$12K for FSD is insane

[muzhik]

But I like my options.. haha.. if you have money to burn, why the hell not. Yes you can dump 10k into TSLA but it's not the same.. I think Tesla should move FSD to account level, not per vehicle.

Navigate on Autopilot
Auto Lane Change
Autopark - works most of the time
Summon - pretty neat
FSD computer
Recognize traffic lights and stop sign

 

[ucmndd]

Agreed. to be clear, the metrics I posted above are for AP (not FSD beta), and is definitely a much larger data sample, which grows with each Tesla sold. And very encouragingly, as more and more "general" people start to own Teslas, we're seeing the AP metrics trend toward more safe, not less.

Incremental “improvement” in this area could easily be attributed to the current packaging of base autopilot and the fact that it’s really only useful for limited stretches in a single lane on a freeway. Put another way, the safest and least accident-prone driving condition there is. The dismally low take rate on FSD combined with hiding otherwise useful features like automatic lane changes behind a $12,000 gate means most autopilot miles are logged in the most dead-simple and safe driving conditions.

I’m not convinced autopilot is markedly better than a human driver if you compare a large sample of statistically similar driving conditions. Yeah it’s easy to say it’s “safer” when you compare a billion single lane freeway miles to a billion other mixed use miles in cities, uncontrolled intersections, mountain highways, etc.

Tesla has no credibility on this subject and is very careful to obfuscate important context in their pretty graphs and safety reports.

 

[Knightshade]

Incremental “improvement” in this area could easily be attributed to the current packaging of base autopilot and the fact that it’s really only useful for limited stretches in a single lane on a freeway. Put another way, the safest and least accident-prone driving condition there is. The dismally low take rate on FSD combined with hiding otherwise useful features like automatic lane changes behind a $12,000 gate means most autopilot miles are logged in the most dead-simple and safe driving conditions.

Except AP was still showing as far safer long before they split up the features that way.

Prior to ~March 2019 you either got nothing or at minimum you paid 5k for EAP (which contains most of the features of the current FSD package, including changing lanes-- minus the NON highway stoplight and stop sign stuff).

They have quarterly safety data going back to mid-2018. AP driving was safer than, and it's safer now.


(I don't really buy the argument "everyones using it properly and ONLY on highways" bit either... every time there's a thread on here about AP doing something weird on a road you aren't supposed to use it-- and someone MENTIONS it's intended only for highways- you get a bazillion owners saying they use it on local roads all the time and the manual warning is just there for legal CYA.... The only time people pretend most owners actually use it ONLY on highways is when they wanna try and make Teslas AP safety data look worse)

Tesla has no credibility on this subject

Weird how nobody likes their data, but nobody ever has any to dispute it.

Nobody can even offer an explanation that makes any sense for why having an extra 8 eyes, with better than human reaction times, looking in all directions at all times wouldn't improve overall safety.

 

[Daniel in SD]

Nobody can even offer an explanation that makes any sense for why having an extra 8 eyes, with better than human reaction times, looking in all directions at all times wouldn't improve overall safety.

Please, you know what the explanation is. It doesn't make sense that driver behavior might change when using Autopilot?
I'm not claiming that it negates the benefits but Tesla's statistics are garbage. I'm not sure who else would have Tesla Autopilot collision data other than Tesla. Admittedly it does seem like a difficult thing to determine.

 

[ucmndd]

Except AP was still showing as far safer long before they split up the features that way.

“Far safer” than what? That’s my point. To my knowledge there has been no study whatsoever on the safety implications of autopilot compared to manual driving over a large dataset of statistically similar conditions. Tesla compares “autopilot on” to “all other driving”, and I don’t care how you cut it or try to defend it, the kindest way to characterize that is “disingenuous”. It’s not a valid nor useful comparison and has no statistical value. It’s misleading marketing by a Silicon Valley carnival barker.

The only time people pretend most owners actually use it ONLY on highways is when they wanna try and make Teslas AP safety data look worse)

This is deluded. It’s foolish to pretend that the overwhelming, vast majority of autopilot miles in the Tesla dataset aren’t freeway miles. That type of driving (again, the safest type of driving) is overrepresented by orders of magnitude when compared to a more general distribution of driving conditions.

 

[nvx1977]

“Far safer” than what? That’s my point. To my knowledge there has been no study whatsoever on the safety implications of autopilot compared to manual driving over a large dataset of statistically similar conditions. Tesla compares “autopilot on” to “all other driving”, and I don’t care how you cut it or try to defend it, the kindest way to characterize that is “disingenuous”. It’s not a valid nor useful comparison and has no statistical value. It’s misleading marketing by a Silicon Valley carnival barker.


This is deluded. It’s foolish to pretend that the overwhelming, vast majority of autopilot miles in the Tesla dataset aren’t freeway miles. That type of driving (again, the safest type of driving) is overrepresented by orders of magnitude when compared to a more general distribution of driving conditions.

Agree that comparing highway to mixed is not useful. But Tesla's data does show valid comparisons:

Note the gray line at the bottom is US benchmark. Then red is no AP and no active safety, Teslas get into fewer accidents. That's kinda interesting. It implies Tesla owners are better drivers. And when you look at orange, where active safety features are enabled. Safety ticks up a bit.

So far, no AP involved, so if you think AP data is biased due to highway miles, then red/orange can be assumed to be a mix, since AP is out of the picture. Without AP but with active safety, Tesla is already 4x above the US benchmark.

If you consider Tesla's data set to be not large enough, just give it a year or two. At Tesla's growth rate of selling cars, it should become very apparent within a few years where AP safety stands. It won't be just early adopters driving Teslas anymore. Will we see the red line start to trend down? Will AP data on its own continue to trend up in safety?

Just calling out a flawed comparison between highway-biased miles vs mixed driving isn't sufficient to refute the safety benefits of the active safety features and AP. I would tend to agree with your cynicism if over time, AP safety trends downwards as the sample size grows. But that's not the case.

 

[COS Blue]

Nobody can even offer an explanation that makes any sense for why having an extra 8 eyes, with better than human reaction times, looking in all directions at all times wouldn't improve overall safety.

Because the software is crap. Driving down I-25 at 80mph, the car slams on the brakes. Those extra 8 eyes saw the shadow of a bridge and freaked out. If there was a car behind me, I would have been rear-ended.

Weird how nobody likes their data, but nobody ever has any to dispute it.

Here's some data:

• Number of times I have slammed on the brakes on the Interstate at 80mph for no good reason: 0
• Number of times my Tesla has slammed on the brakes on the Interstate at 80mph for no good reason: 1

That's just one example -- the system lost my trust almost as soon as I started using it. If I were with a human driver who did that, I wouldn't trust that driver any more. Same thing here.

 

[raptor5244]

Because the software is crap. Driving down I-25 at 80mph, the car slams on the brakes. Those extra 8 eyes saw the shadow of a bridge and freaked out. If there was a car behind me, I would have been rear-ended.

Here's some data:

• Number of times I have slammed on the brakes on the Interstate at 80mph for no good reason: 0
• Number of times my Tesla has slammed on the brakes on the Interstate at 80mph for no good reason: 1

That's just one example -- the system lost my trust almost as soon as I started using it. If I were with a human driver who did that, I wouldn't trust that driver any more. Same thing here.

This is the key issue. It doesn't matter if you have 20 cameras, lidar, radar, etc. if you can't trust the system or can't stand the way it drives.

 

[Daniel in SD]

Agree that comparing highway to mixed is not useful. But Tesla's data does show valid comparisons:

View attachment 758197

Note the gray line at the bottom is US benchmark. Then red is no AP and no active safety, Teslas get into fewer accidents. That's kinda interesting. It implies Tesla owners are better drivers. And when you look at orange, where active safety features are enabled. Safety ticks up a bit.

So far, no AP involved, so if you think AP data is biased due to highway miles, then red/orange can be assumed to be a mix, since AP is out of the picture. Without AP but with active safety, Tesla is already 4x above the US benchmark.

If you consider Tesla's data set to be not large enough, just give it a year or two. At Tesla's growth rate of selling cars, it should become very apparent within a few years where AP safety stands. It won't be just early adopters driving Teslas anymore. Will we see the red line start to trend down? Will AP data on its own continue to trend up in safety?

Just calling out a flawed comparison between highway-biased miles vs mixed driving isn't sufficient to refute the safety benefits of the active safety features and AP. I would tend to agree with your cynicism if over time, AP safety trends downwards as the sample size grows. But that's not the case.

How are collisions measured for the US benchmark? Is it the same way that Tesla measures it?
I don't think anyone is saying that the sample size isn't large enough...
Saying the analysis is flawed is just saying you can't draw any conclusions either way.

 

[ucmndd]

Agree that comparing highway to mixed is not useful. But Tesla's data does show valid comparisons:
…
Note the gray line at the bottom is US benchmark. Then red is no AP and no active safety, Teslas get into fewer accidents. That's kinda interesting. It implies Tesla owners are better drivers.

I’ve gotta admit, I have problems with this comparison too, as it omits critical data.

I agree, it appears Tesla drivers apparently get into fewer accidents than the entire driving population as a whole. But why?

Do Tesla drivers get in fewer accidents than drivers of other comparable cars?

Do Tesla drivers get in fewer accidents than other drivers with statistically similar age/economic demographics?

This is basically a correlation/causation problem. The chart would clearly prefer we draw the conclusion that Teslas get in fewer accidents because of Tesla’s massively innovative and cutting edge active safety features enabled by the AP suite. I don’t think that’s proven at all, or furthermore that they are measurably better than any similar car with similar features driven by a similar driver.

 

[nvx1977]

How are collisions measured for the US benchmark? Is it the same way that Tesla measures it?
I don't think anyone is saying that the sample size isn't large enough...
Saying the analysis is flawed is just saying you can't draw any conclusions either way.

Seems really straightforward.... miles per accident is how it's measured.

Like I keep saying but ppl selectively ignore, taking out all the AP data, Tesla with active safety is scoring 4x higher in miles per accident. Could it be possible that Teslas drive a different mix of local and highway? Sure. I don't see how any other study is going to normalize for that, so the only way is to increase the sample size of Tesla's data so it is less biased (in whatever way it might be biased).

 

[ucmndd]

Nobody can even offer an explanation that makes any sense for why having an extra 8 eyes, with better than human reaction times, looking in all directions at all times wouldn't improve overall safety.

Nobody should have to offer an explanation for this, because the underlying premise is absurd.

Why wouldn’t a plastic bag filled with iPhones be better than my black lab at retrieving a stick from a pond?

 

[nvx1977]

I’ve gotta admit, I have problems with this comparison too, as it omits critical data.

I agree, it appears Tesla drivers apparently get into fewer accidents than the entire driving population as a whole. But why?

Do Tesla drivers get in fewer accidents than drivers of other comparable cars?

Do Tesla drivers get in fewer accidents than other drivers with statistically similar age/economic demographics?

This is basically a correlation/causation problem. The chart would clearly prefer we draw the conclusion that Teslas get in fewer accidents because of Tesla’s massively innovative and cutting edge active safety features enabled by the AP suite. I don’t think that’s proven at all, or furthermore that they are measurably better than any similar car with similar features driven by a similar driver.

This is why I think as more Teslas are on the road, the current factors that might bias the Tesla data in their favor should diminish, thereby revealing a more true comparison with a general US benchmark. Age/economic demographics might continue to be a valid bias until a 25k Tesla is launched.

Until then, all we have are Tesla's current numbers. I don't think their raw numbers are incorrect or widely measured differently. Calling out the comparison of highway-biased AP mileage is completely fair. And I agree that it's difficult to draw conclusions either way. But I do like that AP miles/accident is trending upward over time, especially given how many new Teslas are being added to the fleet every year.

 

[ucmndd]

Like I keep saying but ppl selectively ignore, taking out all the AP data, Tesla with active safety is scoring 4x higher in miles per accident. Could it be possible that Teslas drive a different mix of local and highway? Sure. I don't see how any other study is going to normalize for that

See above for my take - the proper way to normalize is to compare statistically similar drivers in statistically similar cars. If you can then show that Teslas have statistically better outcomes, you’ve actually attributed something to the car/technology. If you show that Lexus drivers also get in 4x less accidents than the whole population, you have to accept that it might be something else…

 

[Big Dog]

At a minimum, Tesla should disaggregate its data to show controlled highways miles separately. (Controlled highways are safer for all drivers vs city towns & open two lane roads.) Tesla has the data and it's an easy search. The fact that Tesla chooses not to do so....

 

[OxBrew]

At a minimum, Tesla should disaggregate its data to show controlled highways miles separately. (Controlled highways are safer for all drivers vs city towns & open two lane roads.) Tesla has the data and it's an easy search. The fact that Tesla chooses not to do so....

Not only that, Tesla needs to disaggregate the TYPE of highway driving.

Are the actual miles that AP is logging (and getting safer) only looking safer because users are increasingly using it only in more safe situations? If so, the accidents per AP mile would be going down only because people use it in safer and safer situations. AP could actually be getting worse, and we have no idea, because Tesla won't tell us.

I fact, as AP gets worse, the stats could actually get better as a result, since people stop using it in all but the safest situations.

I challenge Elon to prove me wrong, and release that level of meaningful data.

All AP data released to date is literally irrelevant.

 

[ucmndd]

Are the actual miles that AP is logging (and getting safer) only looking safer because users are increasingly using it only in more safe situations?

Agree. This was my basic point in my first post. The great majority of the fleet now has a nerfed baby version of autopilot that is just radar cruise and lane keeping. Need to change lanes? Ah, gotta disable autopilot for that. Accidentally crunch into that Elantra in your blind spot? Whoops, silly human. Not Autopilot’s fault!

It’s really only useful for the safest most boring driving scenarios, so that’s where people overwhelmingly use it, and the data is going to continue to reflect that.

 

[Daniel in SD]

Seems really straightforward.... miles per accident is how it's measured.

How is accident defined?

 

[roc-noc]

Agree that comparing highway to mixed is not useful. But Tesla's data does show valid comparisons:

View attachment 758197

Note the gray line at the bottom is US benchmark. Then red is no AP and no active safety, Teslas get into fewer accidents. That's kinda interesting. It implies Tesla owners are better drivers. And when you look at orange, where active safety features are enabled. Safety ticks up a bit.

So far, no AP involved, so if you think AP data is biased due to highway miles, then red/orange can be assumed to be a mix, since AP is out of the picture. Without AP but with active safety, Tesla is already 4x above the US benchmark.

If you consider Tesla's data set to be not large enough, just give it a year or two. At Tesla's growth rate of selling cars, it should become very apparent within a few years where AP safety stands. It won't be just early adopters driving Teslas anymore. Will we see the red line start to trend down? Will AP data on its own continue to trend up in safety?

Just calling out a flawed comparison between highway-biased miles vs mixed driving isn't sufficient to refute the safety benefits of the active safety features and AP. I would tend to agree with your cynicism if over time, AP safety trends downwards as the sample size grows. But that's not the case.

Interesting that there are more accidents in Winter. That makes perfect sense as AP has a difficult time with salt covered roads and during snow or rain storms. Obviously I would expect more accidents during those weather conditions when AP won't work. But it does skew the numbers when you say that driving with AP on is much safer than driving without it. So when that dangerous and risky weather related driving is happening and the cause of more accidents, AP is inactive and not tarnishing its safety record.

 

[nvx1977]

Interesting that there are more accidents in Winter. That makes perfect sense as AP has a difficult time with salt covered roads and during snow or rain storms. Obviously I would expect more accidents during those weather conditions when AP won't work. But it does skew the numbers when you say that driving with AP on is much safer than driving without it. So when that dangerous and risky weather related driving is happening and the cause of more accidents, AP is inactive and not tarnishing its safety record.

Definitely a valid bias!

I live in an area with lots of snow and ice (looks like you do too). Typically the dangerous period is during the storm. By the next day, the sand/salt trucks and the plows have done their job, and the roads are pretty clear, and AP is back in business. But yeah, I can totally see people relying less on AP during a storm, when it is way less effective.

There's a nice bowl shape in the red and orange lines, centering on Q1, which would be the worst of winter. Interestingly, the AP data shows sharp dips every Q4, which is likely also a seasonality issue, but perhaps not winter weather related. Holidays maybe? Thanksgiving and Christmas/New Year traffic.