Weird that Tesla would ignore speed, running red lights, and acceleration in their metrics, but is worried about how hard you turn, and that's a good predictor of insurance claims, and that the #1 predictor of an accident is ignoring an AP warning, 10X larger than unsafe following and 100X unsafe following distance.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Safety Score
- Thread starter ScottNY
- Start date
Knightshade
Well-Known Member
How many states do you sell insurance in?
I ask because you appear to think you know more about insurance claims and costs than a company that sells it to states with a total population north of 100 million people.
I know that when Tesla uses completely invalid methods to calculate their "10X safer than other cars" metric of Teslas on AP, and when they literally call it a "beta calculation" with no data behind it that needs to evolve, I have no reason to trust it as an actual actuarial tool.
What does the population of the states they have coverage in have to do with anything anyway? They only sell to Tesla owners. They only insure in states they have been able to get permission to do so, and famously their largest state they are not allowed to use the score in because the voters in that state are quite reasonably worried that unaudited algorithms like this may be inaccurate or biased.
Is this "beta wide release" or "beta limited release" anyway?
Knightshade
Well-Known Member
So... once again you're SURE you know WAY more about how something you have no personal or professional experience with works than a company that literally does this as a line of business in a bunch of difference places.
Thanks for clarifying!
Thanks for clarifying!
I have lots of personal and professional experience with statistics, safety analysis, and risk evaluation, thanks. I literally do it for a living. One of the unique aspects is that I actually have to show my work to regulators, not just wave my hands over it.
It's easy to see that Tesla does not primarily consider the "safety score" as an estimate of your collision risk. It's a unitless, "beta" metric. Look at the Tesla insurance page: They can't even tell you how the safety score impacts your rates without adding in a completely undefined "safety score from trips," and it's clear how much they focus on mileage as an input and where you live (you'd think Predicted Collision Frequency would already have PREDICTED that...)
Are you really going to defend Tesla not including speeding or red light/stop sign running, yet including false forward collision alerts as a solid statistical model that will really predict their losses and that is PRIMARILY designed to set insurance rates, not filter for FSD "limited access, not wide release" BETA access?
Only time will really tell who is right. Tesla has not used this data for long enough to know if they will be able to keep it. A few months/years of losses from using this number and they'll have to change. Even Tesla screams "BETA" every time they mention it and Elon says it needs to evolve. Like they are already setting up to change it and know it is flawed. You're awfully sure that Tesla is well informed here for a company that slaps BETA on even their insurance offerings and has been doing it for 6 months.
What they should do is get rid of the "Being on AP doesn't count against you".
Since that's the only way I could get a good enough score and I used it even in inappropriate places and didn't bail until I thought it would kill me.
If they did that, a lot less folks would get in to FSD Beta. So then they could just lower the threshold to let in the number of testers they want. Which would probably be 90-95% score. To take AP out of the equation maybe ONLY miles NOT on AP count. Also probably not allow it to be one straight highway trip which is not an accurate measure. Basically only count City Driving with no AP. Since that's the target driving FSD Beta is after any way.
I don't think it's metrics are that far off. It's the other constraints that tilt things wrong. Following to close, braking and hard turns are probably correct most of the time.
They would get better drivers out of it, and less frustrated ones.
Right now they have a biased group that was a little crazy, and knew more how to beat it than actually drive good, like me, rather than truly better drivers.
And for the record, those itching to get on, I may be losing FSD Beta soon with a trade in. I am buying FSD again, but I won't be that itchy to get back in until they have ORDERS of magnitude improvements. It did satisfy my curiosity, but it's a big disappointment.
And you may ask, why buy it again. I'm totally convinced now that the FSD change lane with blinker completely replaces the lack of true Blind Spot. For that safety reason, I'm buying it again. I also do feel Tesla is giving me an extremely generous trade-in and I think it having FSD is part of it.
Knightshade
Well-Known Member
As opposed to how you're currently waving your hands over it right here?
It's even easier to see they absolutely do since they use it to set your car insurance rates
Tesla insurance is available to a population of over 100 million people--and continuing to expand to hundreds of millions more (they filed paperwork for an insurance company in Germany just a couple of weeks ago, which would let them serve most of Europe).... the majority of which have their insurance rates directly impacted by the score.
FSDBeta is currently available to about 60,000 people.
So the idea it's "primarily" an FSD filter, and not for insurance rate setting, is hilariously nonsensical
You'd think a guy who works with stats wouldn't be that bad at math.
Naah, I think that's already pretty clear
No, remember, Tesla sells insurance to 100M people (40M who they cannot use the score on) and the safety score is really accurate.
This is a pretty broken economic model. $12K is $12K, doesn't matter how you got it.
Knightshade
Well-Known Member
Does your vast training in stats make you aware that 60 million people is a larger number than 60,000 people?
It doesn't seem like it does.
Are you unable to parse "primarily"? They use mileage and location before safety score.
Of which 40M are blocked by law from using a score like this.
I never said Tesla DOESN'T use the safety score to calculate rates. I said it was not an effective actuarial tool. Kind of like Tesla sold you Full Self Driving in 2019 but that doesn't mean it ever will Full Self Drive. Tesla does a lot of things before they are sure they are correct. Move fast and break things.
The rules on this stuff varies by State. Like I believe it's not even allowed in California to collect driving metrics to set your insurance rates. I think it's around privacy issues. I think in many states the insurance company can do what it wants. It's not mandatory. As it is here for FSD Beta. But if the apps think you drive better and you get discounts some drivers will opt in. Also because it's imperfect data and more statistical, they purposely do not share to much information on the details. Otherwise they get too many disputes over individual events. There are some evolving standards around this stuff.
Most DriveSafe apps do including Speeding Metric, but they don't do red lights, because they don't have that metric. But Tesla does. I think Tesla purposely left that Speeding Metric out (of being Visible and part of score) for the FSD Beta use case. I'm sure they are collecting it for their database. And Relights too.
I am aware that Tesla has sold about 2M cars total, and only a fraction of those cars have been collecting data for less than a year, so it is completely irrelevant how many people live in the states they offer this product in, nor how many people they have let into FSD "limited access, not wide release" BETA. What matters is how much data they actually have and how well the model correlates the input data to outcomes. And it's "data" not driver behavior, given that we know the data is full of a lot of noise, and is particularly biased by the fact that Tesla has made the safety score opt-in and game-ified.
It appears that you constantly claiming "100M" is trying to make their product and experience seem wider than it actually is.
But the whole argument here is that the Safety Score is NOT for FSD Beta, it's to be an accurate predictor of crash "frequency" and thus is primarily useful to set insurance rates.
Yet you just said it is particularly tuned to FSD use....
Is this in the opt-in disclosure when you request a safety score? They aren't just allowed to collect any data they want.
I'm not that familiar how Tesla is using Safety Score for insurance.
Like I said, if the FSD blinker saves me from an accident, it's priceless.
If I was trying to apply pure economic sense I wouldn't be buying my 3rd Tesla (with or without FSD).
I was half joking that Safety Score was not for FSD Beta (I put wink after it).
But it would not surprise me if it has dual purpose. I would think they have better data by now under the insurance use case though where they have the whole loop of claims on insurance etc.
I'm just guessing they are collecting Speeding and red lights. Does anyone read the fine print on what they opted into. I didn't
If they are not using speeding for insurance, I bet they will eventually.
Knightshade
Well-Known Member
This is grossly false.
They've been collecting massive amounts of data from the fleet for many years... a user can opt out of much of the collection, but lots of owners have not.
The only thing new is that folks in the beta (and tesla insurance) are explicitly allowing that data to be directly linked to your identity (rather than being general fleet data- which if you're actually in any line of work you claim you know would still be useful for making risk correlations)
I get you say that.
You've just utterly failed to support the argument with...anything but the exact hand waving you claimed your "professional" made you totally qualified to NOT need to do.
Infinitely more than you it appears
Tesla's Safety Scoring has been a source of great frustration and is a typical example setting "classroom" standards to the real world. It essentially scores you for everyone else's bad driving habits.
Everytime I drive, I'm fully conscious of how the scoring system works and make every effort to drive accordingly. However, it's virtually impossible to prevent other vehicles from changing lanes or pulling out from side roads/parking lots too closely. But rather that evaluating my response to these situations, my safety score is down graded.
For example, once when driving with the autopilot engaged, a truck-trailer started changing into my lane. The Tesla autopilot didn't know what to do. It couldn't change lanes and didn't brake hard enough to avoid a collision. So, I disengage the autopilot and braked to avoid the collision. To my irritation, my safety score was significantly downgraded even though I took the proper corrective action.
This sort of scenario plays out every day to my great frustration.
Tesla's Safety algorithms are designed around a "classroom" environment and not the real world. For example, when a car changes lanes too closely in front me, the program should evaluate how I respond. Instead, it immediately downgrades my Safety score, even though I brake to increase the safety distance between me and the offending car.
Everytime I drive, I'm fully conscious of how the scoring system works and make every effort to drive accordingly. However, it's virtually impossible to prevent other vehicles from changing lanes or pulling out from side roads/parking lots too closely. But rather that evaluating my response to these situations, my safety score is down graded.
For example, once when driving with the autopilot engaged, a truck-trailer started changing into my lane. The Tesla autopilot didn't know what to do. It couldn't change lanes and didn't brake hard enough to avoid a collision. So, I disengage the autopilot and braked to avoid the collision. To my irritation, my safety score was significantly downgraded even though I took the proper corrective action.
This sort of scenario plays out every day to my great frustration.
Tesla's Safety algorithms are designed around a "classroom" environment and not the real world. For example, when a car changes lanes too closely in front me, the program should evaluate how I respond. Instead, it immediately downgrades my Safety score, even though I brake to increase the safety distance between me and the offending car.
valem
Member
The safety score is designed to determine insurance premium.
If you live in an area where you have to constantly take action to avoid a collision (people cut off in front of you etc.) then you are at a higher risk to end up in a collision.
Safety score is not about testing driver skills, but rather what are the chances they will be involved in an accident.
No matter how good of a driver one is, if you have to engage in evasive maneuvers 10 times a day to avoid a collision, you are statistically more likely to get in an accident than someone that engages in evasive maneuvers once a week.
Insurance premiums are based on statistics.
Tesla advantage in insurance is that they can calculate these statistics in real time.
With all that said, been driving for 6 months now with the safety score for FSD Beta and I am sick of it.
valem
Member
Is anyone on here (who is opted in, playing safety score & waiting for FSD Beta) NOT on 2022.8.x ?
Similar threads
