-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Elon: FSD Beta tweets
- Thread starter Terminator857
- Start date
JupiterMan
Active Member
or perhaps new cars....? are people more likely to look after/drive safely in a newer or more expensive car?
I'm sure everyone will have anecdotal examples, but statistically I don't know the answer.
willow_hiller
Well-Known Member
Sure. If there was a group of people loudly claiming every day that not driving with headlights on was killing children (and trying to get NHTSA to enforce headlight use during the day), then data showing that cars were safer with the headlights off would disprove those theories.
I'm not trying to answer the question "Is FSD Beta improving safety?" I'm trying to answer the question: "Are the current users of FSD Beta an inherent danger to others on the road?" And judging solely by the accident rate, you can positively confirm that the current fleet does not pose any more danger to other drivers than the average driver.
willow_hiller
Well-Known Member
In 2022, Tesla comprised roughly 3.5% of "light vehicle registrations". They're not even close to being representative of the vehicle market as a whole. Sure, millions of Teslas are on the road, but there are over 250 million cars on the road, and the Teslas on the road are the expensive ones.
None of this says that Tesla owners are better or worse drivers, only that they probably aren't representative of the average driver because the cars they're driving are expensive, electric, and whatever other traits might contrast with the average gas car - as well as a preference for those who can charge at home.
AlanSubie4Life
Efficiency Obsessed Member
I think you missed my point, which was that you definitely cannot confirm nor refute this with the data provided. There is no way to compare! Right?
Just think about what the data provided tells you. We just don’t know.
Specific example:
Personally, if I were in a situation where I felt an accident was more likely, or a particularly complex driving situation, I would often not be using FSD.
This is extremely funny. I wonder what he means by “a while?” Conveniently left undefined! (I would guess he means about 4-5 minutes, which is a while, all things considered.) I’m going to guess that with the next version I will need to intervene every couple miles on uncomplicated surface streets.
We have done this many times at this point.
Last edited:
ljbad4life
Member
Tesla has built BEVs for both Toyota and MB, so there's that
ljbad4life
Member
Looking at the Tesla Released numbers FSDb is having accidents at about double the rate of the og autopilot... there's tons of inferences that could be made from that and splitting hairs. I wish there was third party verification of the data.
Do other car makers release the statics of when their ADAS systems crash? comparing autopilot to another ADAS system might be interesting
Do other car makers release the statics of when their ADAS systems crash? comparing autopilot to another ADAS system might be interesting
willow_hiller
Well-Known Member
You're comparing a hypothetical driver with FSD Beta to a hypothetical driver without FSD Beta. You are correct in saying that this data does not allow us to know which is safer.
I'm not comparing hypothetical situations, I'm making an objective assessment of safety given the entire context of the situation. Your specific example is encompassed in that context. Because you choose not to use FSD Beta in a case where an accident is more likely, you are not increasing the overall probability of an accident by having the choice to use FSD Beta.
This is the statement I'm saying we can prove with this data:
"Given the cars being used, the people driving them, the places where they are driving, and the situations in which it is being used, the FSD Beta is not an unreasonable safety burden on other drivers, pedestrians, or other road users."
It's possible that FSD Beta is reducing safety, if you could hold all other variables constand and remove it from the equation, but we cannot know that from this data. But we can certainly know that given the context of the entire Beta test (drivers, times, places, situations, etc), it is not an unreasonable burden on society.
Last edited:
AlanSubie4Life
Efficiency Obsessed Member
We definitely cannot say that.
I am not saying that this is the case, but it is entirely possible it is 10x more dangerous than manual driving in those scenarios! It’s definitely a possibility (albeit unlikely in my opinion), with the data we have available. We just don’t have the accident rate to compare to, so we don’t know.
That would potentially be an unreasonable safety burden, even if the overall increase in accidents were low.
Really was the point of my original post. There are driving situations which should be extremely safe, and some that are quite a bit more hazardous. What if FSD use were positively correlated with safe scenarios? (As you mentioned we already know some factors you list are, but some are not known - and I think they are likely to be more important than driver, car age/condition, etc..)
Last edited:
Todd Burch
14-Year Member
Look at the operational domain.
Autopilot is highway-only, while FSD is surface streets (until very recently). Since the accident rate for surface streets is significantly higher than highways, I actually would have expected FSD to be much more than double the accident rate of highway driving.
willow_hiller
Well-Known Member
I don't know what about my statement you think is not provable with the data provided.
There are 0.31 accidents per million miles where FSD Beta is engaged; whether you want to dispute the accuracy of that fact is separate from the concept that it's a low level of accidents and objectively safe. This figure of 0.31 accidents per mile takes into account all the times that humans intervened and prevented an accident. So it doesn't matter if hypothetically FSD Beta is 10x less safe than manual driving, the end result of the entire Beta program is objectively safe.
EVNow
Well-Known Member
AP is highways only - FSD (till recently) was city only. We should expect more accidents in the city per mile than highways.
As my MYP is 10x faster then my previous Honda I will (via anonymity) say I drive Much faster now and more often. Hypothetically
I don’t know either, but those are 2 different questions.
I think people are more likely to look after expensive cars.
I don’t think people are more likely to drive safely, that implies people do not drive safely in less expensive cars.
Of course those are anecdotal based on my own habits.
With NoA highway driving now incorporated into FSD when FSD is engaged I wonder if Tesla will breakout FSD accident rates between highway and city streets in the next report? With FSD for highway markedly improved over NoA the results could be telling and influence the take rate.
No piece of data can alter real-world experience. FSDb is NOWHERE CLOSE to being as safe as a human driver, much less safer. First off, no one is going to leave FSDb engaged going through a wreck. The car screws up (which it does a lot) and the human disengages right before the accident. Now, this accident is counted not as one while on FSDb, but one while bringing driven by the human. The numbers are false and misleading.
The most important thing about avoiding incoming wrecks is reaction time. The average human reaction time is 0.25 seconds. FSDb is at least double or even triple this number (maybe more tbh). This is absurd. I understand that latency in hard to combat, but this is issue #1 (with lane selection being #2) that they need to conquer to become anywhere close to calling FSDb a “safe” product.
I’m no Dan O’Clown. I do like products that are good and work as advertised though. Someone who reads these numbers may think that they will only have to disengage 0.31 out of every million miles. We all know that is not true. Let’s say it is true though. FSD still would constantly piss people off with its slow and jerky behavior at nearly every unprotected turn.
Oh well. Self driving in 2069 I say.
The most important thing about avoiding incoming wrecks is reaction time. The average human reaction time is 0.25 seconds. FSDb is at least double or even triple this number (maybe more tbh). This is absurd. I understand that latency in hard to combat, but this is issue #1 (with lane selection being #2) that they need to conquer to become anywhere close to calling FSDb a “safe” product.
I’m no Dan O’Clown. I do like products that are good and work as advertised though. Someone who reads these numbers may think that they will only have to disengage 0.31 out of every million miles. We all know that is not true. Let’s say it is true though. FSD still would constantly piss people off with its slow and jerky behavior at nearly every unprotected turn.
Oh well. Self driving in 2069 I say.
According the methodology posted by Tesla for Autopilot at Tesla Vehicle Safety Report | Tesla, they include any accident in which AP was in use within five seconds of the accident. I seems reasonable that Tesla uses the same methodology for FSDb accidents. It would be foolhardy to include only those crashes in which FSDb was active at the instant of collision.
EVNow
Well-Known Member
They are not misleading. Your interpreation of data is wrong.
FSD + Human is safer than just Human. Says nothing about the safety of FSD alone.
Just posted this in the other thread.
AlanSubie4Life
Efficiency Obsessed Member
I do not dispute the accuracy of that number at all.
What if (hypothetically) FSD Beta were used for 50% of vehicle miles traveled ?
What would happen to the overall accident rate for Tesla FSD users?
Relevant to the topic:
What is the overall accident rate for Tesla owners? (We’d need this and some other info to calculate how much of an increase in accident rate FSD Beta hypothetically there would be.)
I think it is very likely lower than 0.5 per million miles since the No Active Safety comes in at 0.6. And many owners use AP for their freeway miles.
But, notably, we don’t seem to know. I looked at the link above to the quarterly safety data (0.2 w/ AP and 0.71 w/o AP for just one quarter so not directly comparable) , and at the data in the Impact Report (Warning: 200MB).
Anyway everything is way hazier than it needs to be. To be clear, I suspect FSD is fairly safe as currently implemented. But I just don’t know!
It would be cool to have a comparison of the FSD Beta users’ overall accident rates vs. the peasants without FSD (but with AP) for example.
Again, this is most definitely NOT known! We do not have the data. It has not been provided.
Similar threads
