FSD Beta v11.x

rlsd · Mar 2, 2023

I disengage when I hear the sound and see the messages at the bottom of the screen because I can't read the messages. The font for the messages is too small to read and it's hard to read at the bottom of the screen when driving. Tesla should change the UI to display large icons having well meaning at the top of the screen instead.

kriskros91 · Mar 2, 2023

So we went through all of February waiting for Investor Day to be some kind of milestone day for FSDb. Multiple tweets by Musk telling us wide release of FSDb 11.x was right around the corner and we not only don't have it we have no clue what is going on. Radio silence is so annoying.

At this point it sort of feels like summer 2022 again when we didn't get a meaningful FSDb update for like 100+ days.

Musk is an absolute engineering/manufacturing savant and seems to hire the best talent available but even his presentation could best be described as aloof or "in the stars". He needs his employees to sort of reel him in constantly to say the actual realistic, meaningful things.

jeremymc7 · Mar 2, 2023

rlsd said:
I disengage when I hear the sound and see the messages at the bottom of the screen because I can't read the messages. The font for the messages is too small to read and it's hard to read at the bottom of the screen when driving. Tesla should change the UI to display large icons having well meaning at the top of the screen instead.

Agree. Which is why I wear half lens readers when driving to be able to read the screen while no lens on top so it doesn't interfere with my driving. Progressives don't work - I've tried.

Mardak · Mar 2, 2023

kriskros91 said:
At this point it sort of feels like summer 2022 again when we didn't get a meaningful FSDb update for like 100+ days

The time from 10.12.2's May 31st 2022 wide release to 10.69.2's September 11th was 103 days to introduce Occupancy Network.

Excluding the holiday update that didn't change FSD Beta behavior, 10.69.3.1's November 21st wide release to now is 101 days, so we are getting close to a new record with 11.x introducing Single Stack since the 10.2 public beta.

The original FSD Beta before Safety Score did have a longer wait from 8.2's March 3rd 2021 release to 9.0's July 10th release incorporating Tesla Vision removing radar, so that still holds the record at 129 days.

But now with no more Safety Score and no more opt-in, maybe Tesla is handling 11.x release differently from "early access?"

gsmith123 · Mar 2, 2023

They've clearly dialed back their risk tolerance with respect to shipping FSD releases to beta testers. We've seen the 11.3 videos from an employee so they clearly have something pretty drivable. Maybe there are some big issues, but to me the long delay it looks like they've just gotten less comfortable shipping something with flaws to public testers.

It makes some sense with highway AP where people might be pretty unhappy with regressions from production.

I'm a bit surprised they didn't just ship all the FSD beta improvements while still keeping production AP on highways. I wonder if the slower cadence of releases / greater reliance on internal testing is having a negative impact on their overall velocity? I hope they've ramped up internal testing sufficiently for that to not be an issue.

Any speculation on what's holding things up at this point? Are they trying to resolve all the NHTSA recall issues and that's taking time? Is NHTSA requiring them to do that? Have some blocking issues other than the recall popped up? Maybe even minor issues that they want to improve through another couple iterations...

My wild speculation with no evidence is that highway FSD has some regressions compared to prod AP under certain circumstances that they want to fix.

JB47394 · Mar 2, 2023

kriskros91 said:
Musk is an absolute engineering/manufacturing savant and seems to hire the best talent available but even his presentation could best be described as aloof or "in the stars". He needs his employees to sort of reel him in constantly to say the actual realistic, meaningful things.

Why was it "in the stars"? They were talking about what they're working on at a strategic level - scaling the company 10 fold. Not as an aspiration, but by showing the concrete work that they're doing today to make it happen. It's exactly the sort of thing that investors should care about. I took today's tanking of the stock as an indication of how many speculators are out there. They loaded up on stock a few days ago in hopes of some bombshell announcement and then dumped it again today when there was no bombshell. The stock is down a bit more than the jump a few days ago, so some actual investors lost interest in Tesla because they saw the presentation as mere aspirational rambling.

As for me, I thought it was amazing. Restructuring paint lines, motors without rare earths, supply chain handling, smaller factories, etc. It gets me excited because they continue to demonstrate how they're willing to keep thinking outside the box and do what needs to be done to achieve their vision. Then I contrast that with the accountants who run most everything else with a vision no grander than making more money.

archae86 · Mar 2, 2023

JB47394 said:
As for me, I thought it was amazing
<snip> motors without rare earths

Yes indeed. That one all by itself is a really big deal. Someone in China is grinding their teeth.

powertoold · Mar 2, 2023

25.2 is really quite good. I'm happy with it until they can polish up 11 to be noticeably superior.

AlanSubie4Life · Mar 2, 2023

powertoold said:
25.2 is really quite good. I'm happy with it until they can polish up 11 to be noticeably superior.

I agree it is the best build we have had and it is smoother in many situations. Which is not to say smooth in any situations (occasionally things are actually smooth but pretty rare). And I am intervening just as often. But the things it gets right are usually done a bit more smoothly.

Seems like we could be waiting a long time for 11 and I am not sure what it will improve, if anything. Seems most likely to only make things worse which I think probably explains things.

D Good · Mar 2, 2023

Beta tester morale is at an all-time low it seems. For good reasons.

sleepydoc · Mar 3, 2023

D Good said:
Beta tester morale is at an all-time low it seems. For good reasons.

I wouldn’t say morale is low, people are just getting impatient waiting for V11 to be released.

Goose66 · Mar 3, 2023

powertoold said:
25.2 is really quite good. I'm happy with it until they can polish up 11 to be noticeably superior.

I remain shocked every time I see a comment like this. I wonder what possible standard you could be using to consider FSDb 25.2 as "quite good." If this was Teslarati.com, I would just assume you were among the horde of stock pumpers. Or perhaps you are simply looking at it from a progressive development perspective and see progress over the last few releases.

But for me, having driven around on successive versions of NOA and FSDb for over 4 years now, "quite good" would be something that is actually useful that justifies the thousands of $ I paid for it. And by that standard, FSDb is ANYTHING BUT "quite good," or even "good." FSDb has absolutely no utility right now. It adds nothing to the driving task - it doesn't reduce workload or fatigue, and it certainly doesn't increase safety. To drive around on FSDb for me is to be constantly fighting with my car. And now that the reporting icon is gone, I question whether my efforts (struggles) in "testing" FSDb are even contributing to the progress of the product. No, after 17 months on FSDb and 4+ years into Tesla's other autonomous driving features (Summon, Smart Summon, Autopark, NoA, etc.), "quite good" is not an adjective I would consider using to describe FSDb at this stage.

VanFriscia · Mar 3, 2023

sleepydoc said:
I wouldn’t say morale is low, people are just getting impatient waiting for V11 to be released.

I'd say a little of both. I, for one, thought Tesla would have made much more progress by this point in the beta. Obvious progress has been made, but it sure does feel slow coming. Maybe my expectations were too high from the start.

uscbucsfan · Mar 3, 2023

Goose66 said:
I remain shocked every time I see a comment like this. I wonder what possible standard you could be using to consider FSDb 25.2 as "quite good." If this was Teslarati.com, I would just assume you were among the horde of stock pumpers. Or perhaps you are simply looking at it from a progressive development perspective and see progress over the last few releases.

But for me, having driven around on successive versions of NOA and FSDb for over 4 years now, "quite good" would be something that is actually useful that justifies the thousands of $ I paid for it. And by that standard, FSDb is ANYTHING BUT "quite good," or even "good." FSDb has absolutely no utility right now. It adds nothing to the driving task - it doesn't reduce workload or fatigue, and it certainly doesn't increase safety. To drive around on FSDb for me is to be constantly fighting with my car. And now that the reporting icon is gone, I question whether my efforts (struggles) in "testing" FSDb are even contributing to the progress of the product. No, after 17 months on FSDb and 4+ years into Tesla's other autonomous driving features (Summon, Smart Summon, Autopark, NoA, etc.), "quite good" is not an adjective I would consider using to describe FSDb at this stage.

If you've been part of the community that long you would know different users have different experiences. FSD B isn't level 5, but I use it about 90-95% of my drives and it's absolutely useful. I've had countless drives where I didn't have to take over. 25.2 is, overall, a very solid level 2 ADAS for me. It reduces workload and increases safety in my experience. When I drive around town, through shopping districts, through downtown it drives for me and keeps in my lane, makes turns, and avoids incidents/potential wrecks.

There are situations where I don't allow it to attempt, with my family in the car. I just disengage and re-engage when I manually drive through it and there are still obviously times where it just randomly shits the bed, but overall I use it more than I drive normally and I miss having it when I'm in a rental car for work. It's worth 10k for me, but obviously I want it to progress. I just don't believe (and I never have since I received 10.3) that it will reach full level 5 automation.

Even when I purchased it (3 times now) I never really believed it would be something I could summon to come get me from the Airport or that I could sleep in the backseat while driving. For me, what it does and my expectations, it was a worthy upgrade option compared to massaging seats or other options I've added to previous cars.

I think a lot of it depends on your perspective and what your expectations were. Elon is full of *sugar* and has delivered incredible products in the same breathe.

kabin · Mar 3, 2023

archae86 said:
Yes indeed. That one all by itself is a really big deal. Someone in China is grinding their teeth.

No doubt China has a strangle hold there. It's promising but for now more likely it's Elon bluster/vapor ware.

kabin · Mar 3, 2023

Goose66 said:
I remain shocked every time I see a comment like this. I wonder what possible standard you could be using to consider FSDb 25.2 as "quite good." If this was Teslarati.com, I would just assume you were among the horde of stock pumpers. Or perhaps you are simply looking at it from a progressive development perspective and see progress over the last few releases.

But for me, having driven around on successive versions of NOA and FSDb for over 4 years now, "quite good" would be something that is actually useful that justifies the thousands of $ I paid for it. And by that standard, FSDb is ANYTHING BUT "quite good," or even "good." FSDb has absolutely no utility right now. It adds nothing to the driving task - it doesn't reduce workload or fatigue, and it certainly doesn't increase safety. To drive around on FSDb for me is to be constantly fighting with my car. And now that the reporting icon is gone, I question whether my efforts (struggles) in "testing" FSDb are even contributing to the progress of the product. No, after 17 months on FSDb and 4+ years into Tesla's other autonomous driving features (Summon, Smart Summon, Autopark, NoA, etc.), "quite good" is not an adjective I would consider using to describe FSDb at this stage.

True dat. The industry is screaming out for nonbiased criteria to add an iota of honesty to unscrupulous manufacturers looking to make a quick buck.

powertoold · Mar 3, 2023

Goose66 said:
I remain shocked every time I see a comment like this. I wonder what possible standard you could be using to consider FSDb 25.2 as "quite good." If this was Teslarati.com, I would just assume you were among the horde of stock pumpers. Or perhaps you are simply looking at it from a progressive development perspective and see progress over the last few releases.

But for me, having driven around on successive versions of NOA and FSDb for over 4 years now, "quite good" would be something that is actually useful that justifies the thousands of $ I paid for it.

I run a very small enterprise software business, and software-wise, 25.2 is very very good. It does almost everything it needs to to accomplish its goals, and it does it without bugging out and reliably in all sorts of environments and weather conditions. It's rather incredible... The missing / unknown piece is training, labeling, and NN architecture. NNs are known to be good at being creative but not at factual information: essentially NNs are not known to be great at real-world AI.

Whether Tesla can auto-label and train the data in a way that forces the NNs to behave in a predictable, repeatable, and reliable way remains to be seen. When Tesla first released Tesla Vision-only, my mind was blown and still is, because people who have worked with computer vision know how difficult it is to get such reliable and accurate outputs from only cameras. It gave me hope that Tesla will eventually get there, but the diversity of environments and road geometry may be a whole nother level of insane.

bayareaever · Mar 3, 2023

powertoold said:
I run a very small enterprise software business, and software-wise, 25.2 is very very good. It does almost everything it needs to to accomplish its goals, and it does it without bugging out and reliably in all sorts of environments and weather conditions. It's rather incredible... The missing / unknown piece is training, labeling, and NN architecture. NNs are known to be good at being creative but not at factual information: essentially NNs are not known to be great at real-world AI.

Whether Tesla can auto-label and train the data in a way that forces the NNs to behave in a predictable, repeatable, and reliable way remains to be seen. When Tesla first released Tesla Vision-only, my mind was blown and still is, because people who have worked with computer vision know how difficult it is to get such reliable and accurate outputs from only cameras. It gave me hope that Tesla will eventually get there, but the diversity of environments and road geometry may be a whole nother level of insane.

Does anyone know the status of Dojo in all this? Seems like a key missing piece but not being talked about here or by the company lately. The timeline for dojo being implemented seems to have passed a few times already (‘Elon’ timeline at least).

BBTX · Mar 3, 2023

bayareaever said:
Does anyone know the status of Dojo in all this? Seems like a key missing piece but not being talked about here or by the company lately. The timeline for dojo being implemented seems to have passed a few times already (‘Elon’ timeline at least).

I don't know the timeline, but I do know from the Investor Day presentation that as of a couple days ago, they're still not using it yet. They gave stats on their "traditional" GPU clusters and talked about Dojo eventually improving on it.

BBTX · Mar 3, 2023

Goose66 said:
And now that the reporting icon is gone, I question whether my efforts (struggles) in "testing" FSDb are even contributing to the progress of the product.

It's pretty clear from various presentations that they mine data from at least disengagements, if not also interventions. So in that sense, yes, continued testing and intervening/disengaging as necessary still contributes to the training of the NNs.

FSD Beta v11.x

Active Member

Member

Active Member

Active Member

Member

Active Member

Member

Active Member

Efficiency Obsessed Member

Member

Well-Known Member

Member

Member

Active Member

Active Member

Active Member

Active Member

Member

Member

Member

Similar threads