Comment by slownews45 | Hacker News Reader

Comment by slownews45 | Hacker News Reader

Parent Full thread

Then don't drive with other drivers - NON-autopilot drivers eating, texting (constantly!), seeming to look up directions and much more (falling asleep).

https://www.youtube.com/watch?v=cs0iwz3NEC0&t=2s

The fact that you promote this absolute recklessness and demand users turn off safety features that may, net net save lives is depressing.

Finally, folks get very confused between things like autopilot and other features (autosteer beta with FSD visualization) etc. I think some teslas now even show a message warning that cruise control will not brake in some cases? Anyone have that.

I've been following the unintended acceleration claims around tesla's as well. Most seem pretty bogus.

For what its worth here is data we currently have from Tesla:

"In the 2nd quarter, we recorded one crash for every 4.41 million miles driven in which drivers were using Autopilot technology (Autosteer and active safety features). For drivers who were not using Autopilot technology (no Autosteer and active safety features), we recorded one crash for every 1.2 million miles driven. By comparison, NHTSA’s most recent data shows that in the United States there is an automobile crash every 484,000 miles."

So you want to go to a 10x increase in crashes. (1 per 484 vs 1 per 4M). I realize this is nt one to one, but even outside of this we are seeing maybe a 2x improvement (if your normalize for road conditions etc).

Nission Maxima had something like 68 deaths per million registered vehicle years. Will be interesting to see how this comes out for Teslas.

That data is not comparable, the autopilot in question there can only be activated in simple situations like a straight highway when you aren't changing lanes. While the human data is for all scenarios.

Tesla is not being transparent enough with their data for us to know anything.

No one has shown any credible data that the fatality rate driving a tesla is higher than anything. They go out of their way to ignore cars getting into pretty clearly lots of accidents.

My expectation would be that Suburu WRX / Infinit Q50 / Elantra GT type drivers are crashing more than folks using autopilot? Maybe ban them first?

Anyways, we should in a few years get some NHTSA data on this, though there is some speculation that given their approach towards tesla if its possible they will delay it and stop updating vehicle model statistics if its favorable towards Tesla.

Here’s some credible data that your assumptions are probably wrong:

https://medium.com/@MidwesternHedgi/teslas-driver-fatality-r...

The comparison there is as bad as elons in the other direction. They are digging into every missing bit of data and mistake for tesla in a different data set than other brands. I wonder of you could redo now with official data.

That page is not credible at all.

IIHS reports are publicly available [1] and the numbers there don't match what's in the article at all. Large/very large luxury vehicles have an overall fatality rate of 16-21, ranging from 0 to 60 depending on model. The overall average for all vehicles is 36. The Audi A6 for example, is reported as having 0 fatalities in the article, while in the report the actual number is 16.

The other source used, tesladeaths.com, lists a ton of accidents where AP was not deemed responsible. It actually states the real number if you pay attention - 10 confirmed deaths in the lifetime of Tesla as of 2021 - yet somehow the article claims 11 deaths up to 2016.

[1] https://www.iihs.org/api/datastoredocument/status-report/pdf...

That's one subset.

What people can credibly claim, because Tesla's stats are entirely loaded and biased, is that any claim AP is safer than humans is just that, a claim. Because AP will (usually) disengage in situations where it won't do well, something humans can't.

The closest you could come (maybe, who knows, because Tesla hordes the raw data like Gollum, despite their "commitment to corporate transparency") is:

"In optimal and safe driving conditions, AP may be safer than human drivers, and in less optimal conditions, uh, we don't know, because it turns off. And we won't count those miles against AP."

Tesla should actually, if they have the courage of their convictions, show how many miles are driven where AP/FSD would refuse to engage.

> Tesla's stats are entirely loaded and biased

Do you know that, or do you suspect that?

> how many miles are driven where AP/FSD would refuse to engage

I've used FSD for the vast majority of the nearly 40k miles on my 2017 Tesla model S. Highways, boulevards, side streets, even dirt roads on occasion.

It's a powerful but subtle tool. Used correctly, I have absolutely no doubt that has made my driving experience safer and less stressful.

It definitely requires an engaged, alert driver.

Where I suspect you and I agree is the impact of Musk's language and claims around the technology.

If he would fully shut the hell up about it, I think it's quite likely that there would be way less ill will toward the product.

It's known. The logical fallacy is accurate re "miles driven" versus "total possible miles". Tesla's statisticians cannot possibly be unaware of this if they have anything beyond a middle school education, yet they repeatedly continue to tout this stat without so much as a footnote.

I realize that may still, to some, be "suspect", not "known", so yes, if you're asking, "Is there public or leaked internal documentation saying 'we realize these stats are misleading and don't care'", then no, there's not.

The burden of evidence is on Tesla to show their car is significantly safer in an apples-to-apples comparison. They haven't done that so far.

And this is actually quite intriguing: even if Tesla was 10x safer, it would be extremely difficult for them to prove it. Because they'd have to do a multi-year feature freeze on Autopilot etc. while collecting data.

See in total, Tesla has sold around 2 million cars. Comparable cars have a fatal accident rate of around 10 per 1 million vehicle years. So the target for Tesla with the current number of cars is to have less than 2 fatalities per year, on average. To show with statistical significance that you have achieved that, you would need data for around 5 years without doing any changes. Maybe only 3 or 4 years given that the number of Tesla's keeps increasing, but still.

Really, it's only when you're doing 10 million cars per year, like Volkswagen or Toyota, that you can make statistically meaningful statements about safety while also updating your car's software frequently.

> NON-autopilot drivers eating, texting (constantly!), seeming to look up directions and much more (falling asleep).

FSD is something people are _encouraged_ to use, not _discourage_.

This is an entirely disingenuous take.

And Tesla's stats have been, _repeatedly_, shown to be entirely disingenuous.

Human drivers don't have a "disengage" mode, other than pulling off the road, when conditions are "suboptimal". AP disengages. FSD disengages. And suboptimal conditions are, surprise, surprise, the conditions in which more incidents occur. So the whole "AP is safer because when it's safe to use AP it's safer" circuituous reasoning, and every stat Tesla publishes, while withholding as much information from the public as it can (while touting their "corporate transparency is a core value" spiel), should safely be thrown in the trash.

> I think some teslas now even show a message warning that cruise control will not brake in some cases?

In my experience, that warning about not braking occurs when you are in TACC mode but are pressing on the accelerator to go temporarily faster than the set speed.

Unfortunately some people may have gotten into the habit of hovering their foot over the accelerator to be ready to override phantom braking, so maybe some even rest their foot on the pedal without realizing.

Fortunately, removal of radar will reduce the frequency of phantom braking, so hopefully this habit will fade.

Reply on news.ycombinator.com