Price drop on car - Resetting the price on FSD - Anyone else feeling this is highly disingenuous on Tesla's part?

[arbitrage000]

Teslabull tweeted that the parking features are back in FSDb V11.3. So at least that is some progress in getting parking back to the non-USS fleet. Although 11.3 is only to employees so it may still be awhile.

I think the OP made the right choice in cancelling especially with some strong rumors that MX will have HW4 first and may already be in production with them stockpiling them while they sell off the existing inventory.

I read about Tesla not honoring FSD pricing back on the MY threads in Jan. They also didn't honor the MY 7 seater price increase nor the MSM paint increase but those were minor compared to the 21% price drop on the car itself.

 

[aataskin]

Not familiar with the details of the Baidu, but pretty sure that the Mercedes is limited to under 40 mph. And I believe only on specific roads.

A number of companies have gone down the path of hyper-accurate maps and I personally think that it is a huge mistake. That really makes it hard to handle even the simplest of corner cases, such as construction zone reroute, which Tesla seems to be doing a fairly good job at these days.

And should I remind you that Level 3 has "human override is still required"

And yes, we do have a number of autonomous, limited area solutions in the US. But as I believe Elon has said, driving on known streets is easy, driving across the US is hard.

But how is driving across the US can be easier with just the cameras?

On a separate note, I wonder if anybody on this thread knows what exactly happened to Andrej Karpathy? A well-known AI guru who believed that sensors are a liability and who is not at Tesla anymore as of a few months ago. Does this mean anything for HW4? Or for HW5? Sensors coming back finally?

 

[2101Guy]

But how is driving across the US can be easier with just the cameras?

On a separate note, I wonder if anybody on this thread knows what exactly happened to Andrej Karpathy? A well-known AI guru who believed that sensors are a liability and who is not at Tesla anymore as of a few months ago. Does this mean anything for HW4? Or for HW5? Sensors coming back finally?

He went on vacation “sabbatical”.

At which time, several of us (mainly me) posted that we knew….not guessed…knew….he would NOT be coming back.

And well…

 

[ewoodrick]

But how is driving across the US can be easier with just the cameras?

On a separate note, I wonder if anybody on this thread knows what exactly happened to Andrej Karpathy? A well-known AI guru who believed that sensors are a liability and who is not at Tesla anymore as of a few months ago. Does this mean anything for HW4? Or for HW5? Sensors coming back finally?

Because doing it with vision, as opposed to hyper-accurate maps allows for the unknown. Why is it so hard for people to understand that vision works because that's the way we do it every day.

Go look at some RADAR or LIDAR outputs and try to drive it. You can't, they are missing too much information. After all, our roads are designed for driving by vision, not LIDAR.

One part of driving is being able to recognize what it around you and then decide if it is going to impact your travels. Look at the RADAR/LIDAR screens and see how well you can classify things. The look at some of the Tesla visualizations. You'll see Tesla is already identifying and locating things as well and often better than the RADAR/LIDAR solutions.

AI is to the point now where you can give it a picture and it can effectively identify just about everything on the picture. Take that information move to the next few images and you are determining the location and speed of everything. Tesla's doing all that today.
The problem with FSD is not really identifying things that much, it's what you are going to do with the data.

And when you bring up RADAR/LIDAR, ALL you are talking about is being able to see things. Nothing about what to do with it.

 

[aataskin]

Because doing it with vision, as opposed to hyper-accurate maps allows for the unknown. Why is it so hard for people to understand that vision works because that's the way we do it every day.

Go look at some RADAR or LIDAR outputs and try to drive it. You can't, they are missing too much information. After all, our roads are designed for driving by vision, not LIDAR.

One part of driving is being able to recognize what it around you and then decide if it is going to impact your travels. Look at the RADAR/LIDAR screens and see how well you can classify things. The look at some of the Tesla visualizations. You'll see Tesla is already identifying and locating things as well and often better than the RADAR/LIDAR solutions.

AI is to the point now where you can give it a picture and it can effectively identify just about everything on the picture. Take that information move to the next few images and you are determining the location and speed of everything. Tesla's doing all that today.
The problem with FSD is not really identifying things that much, it's what you are going to do with the data.

And when you bring up RADAR/LIDAR, ALL you are talking about is being able to see things. Nothing about what to do with it.

But you don't need to choose one or the other, which is what Tesla is doing. These different technologies work differently and allow for redundancy which is the name of the game if you really want the machine to take over. USS will warn you against a curb which your camera failed to notice because it fogged up - or just failed. Radar will warn you against an imminent collision with a light blue truck against a light blue sky background. Camera will see a pole which the USS will not notice. All these things use different wavelengths and because of this can achieve much more by complementing each other. So why handicap yourself? The guys who got further than Tesla along this path, chose not to.

 

[eli_]

But you don't need to choose one or the other, which is what Tesla is doing. These different technologies work differently and allow for redundancy which is the name of the game if you really want the machine to take over. USS will warn you against a curb which your camera failed to notice because it fogged up - or just failed. Radar will warn you against an imminent collision with a light blue truck against a light blue sky background. Camera will see a pole which the USS will not notice. All these things use different wavelengths and because of this can achieve much more by complementing each other. So why handicap yourself? The guys who got further than Tesla along this path, chose not to.

How do you achieve the correct balance? You can always add MORE sensors, they could have 24+ cameras, even more USS (to fill in blind spots), radars and lidars pointed in every direction. You can fill an entire roof rack with equipment. Obviously there has to be a trade-off.

If I can run vision into my occupancy network and get a decent point cloud out of it some 99.9% of the time then I'm not going to keep drilling holes in bumpers and putting ultrasonics in. If that tech existed at the time when USS was first adopted in vehicles, like in 2003, no one would have ever bothered with USS.

 

[aataskin]

How do you achieve the correct balance? You can always add MORE sensors, they could have 24+ cameras, even more USS (to fill in blind spots), radars and lidars pointed in every direction. You can fill an entire roof rack with equipment. Obviously there has to be a trade-off.

If I can run vision into my occupancy network and get a decent point cloud out of it some 99.9% of the time then I'm not going to keep drilling holes in bumpers and putting ultrasonics in. If that tech existed at the time when USS was first adopted in vehicles, like in 2003, no one would have ever bothered with USS.

Good question. I guess if the two groups that have succeeded (Merc and Baidu) are both, independently, using all of this tech then they must have deemed it important.

To me, if, say, Baidu came out now and said, look, we already have successful self-driving taxis and we have now established that dropping the USS doesn't hurt the reliability so we are doing it for our second generation taxis. If _they_ did this I would have no problem with it because they have credibility. They proved they know what they are doing.

Tesla, on the other hand, started with all those sensors and radars, didn't quite get there unlike these other two groups, then switched direction towards dropping the sensors and the radar, didn't get there again, the guy who was driving that effort is now gone, and now the radar that was not needed just a couple of years ago is rearing its ugly head again in HW4 (which partly answers your question - at least the radar is in fact needed). But then if the radar is now again needed, could it be that the USS are needed as well? We get parking assistant back immediately and get a much better chance at catching up with those other guys who both proved that this suite of sensors works.

 

[SO16]

the guy who was driving that effort is now gone

Andrej was head of AI.

Ashok Elluswamy is and has been head of FSD.

There is also a difference in the quality of radar between old and new.

 

[ewoodrick]

Tesla, on the other hand, started with all those sensors and radars, didn't quite get there unlike these other two groups, then switched direction towards dropping the sensors and the radar, didn't get there again, the guy who was driving that effort is now gone, and now the radar that was not needed just a couple of years ago is rearing its ugly head again in HW4 (which partly answers your question - at least the radar is in fact needed). But then if the radar is now again needed, could it be that the USS are needed as well? We get parking assistant back immediately and get a much better chance at catching up with those other guys who both proved that this suite of sensors works.

Tesla is NOT aiming for limited range taxi service (such as Cruise in the US). That's been mentioned by Elon numerous times. Tesla is only headed for the whole enchilada.

Question: If Baidu was so perfect, why isn't it in passenger cars? Probably same as Cruise vehicles, they don't work outside their geofence. Their coverage area seems to be about 30 sq km max.

Go look at Cruise's recent history. It seems as if they are doing a great job of bringing all traffic in San Francisco downtown to a complete halt lately.

If you look at the details between autonomous taxi service and full self-driving, while there's a lot of similarities, there are huge differences. As I mentioned before, hyper-accurate mapping is used in the taxi service vehicles. They are also hyper-tuned for the road conditions that they drive on. For instance, they may not understand something like a roundabout or an Interstate on ramp.
Autonomous taxi services implement a subset of FSD with a lot of shortcuts. Ask yourself why Cruise, which just boasted a million autonomous miles Cruise’s robotaxis have driven 1 million miles with nobody behind the wheel | Engadget isn't in every city. It basically ONLY knows how to drive in San Francisco, Austin and Phoenix. And in those towns, the range is only a few miles, I've walked from end of Cruises coverage to the other before. Self Driving Cars | Driverless Rides | Cruise

Again, Baidu vs Tesla is truly comparing apples to oranges,

 

[eli_]

Good question. I guess if the two groups that have succeeded (Merc and Baidu) are both, independently, using all of this tech then they must have deemed it important.

To me, if, say, Baidu came out now and said, look, we already have successful self-driving taxis and we have now established that dropping the USS doesn't hurt the reliability so we are doing it for our second generation taxis. If _they_ did this I would have no problem with it because they have credibility. They proved they know what they are doing.

Tesla, on the other hand, started with all those sensors and radars, didn't quite get there unlike these other two groups, then switched direction towards dropping the sensors and the radar, didn't get there again, the guy who was driving that effort is now gone, and now the radar that was not needed just a couple of years ago is rearing its ugly head again in HW4 (which partly answers your question - at least the radar is in fact needed). But then if the radar is now again needed, could it be that the USS are needed as well? We get parking assistant back immediately and get a much better chance at catching up with those other guys who both proved that this suite of sensors works.

Literally everyone in the auto industry uses USS but Tesla culture is all about first principals so they're going to arrive at a different conclusion sometimes, if you work there and your only reasoning is you need a sensor because Mercedes has one, you're not going to last very long. Also we're talking about different products, these Teslas are super-optimized millions/year scale products and they are "works anywhere" driver assistance convenience features, and not "proof of concept" robotaxis.

If the cameras are occluded then the system is just flagged as unhealthy and the driver takes over. It's not like adding radar or ultrasonics is going to change the reliability of the overall system because if your cameras are the primary sensor and they're blinded then you can't drive at all, there's no amount of range-finding with radar or ultrasonics that can get around the fact your cameras are fogged up, you wouldn't even be able to see the lane lines or traffic lights. When cameras don't work you are just SOL.

 

[aataskin]

Literally everyone in the auto industry uses USS but Tesla culture is all about first principals so they're going to arrive at a different conclusion sometimes, if you work there and your only reasoning is you need a sensor because Mercedes has one, you're not going to last very long. Also we're talking about different products, these Teslas are super-optimized millions/year scale products and they are "works anywhere" driver assistance convenience features, and not "proof of concept" robotaxis.

If the cameras are occluded then the system is just flagged as unhealthy and the driver takes over. It's not like adding radar or ultrasonics is going to change the reliability of the overall system because if your cameras are the primary sensor and they're blinded then you can't drive at all, there's no amount of range-finding with radar or ultrasonics that can get around the fact your cameras are fogged up, you wouldn't even be able to see the lane lines or traffic lights. When cameras don't work you are just SOL.

To all of you guys - so how many years do you give until Tesla is finally able to provide park assist again (this is the only one I actually care about), and to provide FSD?

 

[2101Guy]

To all of you guys - so how many years do you give until Tesla is finally able to provide park assist again (this is the only one I actually care about), and to provide FSD?

Define "FSD". Because in Elon's terms, FSD was provided 12/31/2022.

 

[aataskin]

Define "FSD". Because in Elon's terms, FSD was provided 12/31/2022.

L3, legally approved in the US, Germany or the UK or another credible jurisdiction.

 

[ewoodrick]

To all of you guys - so how many years do you give until Tesla is finally able to provide park assist again (this is the only one I actually care about), and to provide FSD?

It's only been gone for a few months. Just gotta wait for the next software release which is in the "any day now" mode.

I can't wait for it to come out, just to shut up the complainers. As has been mentioned numerous times, it's going to get fixed soon, just chill.

 

[ewoodrick]

Literally everyone in the auto industry uses USS but Tesla culture is all about first principals so they're going to arrive at a different conclusion sometimes,

Let me restate your response slightly....

Literally everyone in the auto industry uses internal combustion engines but Tesla culture is all about first principals, so they're going to arrive at a different conclusion sometimes.

Aren't we glad that Tesla doesn't follow the drumbeat of the other manufacturers?

 

[WyoDude]

Let me restate your response slightly....

Literally everyone in the auto industry uses internal combustion engines but Tesla culture is all about first principals, so they're going to arrive at a different conclusion sometimes.

Aren't we glad that Tesla doesn't follow the drumbeat of the other manufacturers?

But I want all of the upsides of Tesla’s innovation and speed, with none of the downsides and I expect that if they ever say anything publicly that is 100% accurate in perpetuity. Is that too much to ask for??

I want them to be completely different from all the bad parts of legacy automakers but also do all the things that legacy companies do that make me feel good inside even if it’s just shielding me from the truth.

I don’t understand why Tesla can’t build a perfect car, software, company, process, customer service, service center experience, PR policy, manufacturing process, all for the price of a Camry, they’ve been doing this for over a decade!

Isn’t the customer king, shouldn’t I be able to get everything I desire, like yesterday?

 

[aataskin]

Let me restate your response slightly....

Literally everyone in the auto industry uses internal combustion engines but Tesla culture is all about first principals, so they're going to arrive at a different conclusion sometimes.

Aren't we glad that Tesla doesn't follow the drumbeat of the other manufacturers?

What we are not glad about is that Tesla can turn off a useful feature without providing an alternative. And given its history of blowing deadlines, by years in the case of FSD, it makes us apprehensive about buying a car that we otherwise like a lot. Parking assists are not crucial for everybody but for some of us they are. I don't think there's a single car on my street without parking sensors. And since the company is already finally admitting the mistake of removing the radar, maybe they can do the right thing and bring the USS back as well until such time when they figure out how to provide parking assists without them. This is what @2102Guy has pointed out well before I got into this thread, and it's so trivial it shouldn't really be a point of discussion.

About marching to the drumbeat. In addition to parking sensors other manufacturers also provide airbags. So should Tesla disable them as well because one day really soon the VISION will get so good that the car won't be involved in any car crashes so why install them? Where do you draw the line?

 

[Joboo7777]

The only thing to consider is that if/when FSDb becomes more functional and consistent I would expect Tesla to increase the subscription cost. I wouldn’t be surprised if they increase the subscription cost when V11 is released. So Purchasing FSDb up front even at 15k could have more upside than current calculations.

 

[ewoodrick]

About marching to the drumbeat. In addition to parking sensors other manufacturers also provide airbags. So should Tesla disable them as well because one day really soon the VISION will get so good that the car won't be involved in any car crashes so why install them? Where do you draw the line?

Uh, Because it's the law?

And uh, because not everyone drives with FSD on?

And most importantly, because even if the car drives as safe as possibly, there are still idiots out there that will do stupid things. Not much can be done if someone runs a red light at high speed in a limited sight area.

"Safest" does tend to be a big part of Tesla's design philosophy. Current example is how many cars can you drive off the end of a 200-foot cliff and no one dies? That's because of all the safety features that Tesla has built in.

If you want to be safer on the road, drive a Tesla, if you want to be even safer, turn FSD on. That's what the stats say.

 

[ewoodrick]

And given its history of blowing deadlines,

When has Tesla ever blown a deadline? AFAIK Tesla doesn't give deadlines to the public. It gives ESTIMATES.

Those are VERY different things.

This is the general problem with the complainers on here. they read something and assume that it's law.