Live data from Hacker News

Tesla's 2M car Autopilot recall is now under federal scrutiny

arstechnica.com

51–60 of 90 posts

Re: Tesla's 2M car Autopilot recall is now under federal scrutiny

#51

Earlier quoted context omitted.

It can be best in class, yet still rarely deploy... It can even be best in class when tested, rarely deploy, and save more lives than competitors. It just needs to correctly detect cases like this[1]. [1]: https://youtu.be/MTX0MvqBqL0?t=10

You claim that Tesla is gaming the system. I'm not aware of a definition of gaming in which the system is overall safer or equivalently safer than other top-tier competitors. That seems objectively better than deploying airbags when they will harm the occupants or total the vehicle when not-needed.

It only makes sense if everyone else is gaming the system, and then it just proves it's still the best lol.

Re: Tesla's 2M car Autopilot recall is now under federal scrutiny

#52

> , it only records an accident if the airbags deploy, I've often wondered if Tesla is gaming the figures by only deploying the airbags in really bad crashes, whereas other manufacturers might deploy the airbags for pretty minor bumps. This in turn makes them look really good on 'number of accidents where airbags deploy'.

That's how airbags work in general.

A slight bump won't set off your airbag in any car.

Re: Tesla's 2M car Autopilot recall is now under federal scrutiny

#53
post #41

Earlier quoted context omitted.

> The photo is of a Jan 2018 crash So still more than 5 years out of date. Would you accept a photo of a 5 year old iPhone on an article about some iPhone limitation? Would you accept a photo of a 5 year old version of Uber on an article about some Uber limitation? Would you accept a photo of a 5 year old robot vaccum on an article about some robot vac limitation? These things all change drastically each year. Five y…

It's a picture of a crashed car. Unless you're claiming the system doesn't crash now, a picture from 2018 is just as good as a picture from last week. It's not a main piece of evidence, it's an example. The evidence is in the text of the article where it talks about hundreds of crashes. As to your questions, as long as the issue in the article applies to the model in the picture, all of those are fine. I'd want a new…

It's like showing a video of an iPhone taking ages to load a web page or a very grainy photo that it took.. but it's from a 5 year old iPhone.

FSD has come a LONG way in 5 years.

https://www.youtube.com/watch?v=43Lrrhn0CMk

Re: Tesla's 2M car Autopilot recall is now under federal scrutiny

#54

> , it only records an accident if the airbags deploy, I've often wondered if Tesla is gaming the figures by only deploying the airbags in really bad crashes, whereas other manufacturers might deploy the airbags for pretty minor bumps. This in turn makes them look really good on 'number of accidents where airbags deploy'.

Tesla frequently loses telemetry in serious and fatal crashes [1]. For a specific example you can see here [2] listed as 13781-3074 where the fatal crash was detected by a complaint rather than telemetry. Of reported conclusively fatal accidents, only ~45% were detected by telemetry.

From that alone, we can conclude that their telemetry is bad at fatal crash detection. However, when we contrast it against their regular crash reporting where ~90% are telemetry, we can further conclude that either crashes are massively under-reported or fatal accidents are uniquely under-reported (even ignoring the unknown base rate vs reported rate problem). In either case, the massive unexplained discrepancy in the most visible, fatal crashes demonstrates that their data collection procedures are grossly inadequate to conclude any positive safety statistic.

The fact that Tesla repeatedly markets safety statistics based on grossly inadequate data collection procedures without any attempts to identify the true, higher rate, and without qualification is criminally repugnant.

[1] https://static.nhtsa.gov/odi/ffdd/sgo-2021-01/SGO-2021-01_In...

[2] https://www.washingtonpost.com/technology/interactive/2024/t...

Re: Tesla's 2M car Autopilot recall is now under federal scrutiny

#55
post #30

Earlier quoted context omitted.

Yeah, except of those people who have bought (at a significant price) those features, and understand that they're nowhere near their "aspirational" goals (When Tesla has been called out by the SEC and DMVs for some of Musk's more absurd statements, the company has quietly whispered to them "these are aspirational goals", and in one case, to the California DMV, "Musk's visionary statements are not reflective of engine…

The aspirational claims are not about Autopilot, but FSD (and I agree that having people pay for FSD in 2015 is borderline fraudulent. It is a consumer protection matter, not a safety matter.) Again, the statements to California DMV or SEC or Congress is one thing. Whether the owner is confused by the abilities of their vehicle post purchase and driving for a week is a distinct matter. > Those drivers are "it's okay…

> People break traffic laws all the time because they are selfish and it is convenient for themselves. Tesla drivers are no different.

Never said the first scenario is acceptable, and neither is the second. What's worse is this active "I'm going to let it do what it wants because it'll get better."

> Have you seen a single consumer, not an employee, who does this for training data for Tesla?

YouTube has plenty of examples. TikTok. "Training my Tesla" and similar searches will net you plenty of results.

Re: Tesla's 2M car Autopilot recall is now under federal scrutiny

#56

Earlier quoted context omitted.

It can be best in class, yet still rarely deploy... It can even be best in class when tested, rarely deploy, and save more lives than competitors. It just needs to correctly detect cases like this[1]. [1]: https://youtu.be/MTX0MvqBqL0?t=10

You claim that Tesla is gaming the system. I'm not aware of a definition of gaming in which the system is overall safer or equivalently safer than other top-tier competitors. That seems objectively better than deploying airbags when they will harm the occupants or total the vehicle when not-needed.

Tesla is not gaming the system by having a good airbag system that only deploys when necessary (that's actually really good!).

Tesla is gaming the system by excluding accidents where the airbags don't deploy from the "Autopilot/FSD accidents" dataset, thus artificially deflating the number of accidents.

Re: Tesla's 2M car Autopilot recall is now under federal scrutiny

#57

Earlier quoted context omitted.

You claim that Tesla is gaming the system. I'm not aware of a definition of gaming in which the system is overall safer or equivalently safer than other top-tier competitors. That seems objectively better than deploying airbags when they will harm the occupants or total the vehicle when not-needed.

Tesla is not gaming the system by having a good airbag system that only deploys when necessary (that's actually really good!). Tesla is gaming the system by excluding accidents where the airbags don't deploy from the "Autopilot/FSD accidents" dataset, thus artificially deflating the number of accidents.

This is a point you certainly could make. But it is not what the original commenter wrote.

Re: Tesla's 2M car Autopilot recall is now under federal scrutiny

#58
post #54

> , it only records an accident if the airbags deploy, I've often wondered if Tesla is gaming the figures by only deploying the airbags in really bad crashes, whereas other manufacturers might deploy the airbags for pretty minor bumps. This in turn makes them look really good on 'number of accidents where airbags deploy'.

Tesla frequently loses telemetry in serious and fatal crashes [1]. For a specific example you can see here [2] listed as 13781-3074 where the fatal crash was detected by a complaint rather than telemetry. Of reported conclusively fatal accidents, only ~45% were detected by telemetry. From that alone, we can conclude that their telemetry is bad at fatal crash detection. However, when we contrast it against their regul…

Recently disassembled a 2023 Nissan, and found that the telemetry/emergency crash reporting system has its own internal battery, sim card, and LTE antennas, allowing it to report a crash even when the cars electrical system is totally down due to large numbers of burnt/broken wires.

I wonder if Tesla will consider that for future models...

Re: Tesla's 2M car Autopilot recall is now under federal scrutiny

#59
post #33

Earlier quoted context omitted.

> it's also not Tesla's responsibility to nanny people Yes, it is. An L2 driver assist system must ensure that the driver is always actively supervising, and Tesla's systems failed to do that. That's why they recalled 2 million cars.

It seems to me that there are 10s of millions of "L2 driver assist" as defined by synopsys that have zero driver supervising checks such as most cars out there with lane-keep and cruise control.

And you believe that federal regulators have simply never heard of these other systems? A more likely explanation is that the driver attention monitoring in a Honda or Cadillac or whatever works better than Tesla's.

Re: Tesla's 2M car Autopilot recall is now under federal scrutiny

#60

One of the big things to remember from all of this is that "if airbags do not deploy, Tesla does not consider it an accident" when it is touting its (already inaccurate and misleading for other reasons) "safety". As an automobile manufacturer, there is no good conscience reasoning that Tesla is unaware that active restraint and airbag systems of today are far more nuanced and weigh multiple criteria when deciding to…

To wit, "incident was so catastrophic that restraint systems could not deploy"/"incident was so catastrophic it could not report" is not theoretical. In fact it is the case in the majority (yes, majority) of reported conclusively fatal crashes [1].

[1] https://static.nhtsa.gov/odi/ffdd/sgo-2021-01/SGO-2021-01_In...

Post reply on HN