties benefit from said improvements. This is not naivete on my part; it is simply a business model that we all can understand, whether or not we feel it is technically correct or fairly priced.

I don't really expect you to agree with this because I understand your highly negative take on all things Tesla. You're a smart guy as are most people here, but we can disagree on the interpretation and the probability of eventual success. I think my take on Tesla is much more balanced; I have my criticisms and strong suggestions (which they're not likely to read or take), but there are some very interesting, potentially positive things going on and I don't fail to recognize them. Twenty years from now, if you write a book on the development of ML/NN artificial intelligence technology (however it turns out), you'd be remiss in failing to devote some space to this FSD beta program and the associated massive, fleet-wide collection of real-world cases.
 
Last edited:
  • Like
Reactions: edseloh
"Or so" hopefully means it will be released when it is ready, not when a certain number of days pass (hopefully).

I just don't really understand this line of thinking. Maybe I was expecting something different but around the time I got into beta it was 10.3.1 and the public firmware tended to be the same as, or a little behind FSD. At that time we were all seeing very tangible, noticable differences between versions of FSD and were very excited... Then updates started to slow down significantly because of FSD and we started to get more and more behind the public version. We are waiting "one more week" for an outdated main firmware. Why is that acceptable? We are the owners that are the most loyal and paid the most for the features in our cars yet we will probably get the 2022.24 version in like late September or October at this rate.
 
They are in the back seat and get flung forwards into the back of the driver seat.

He needs a dog hammock. Although it won’t solve this problem entirely it could help.

Also phantom braking is really not that bad. Super annoying. But not extreme.
I have the Tesla dog hammock. They are chocolate labs. It IS that bad when it occurs at 80-85mph which is within the speed limit where I was driving.
 
I just don't really understand this line of thinking. Maybe I was expecting something different but around the time I got into beta it was 10.3.1 and the public firmware tended to be the same as, or a little behind FSD. At that time we were all seeing very tangible, noticable differences between versions of FSD and were very excited... Then updates started to slow down significantly because of FSD and we started to get more and more behind the public version. We are waiting "one more week" for an outdated main firmware. Why is that acceptable? We are the owners that are the most loyal and paid the most for the features in our cars yet we will probably get the 2022.24 version in like late September or October at this rate.
Well it is technically an opt in Your decision choice. Not based on what you paid but Your decision to participate. Less variables is a better solution so implementing them into FSD just goes slower. The more critical challenges will take much longer in the future as will the time frame to release new versions. We like to point out the failures but we are in a hurry to get to the next version.
 
Well it is technically an opt in Your decision choice. Not based on what you paid but Your decision to participate. Less variables is a better solution so implementing them into FSD just goes slower. The more critical challenges will take much longer in the future as will the time frame to release new versions. We like to point out the failures but we are in a hurry to get to the next version.

It is an opt-in but now that I have it I don't want to lose it because who knows when the "release candidate" version will actually be released. I'm getting an MXP soon with FSD for my wife but I won't be opting in so that at least one of our cars gets the latest public version.

I wasn't expecting a trade-off when I first got in and now I'm so used to FSD that I couldn't go back. I find that it actually does really well in most situations now except almost daily I notice some sort of "edge" case situation where I won't even challenge FSD cause I know it will mess up. Construction being the most obvious one.

Honestly I'm in favor of them just releasing an update every like 2 weeks (not Elon time but real time) even if there are very minimal updates if only to just stay up to date. The differences in behavior are nowadays pretty subtle to begin with. Maybe this UPL update will be big but still sort of an edge case that they are specifically trying to fix.
 
Maybe there were roundabout improvements as mentioned back in June:
Interesting that FSD Beta 10.13 notes don't have any explicit mentions of roundabouts. But maybe these roundabout improvements are part of a general solution related to unprotected left turns?
That would be nice because for me, the roundabouts are one of the worse problems with current fsd version. It borders between humorous and dangerous when I try it. Maybe other folks have better luck, but I don't.
 
  • Like
Reactions: jabloomf1230
I’m guessing the current NHTSA investigation is not because PB ISNT considered a safety issue by them.

Then again, perhaps a poster on the internet knows more than the NHTSA persons.
It’s hard to know. That’s why they are investigating! Which they should definitely do! Just stating that in my experience I have never experienced a serious and dangerous slowdown. Just a lot of super annoying and shocking ones which are very disconcerting and horrible but not a safety issue, since I was driving the car.

It is possible there are phantom braking incidents which are safety issues, but I have never seen one documented (except for the well known recall). It would be great to see one. I would describe an issue as more than a 10mph reduction in speed in less than a second. Roughly. We could discuss that metric to figure out what is reasonable (maybe 7mph in less than a second?). Emergency stop is 60 to 0 in less than 3 seconds, which is obviously super dangerous. So that is 20mph per second.
 
Last edited:
It is an opt-in but now that I have it I don't want to lose it because who knows when the "release candidate" version will actually be released. I'm getting an MXP soon with FSD for my wife but I won't be opting in so that at least one of our cars gets the latest public version.

I wasn't expecting a trade-off when I first got in and now I'm so used to FSD that I couldn't go back. I find that it actually does really well in most situations now except almost daily I notice some sort of "edge" case situation where I won't even challenge FSD cause I know it will mess up. Construction being the most obvious one.

Honestly I'm in favor of them just releasing an update every like 2 weeks (not Elon time but real time) even if there are very minimal updates if only to just stay up to date. The differences in behavior are nowadays pretty subtle to begin with. Maybe this UPL update will be big but still sort of an edge case that they are specifically trying to fix.
There is a lot of testing and overhead involved in a release, so that if the changes are minors, the team would have gone through a lot of effort for very little payback.
 
Would love to see a video of this. It is all very curious. Other than the safety recall early on, it just seems like you can slam the accelerator after half a second.
When cruising along the interstate I do not typically cruise with my foot over the accelerator I like to relax that foot. I do re-accelerate but that is not a solution to this problem. It does not require a video, it just happens mate, and often.