> The auto safety regulator said the Tesla software allows a vehicle to "exceed speed limits or travel through intersections in an unlawful or unpredictable manner increases the risk of a crash." Tesla will release an over-the-air (OTA) software update, free of charge.
I'm not sure how to reconcile that statement with how FSD beta works. Fine, the speed limit thing is easy to fix-- just cap the max speed to the speed limit.
But the "[traveling] through intersections in an unlawful or unpredictable manner" is inherent to the FSD beta. Most of the time it does "fine", but there's some intersections where it will inexplicably swerve into different lanes and then correct itself. And this can change beta-to-beta (one in particular used to be bad, it got fixed at some point, then went back to the old swerving behavior).
I'm hopeful about Tesla FSD, and don't think it necessarily needs to be perfect, just significantly better than humans. So I'm rooting for Tesla FSD/Autopilot overall. I just don't see how given the findings there is a solution without removing removing the entire FSD feature.
If they accepted a software update, I would assume it would be about things that are deliberately programmed in, rather than just fundamental limitations? But I agree it's a bit odd. The longer description says
> The FSD Beta system may allow the vehicle to act unsafe around intersections, such as traveling straight through an intersection while in a turn-only lane, entering a stop sign-controlled intersection without coming to a complete stop, or proceeding into an intersection during a steady yellow traffic signal without due caution. In addition, the system may respond insufficiently to changes in posted speed limits or not adequately account for the driver's adjustment of the vehicle's speed to exceed posted speed limit. [https://www.nhtsa.gov/vehicle/2020/TESLA/MODEL%252520Y/SUV/A...]
"entering a stop sign-controlled intersection without coming to a complete stop" sounds a lot like the thing they already issued a recall over in Janary 2020 (further down the same web page), so it seems a bit odd that it would still be an issue. And "due caution" to yellow lights seems like it could just be tuning some parameter. On the other hand, failing to recognize turn-only lanes sounds more like a failure of the computer vision system...
Seems like they should come up with a new phrase to distinguish between making modifications to the car at a dealer or service provider vs. a downloadable update, right?
Yes, this is just a historical term that nobody bothered to change.
We have to remember that Tesla was the first company to really do this and even today almost no other company does it. Most can update some parts of their system, but almost non have anywhere close to the integration Tesla has.
So for 99% of recalls, it is a physical recall, its just Tesla where most of the time it isn't.
Almost all of these "recalls" that make the news are just software patches. The only difference between these and a normal patch is that customers get an email about it.
I can understand why people might think that all these recalls require going back to the shop, that's how most legacy makers work to this day.
Because it gets more clicks. Reading about how Tesla will lose lots of money because they have to recall 350k vehicles is juicy story, specially if its about removing a major feature, a slight software update to that feature is boring.
I'm astonished consumers are still paying fifteen thousand dollars to be guinea pigs for this bullshit. No car that currently has tires on the road is going to let you fall asleep in your driveway and wake up at work.
> How many 9's do we need before we can say it's reliable enough to trust it? A few more, for sure.
Yeah, 1% is definitely not going to cut it. What are the odds of dying when a human is at the wheel? Something like 0.000025% if my napkin math is right and my assumptions in the right ballpark.
The technology is mostly fine. The bullshit is mostly the false expectations they set, and the subsequent risks they choose to take in implementation in order to minimize their expectations miss. Other driving assistance systems try to do less, but they do it honestly.
Does that mean "stay in the correct lane on the interstate and take the proper exits without hitting anything" or does that mean "begin inside your garage and end inside another garage five hundred miles away without touching anything"? The first one is nearly trivial.
It stops being bullshit when they stop telling the user they need to pay attention to the road, and not one second before.
I wouldn't describe it as trivial. My Tesla would phantom brake consistently in a lot of spots - overpasses were a very big trigger.
Also anytime I'd pass an exit I wasn't taking in the right lane it would veer and slow down aggressively, same with trying to speed match a car rapidly accelerating on an on ramp.
Bottom line, there's absolutely nothing that isn't driver assist; everything requires a significant amount of vigilance when you aren't driving in a straight line, in the left lane, on an empty highway with no one entering and exiting.
Imagine a scenario: someone's on a bike, running down the street towards the pedestrian crossing. There's a vehicle (truck, bus...) on the left covering the view. The cyclist swiftly crosses the road, the car doesn't detect anything as there's nothing to detect (because of the truck/bus), bam!
Many drivers would see the cyclist before and be very careful when passing the other vehicle (although we don't predict correctly 100% of the time either). I haven't had a chance to test any self-driving system yet so I'm legitimately interested, do those systems reliably detect such things?
Not to mention people (esp. kids) racing through intersections on electric scooters, casually ignoring traffic rules...
FSD requires you to be 100% in control of the vehicle for the entire drive, and paying full attention to the driving, and also do nothing.
This is demonstrably not a task that anyone can do, let alone Joa[n] Average car driver. Highly trained pilots are not expected to do that, and that's when autopilot is being used in a vehicle that can provide significant amounts of time to handle the autopilot going wrong - seriously, when autopilots go wrong in aircraft at cruising altitude pilots can have 10s of seconds, or even minutes, to handle whatever has gone wrong, Tesla's FSD provides people a couple of seconds prior to impact.
That said in countries other than the US people can reliably use trains and buses, which also means that they don't have to intervene in driving the vehicle.
> That said in countries other than the US people can reliably use trains and buses, which also means that they don't have to intervene in driving the vehicle.
Most of Europe doesn't have ideal public transport either, I'd imagine South America, Africa being the same or even worse in this aspect. It gets drastically worse the moment you want to go somewhere in the countryside.
Yeah I don't understand the "legitimate" use case. I would imagine people who actually shell out the money get their car onto the interstate then strap a weight to the steering wheel until their exit comes up. Having to watch the road without the stimulation of actively driving is worse than having no assistance at all.
I've missed trains that left a station ahead of schedule by a minute or two in my regional metro system. I've also had busses go past a bus stop earlier than the posted time.
The Swiss train up already does this. It also tells you exactly where to go with a plan of the station, how full it is where in the train, where the wagon for food is and tons of other things.
Beyond that, you may be surprised how many people would wake you up for your stop. Especially in US Commuter Rail systems people tend to follow similar habits and get to know each other over the years. Even on a normal city bus I quickly (months) got used to the pattern of the same people getting on and off at their various stops. If you’re asleep at your stop someone can usually nudge you
Hell, right now I'd pay a few bucks to disable AP so I could use the old school cruise control. I still don't understand Tesla's opposition for making that an option whether you have AP or not.
they meant "save a few thousand dollars so my Tesla can just have the ordinary cruise control that it comes with, without the autopilot which I believe to be unusable"
Prices and packages aside, I'm asking can you turn on basic cruise control without turning on any form of automatic steering?
To me, design hubris is forcing someone to accept a choice-limiting or poorly documented form of automation in order to accept a more basic form. I have a Fiat 124 Spider... love it, but it has one "feature" that I absolutely hate, which is that if you hold the clutch down and brake while stopped on an uphill incline, it holds the brake for 2 seconds after you take your foot off. Presumably this is for people who don't know how to balance clutch on a hill... although the overlap between those people and the people who'd buy this car must be vanishingly low. You can disable this dumb thing - but the only way to do it is to pull the fuse that controls antilock brakes. It's infuriating.
> Prices and packages aside, I'm asking can you turn on basic cruise control without turning on any form of automatic steering?
Yes.
BTW, the automatic steering is very easy to override by just turning the wheel. When I change lanes I just turn the wheel and then turn autopilot back on.
BTW, I also had a Subaru that held the clutch on a hill. I really appreciated it in San Francisco. Personally, I don't understand the hate. I never felt like I needed it, but I remember my mom complaining about other people rolling back in a steep driveway at my preschool in the 1980s. Maybe the feature is needed for those places in the world where everyone drives stick, even the people who are clutzes?
I learned to drive stick on the hills in LA, and owned a manual in SF... not rolling backward is part of being a good driver. I'm just saying I'd like the option to turn it off w/o disabling antilock brakes. For me, the feature creates unpredictable stalls or dangerous situations, because you never know if it's engaging. It engages on a 4% grade or something; it's hard to know if you're at 3% or not before you take your foot off the brake and find your car not moving as you let out the clutch. If you assume the brake is on and it isn't, you let the clutch out further to override it (also not marked how much throttle that takes) and you shoot forward into the car in front of you. If you assume it's off and it's actually on, you stall. People who live in parts of the world where there's only manual transmissions know how to drive a manual car. I think it was made for people who've only driven automatic but chose a manual this time. The Fiat 124 up until last year was the only car sold in America where the automatic model was still more expensive - an add-on. AFAIK.
I don't care what features they add, and I've heard other people not be bothered by it. Just give me a way to turn them off!
In the Subaru when it would engage it would turn on a light on the dashboard. It also only engaged on very steep hills where you would always be stomping on the gas no matter what.
I think the Subaru just had a much better execution of it than the Fiat.
Edit: Maybe call the dealership and see if the level sensor can be adjusted? It really should only hold the brakes if you are on a hill that is so steep that you are naturally worrying about rollback.
That's an interesting idea, mangling the level sensor ;)
The other quick-fix is to always hold the e-brake up just enough to turn on the light when you think the grade might be >3%. This is apparently because whoever designed the system assumed that people who use the handbrake on a hill know how to release it and drive better than people who don't - whereas if you grew up driving on really steep hills in the US you weren't taught that method. The other way around it is to hold the brake w/o the clutch, keeping the car in neutral, and then unbrake-clutch-shift-gas really fast before the car rolls backward (more dangerous, but acceptable as heel/toe practice). I really wish it only happened on a very steep hill; it's the very mild slopes that caused me the worst surprises when I first bought the car and almost got into several accidents.
The "still" is what surprises me. Back in 2016 or 2017, I was excited. Musk still seemed like a pretty stable person back then and Tesla had basically executed on its promises and Musk said that by the end of 2017 you'd be able to drive from NY to LA without a single touch on the wheel. Everyone had said that EVs were doomed and Tesla really showed otherwise. Why not believe him? I like being hopeful about the future.
At this point, it's looking pretty far off, especially for Tesla. Even Cruise and Waymo are having issues and they have far better sensors than Teslas. It seems silly to be paying $15,000 for something that realistically might not happen during my ownership of the vehicle.
Even if it does happen, I can purchase it later. Sure, it's cheaper to purchase with the car because Tesla wants the money now. However, I'd rather hedge my bets and with investment gains on that $15,000 it might not really cost more if it actually works in the future. 5 years at 9% becomes $23,000 and I don't think Tesla will be charging more than $25,000 for the feature as an upgrade (though I could be wrong). If we're talking 10 years, that $15,000 grows to $35,500 and I can potentially buy a new car with that money.
Plus, there's the genuine possibility that the hardware simply won't support full self driving ever. Cruise and Waymo are both betting that better sensors will be needed than what Tesla has. Never mind the sensors, increased processing power in the future might be important. If current Teslas don't have the sensors or hardware ultimately needed, one has paid $15,000 for something that won't happen. Maybe we will have self driving vehicles and the current Teslas simply won't be able to do that even if you paid the $15,000.
It just seems like a purchase that isn't prudent at this point in time. The excitement has worn off and full self driving doesn't seem imminent. Maybe it will be here in a decade, but it seems to make a lot more sense to save that $15,000 (and let it grow) waiting for it to be a reality than paying for it today.
I mean, to be honest, I'm completely unwilling to pay any additional cost for a license to proprietary software that on my vehicle, and I won't even consider a brand that offers paid software upgrades to their vehicles, entirely on principle. See the recent case around BMW offering subscriptions for heated seats. Just no, absolutely not. Reject faux ownership.
Both of my cars patched last night with “Bug fixes” as the changelog. I don’t think I’ve seen an update that only said that before. I suspect that was this “recall”. If that’s the case, then the recall (patch) is probably mostly done by now.
If the changelog to my car's safety critical system was limited to "misc bug fixes", I would never touch it again. Either have extreme transparency into the system, or don't have the system.
I leverage this complaint against my own car. The AEB radar just stopped passing start up self tests for a while, and I had no idea why, and could not find out without going to the dealership. While waiting to be able to do that, it started working again. I don't consider it a backup, or a helper or anything. It's an independent roll of the dice for an upcoming crash that I already failed to prevent. If it does anything in that case then that's probably an improvement, but I don't exactly trust it to save my bacon.
Yes, the description of the remedy in the recall says:
The remedy OTA software update will improve how FSD Beta negotiates
certain driving maneuvers during the conditions described above, whereas
a software release without the remedy does not contain the
improvements.
It seems like there is a huge disconnect between people who know about FSD from using it and people who know about FSD from things they read on the internet.
I have FSD, I use it every single day. I love it. If every car on the road had this the road would be a substantially safer place.
I'm reminded of a thread on HN where someone did a cross country trip and their car only phantom-braked a couple of times, and called that a huge success.
Ha, my new Model 3 phantom braked the first time on the trip home from the service center, which was only 10 miles. I doubt it had even gone over 10 miles on the odometer at that moment.
How were you able to put it into autopilot? Typically, the autopilot is calibrating itself the first 60 or so miles of driving, and it does not let you engage it.
Good question! It didn't argue one bit, literally drove it off the lot at the service center and hit the freeway and turned AP on (it's my second Model 3, so I already have such habits).
It's entirely possible it had more miles than 5 on the odometer when I picked it up. Officially the paperwork says 15, the associate who gave me the car said it was actually less, and the odometer isn't very prominent so I never even looked. Maybe it had 50 miles and was a reject ;). I should go check TeslaFi, since I reenabled that the day I bought the car. I confess that the only way I ever know how many miles my car has is when I see it on TeslaFi.
It doesn't really matter, we have multiple incidents of FSD causing accidents due to outright mistakes, not even "a human would have messed up here too" situations.
Sure, but say I consider myself a diligent driver and I've never caused an accident... what does FSD have to offer me? Yet another random opportunity for a car to fail me. Why would I surrender control for that?
I’m the type of person who keeps a lot of car lengths between me and the car ahead like you are /supposed to/ but I’ve also had someone panic stop in from of me and, due to road conditions, I cannot stop in time and have hit them. Not hard. But did all the same.
FSD and similar aren’t magic bullets that would prevent that. People like you who seem to think they are are also part of the problem.
Would a completely automated roadway be fucking awesome for safety and productivity and relaxation? Yes. Is it going to happen anytime soon? Doubtful.
Blows my mind how many people are totally willing to risk their own safety and everyone else’s using known buggy as fuck software. On the road. At 70mph. In a 2 ton vehicle.
… but it hasn’t been an issue for you! You say. Ok.
But I also watch dumb asses lane hop with barely enough room to pass likes it’s a video game and then act surprised when they wreck. So I guess I’m giving people too much credit.
If someone brake checks and you hit them, you're just as much of a problem. Learn defensive driving and stop driving so close to another driver. Keep a safe distance.
But this backlash to my comment about hard braking and wrecks is kind of hilarious.
Basically, my point is that Tesla has made egregiously bullshit claims about FSD and autopilot. They have given them misleading names. And people who want to “live in the fuuuuture” have eaten it up and are using the Teslas like they are actually capable of driving themselves 100% of the time. In all conditions.
If anything, it has made people who rely on it seemingly worse, inattentive drivers.
So while I’m glad y’all enjoy your purchased vehicle, don’t pretend all gripes and complaints lobbed at them are invalid or someone else’s fault. That’s asinine and childish.
I have the same experience. I think if there was more transparency in the data around FSD related accidents the conversation would be different. The last evidence we have is Tesla stating that there have been no fatalities in 60 million miles of driving under FSD. Pretty good so far.
Yeah, let's now hear the stats on "fatal accidents within 30s of FSD being enabled".
You really just can't trust Tesla at all, about anything. There's no integrity there. They won't even be honest with the name of their headline feature!
For Tesla's collection of data on accidents with Autopilot technologies, they considers Autopilot to have been enabled during the crash if it was on several seconds beforehand. These systems may hand control back to the driver in crash scenarios moments before impact, but that doesn't mean the data will report that the system was not enabled for the incident.
There is no such evidence beyond unsupported proclamations by the vendor who, by the way, has compulsively misrepresented the capabilities of the product for years. The only evidence available that is not completely tainted by a complete conflict of interest is basically Youtube videos and self reported information [1] by investors and super fans that on average show critical driving errors every few minutes and even those are tainted by self interest to over-represent the product.
Tesla’s statements should only be believed if they stop deliberately hiding the data from the CA DMV by declaring that FSD does not technically count as a autonomous driving system and is thus not subject to the mandatory reporting requirements for autonomous systems in development. If they did that then there could actually be a sound independent third party statements about their systems. Until that time their claims should be as trusted as Ford’s were on the Pinto.
Absolutely need more data, but if a public company makes untruthful claims about the past they are in big trouble. They are mostly free to make claims about the future without much consequence.
This is why startups are allowed to project massive sales next year but Elizabeth Holmes is headed to jail.
If it turns out that FSD has had a fatal crash and Tesla lied about it Musk is headed to jail too.
We don’t live in a “just world”. There’s a reason a lower tier person like Holmes gets some prison and psychos like, say, Trump or Musk barely get a slap on the wrist.
Regardless of whether or not that's true, that's also not the impressive number you think it is. Human-driven cars have a fatality rate of about 1 per 100 million miles travelled, and that's including drunk drivers, inattentive drivers, and all environments, not just the ones people are comfortable turning on these features in.
FSD is terrible. I have it, it's dangerous. If I had to go 5 miles through city streets I'd sooner trust a human after 3 shots of tequila before I'd trust FSD every time.
That other drivers are beta testers for software controlling a large mass of metal and combustibles hurtling down asphalt at _ludicrous_ speeds is something you'd write into a dystopia story.
"This study investigated the driving characteristics of drivers when the system changes from autonomous driving to manual driving in the case of low driver alertness.
...
In the results, a significant difference was observed in the reaction time and the brake pedal force for brake pedal operation in the low alertness state, compared to the normal alertness state of the driver, and there were instances when the driver was unable to perform an adequate avoidance operation."
Not sure if it's the latest, and nowadays I rarely use it, and when I do it's just to show off to curious friends, but I live in an east coast metro and I've never been able to use FSD for more than 20 minutes without it making a dangerous mistake - 90% of the time it's been a mistake making a left turn, turning into the wrong lane or oncoming traffic after making the left, not correctly respecting the traffic lines or the left turn lights and other very scary situations.
So it’s good enough that you let it control and become less attentive?
I was thinking that you feel in what kind of conditions it behave correctly (highway), low traffic and then use it in progressively more difficult situations as it prove to you it can handle them consistently.
I know it’s not ready yet. I heard as saw so many positive reviews (but still with corner cases and strange behaviours)
I've not had any FSD issues on the highway besides phantom braking, it keeps lanes correctly, safely adjusts to traffic speeds and it merges into offramp lane without issue (though it still makes me nervous), once it gets into city streets though...
Disagree. I've had the beta for a year and drive in NYC. It has been mostly flawless considering all scenarios it has to handle. It does make the occasional mistake but (1) it's a beta and (2) I'm responsible for monitoring and correcting it. Agree that those that expected more should be entitled to a refund if they so desire.
Consider the possibility that you might both be right. Two people may have vastly different experiences with FSD, depending on the routes they take and other variables.
Ok but don't you have to weigh that against the potential of avoiding accidents by reacting quicker then you as a human can?
I have defiantly seen cases where FSD Beta stopped and the driver didn't understand instantly understand why.
Or simpler cases where the car follows a lane while the driver wasn't paying attention (adjusting the radio or whatever). Those can easily lead to an accident but are unlikely to if you are in FSD.
All the available scientific evidence suggests these systems diminish driver focus and attention which definitely makes them less safe since these systems can't be operated safely without an attentive driver.
I have seen some evidence bases on some limited lane keeping systems that suggest this. But not on systems that are as advanced as FSD Beta for inner-city with the same amount of forcing the drives attention systems.
It seems there’s a huge disconnect between people who know how to drive and are confident in their skills, and those who are scared to drive and want to offload it to Musk.
I’m actually afraid to drive behind a Tesla and either keep extra distance or change lanes if possible. I still have more faith in humans to not randomly brake them in beta FSD.
It’s one thing to put your own life in the hands of this beta model, and it’s another to endanger the life and property of others.
Yeah it's not that I'm scared of my own driving, I mean that if other people had it it would be safer. Things which have happened in the last 3 days of moderate driving around a city:
1) A car gets impatient at some traffic turning right on a green light in a construction zone, and comes into my lane meaning into oncoming traffic. We have to swerve out of the way and slam on the brakes to avoid them.
2) A guy gets impatient behind me slowing down over a speedbump, tailgates me a few inches behind and then passes me by cutting into oncoming traffic in a no passing zone, then cuts me off, again with a few inches to spare, and speeds through the neighborhood where we eventually meet at the stoplight at the end of the road.
3) Every single day people run red lights. Almost every light where there are people turning left at a green there are 4-5 cars that go through the red light after it turns.
My safety score on my tesla is 99. I am an extremely safe driver. I wish FSD was more common because so many people are terrible drivers.
Your Tesla tells you you're a safe driver. Well, geez, I'm convinced that the $55,000 (maybe+?) you spent tells you things you like to hear. I mean, one of the criteria is "Late Night Driving". Why not driving between the hours of 4pm and 6pm? When there are a lot more cars on the road, which is just statistically less safe.
And looking at the rest of the criteria, they're ok, but hardly comprehensive. This is like, the bare minimum. It doesn't measure what I would call "doing stupid shit". Like crossing several lanes of traffic to get in a turn lane. Forcibly merging where you shouldn't. Nearly stopping in the middle of traffic before merging into a turn lane at the last minute. Straddling lanes because you're not sure if you really want to change lanes or not. Making a left turn from a major artery onto a side street at a place that is not protected by a light. Coming to a near complete stop for every speed bump, then rushing ahead to the next.
And a host of other things that demonstrates the person does not consider other people on the road at all.
Here's an entire article about how to game the safety score:
One of the tips is to "drive around the neighborhood when traffic is light".
And the car doesn't ding autopilot for behaviors it would knock a human for. Because the assumption is that the car would know better I guess. But then why isn't the safety score simply a deviation from what autopilot would do in a situation? If autopilot would brake hard to avoid a collision, shouldn't you?
The Tesla doesn't just tell me I'm a safe driver, it also gives me a cheaper insurance rate due to the safe driving. The safety score is related to my monthly insurance premium.
So maybe it's to stroke my ego? But they're also putting their proverbial money where their mouth is.
The insurance you get from Tesla, which has a vested interest in its own Safety Score. It's not giving a cheaper insurance rate "due to the safe driving", it's giving a cheaper rate due to having a better score on the metrics it decided.
You see how that's circular, right. It does not mean you are a safe driver.
…what? No I am not following that. The safety score is a representation of the metrics used to determine the insurance rate. This is not in any way circular. The insurance rate and the safety score are representative of the same thing: safe driving.
They're both from Tesla. There's no real proof that they're representative of safe driving. And you kind of want it to flow both ways: "the safety score represents good driving, so I get cheaper insurance" and "I get cheaper insurance so the safety score must represent good driving".
You're trying to use each of these things to validate the other so you can then claim it's something entirely else.
I think it's telling that autopilot is allowed to drive in a manner that would otherwise negatively impact your safety score. Either autopilot is unsafe or the safety score doesn't measure actual safe driving.
You are talking to someone who is excited about… The Future (tm). And will defend their purchase and choices to the death. So I doubt there is any reason to continue. The fact they even gave us their Tesla given “safety rating” like it was something to give a shit about says a lot.
We are all glad your car is giving you head pats and game badges.
Also, the Tesla driving score reminds me of the Windows “Experience Index” score. Which was complete and utter toss. But made people feel good about their systems.
Man I think you're really confused about what the "safety score is". Here it is stated a simpler way:
Tesla tracks driving style and gives you an insurance rate based on those driving habits.
My insurance rate is about $90, which is the lowest the rate goes, because the things it tracks (following distance, for one) corresponds to a lower liability risk for the insurance underwriter, which in this case, is tesla.
The monthly price is directly correlated to driving habits. You seem to be getting confused about "safety score".
I'm genuinely confused as to what about this doesn't make sense to you.
Tesla is the underwriter. They give a monthly premium based on driving habits.
Tesla is the insurer. Tesla is also the manufacturer of the insured item.
You are using as proof of the claim, the fact that the group making the claim says they're right. This is big "We've investigated ourselves and found nothing wrong" energy.
> I’m actually afraid to drive behind a Tesla and either keep extra distance or change lanes if possible.
Irrespective of the Teslas or FSD:
If you're afraid of an accident due to the vehicle in front of you braking regardless of the circumstances, then you're following too closely. It doesn't matter if the braking event is anticipated or unexpected. If you're not confident in your ability to avoid an accident if the vehicle in front of you slams on their brakes then you are following to closely.
> If you're afraid of an accident due to the vehicle in front of you braking regardless of the circumstances, then you're following too closely.
I know what you're saying, but that is not what I'd meant. Also, I don't follow too closely.
The difference here is that almost always a driver will brake depending on the happenings in front of them. So, if you pay attention to not only the car in front, but in front of them and in the neighboring lanes and so on, you can sense and also detect patterns in how a particular driver is driving. There are many skittish drivers who brake every 1 second and some don't, and so on. Basically based on your driving experience you can predict a little.
The problem here is that this stupid POC FSD will brake randomly or change lane randomly or whatever, so there is no way you can predict, and hence my concern and issue with it. I just prefer to change the lane, but that's not always an option.
> Basically based on your driving experience you can predict a little.
Yes, we all do that AND you're using that predictability to take liberties in safety such as following too closely. FSD's unpredictability exposes the vulnerability in your driving process and makes you feel uneasy.
You (and everyone else) can follow too closely AND FSD can be an unsafe steaming pile of crap. It's not an either or situation.
I have it (subscription, just wanted to try it out). I won't use it again. It confidently attempted to drive me into a rather solid object with very little warning. If every other car on the road had this, I would sell all my cars and forbid family members from getting anywhere near the road.
To me it's weird you can voice an opinion like "I won't use it" and then say it's nice on the highway, as if you've given it enough of a trial there to endorse it when you think otherwise it's going to be killing people.
Not that weird, it's just weird to recommend it after you've seen it almost kill you. I doubt he's put in the requisite time/miles/effort/observation/whatever to know that it does great on the highway but not elsewhere.
I definitely agree, in which case I wouldn't find myself endorsing it for any use. Just because I drove X amount of miles without a problem on the highway doesn't mean it's any less likely to drive into a solid object, why would he recommend it?
I have a feeling the differences between experiences is heavily influenced by where you are located. There is no way they are adequately able to train across all of the locations FSD is expected to be able to work.
Australian here. We drive on the "wrong" side of the road. I have no idea if FSD even accommodates this different scenario. And if it did, and being from Melbourne, having diabolical yet somewhat logical hook turns (turning from the furthest left lane to turn right) I will bet would not even fall in FSD scope.
I suspect that the NHSTA know about FSD from all of the above and more.
Even if Tesla’s current implementation were objectively safer than the average human driver, it would still represent a net negative on safety because of the negative impact the glaring flaws will have on peoples’ confidence in self driving tech.
In fact, there is a huge disconnect between people who design similar systems (Vision, safety, self driving, robotics, simulations) and those who use them. The criticism of Tesla is coming from experts, not jealous competitors.
The problem with systems where 99% is a failing grade, is that 99% of the time, they work great. The other 1% you die. No one is against FSD or safer tech. They are against Tesla's haphazard 2d vision-first FSD.
Wanna hear about this new fangled tech that avoids 100% of accidents, has fully-coordinated swarm robotics between all the cars, never swerves and can operate in all weather + lighting conditions ? It's called a street car.
The words 'road' and 'safety' should never be uttered in the the same sentence.
Um... no one died. I understand your point is hyperbole, but you're deploying it in service to what you seem to claim is serious criticism coming from experts. Which experts are claiming FSD beta killed someone? They aren't. So... maybe the "experts" aren't saying what you think they're saying?
> Wanna hear about this new fangled tech that avoids 100% of accidents, has fully-coordinated swarm robotics between all the cars, never swerves and can operate in all weather + lighting conditions ? It's called a street car.
And this is just laughably wrong. Street cars (because they operate on streets) get into accidents every day, often with horrifying results (because they weigh 20+ tons). They are surely safer than routine passenger cars, but by the standards you set yourself they "aren't safe" and thus must be banned, right?
I think a large issue is people are really bad at risk assessment.
Also, why the hell did they try doing camera only? Fucking dumb as hell. I would have trusted something like FSD a lot more if they had also put in very advanced radar etc. trying on fucking /cameras/ only for newer models should be a criminal offense.
When the model S was at its sales peak from 2013-2020, Lidar was still pretty expensive. So he put cameras in the original Teslas.
When Karpathy took over in 2018-ish, Tesla realized that all their advantages (dataset size, quality of researchers) was in the 2D realm. Waymo had been doing lidar for longer, seemed to have better 3d engineers & could keep upgrading lidar devices as they went.
Tesla's insistence on 2D cameras has to do with stubborn desperation more so than a willingness to play catch up. Elon needs 2D cameras to 'win' because even as early as 2019, he thought that it was too late to start playing catch up in 3D.
Ditto. Most fun I've had in a vehicle in my whole life. A robot drives me around every day, and the worst misfeatures are that it's too timid at stop signs and left turns, makes poor lane choices sometimes (oops, can't turn, will go around), occasionally zipper merges like a jerk (yes, technically you can use that lane, but if you try you'll get honked at -- car doesn't care).
But the company is run by an asshole who people love to hate, so... everything it does is Maximally Wrong. There's no space for reasoned discourse like "FSD isn't finished but it's very safe as deployed and amazingly fun". It's all about murder and death and moral absolutism. The internet is a terrible place, but I guess we knew that already. At least in the meantime we have a fun car to play with.
This is a "me or your lying eyes" argument. It drives my kids to school. Why is it so important to you that I not enjoy the product I purchased and love? You realize you aren't going to win an argument like that, right? You think I'm going to just wake up and realize that it can't do what I see it do?
It doesn't matter how you rationalise it to yourself or how irresponsible you are in using it. Your car does not do full self-driving. It is stranded on Hardware 3. It will not get Hardware 4 unless a lawsuit enforces it: https://electrek.co/2023/02/15/tesla-self-driving-hw4-comput...
It literally drives itself. You're hiding behind a semantic argument about what "full" means, and how autonomy is deployed. It's the "Money Printer Go Brrr...." meme.
You don't seem to be able to understand that the product as deployed has value to its purchasers, and that's leading you to some really emotionally dark places. I mean, it shouldn't be this upsetting to you that I like my car, and yet it clearly is. Just... maybe just don't? Let it go. You know these cars are safe as deployed. Let the Tesla nuts have their fun.
> that's leading you to some really emotionally dark places
Okay, this is just sad now. It's like a kind of Stockholm syndrome. Cult members go one of two ways when confronted with realities which contradict their beliefs: they either finally recognise the silliness of their beliefs or they become even more ardent believers.
It's better to try to find a way to be objective. You shouldn't let people con you and you certainly shouldn't propagate the lie.
HN should severely downrank any "opinion" posts from anon accounts.
I have a Tesla with FSD and it's incredible. Though it drives way worse than me, it drives pretty darn good, and will save a ton of lives and change the world by freeing up a lot of time for people to do more important things.
Is your post supposed to be read as genuine or sarcastic? It looks to me like a strawman of a Tesla owner stereotype, and I don’t want to accidentally eat the onion here by taking it at face value.
Your post appears to be saying that Tesla’s FSD is way worse than human drivers, not safer than them, but we should still welcome it for the convenience. Again, this reads like a strawman. I would like to think the average Tesla owner would not endorse that statement.
The goal isn’t to save time at the cost of lives, it’s to save both, non?
A lot of people say they’re significantly better than average driver. It’s such a stereotypical thing to say, that it probably has negative truth value on average, since it reflects broad personal confidence more than actual driving skill.
You still sound like you’re trying to present the weakest possible straw man for Tesla critics to attack. Stop this bad faith attempt to make Tesla look bad; you’re dragging down the discourse. Nobody is going to take your bait and feed you the expected counter arguments to the nonsense you expressed up-thread.
> Remedy: Tesla will release an over-the-air (OTA) software update, free of charge. Owner notification letters are expected to be mailed by April 15, 2023. Owners may contact Tesla customer service at 1-877-798-3752. Tesla's number for this recall is SB-23-00-001.
Surely the letter could be displayed on screen like most other tech displays some text of sorts before an update occurs.
Are these letters designed to satisfy the legalese types, or is paper still required to make sure tech companies dont make post update changes to the letter contents?
When GP says "the update could be displayed on-screen," they are referring to the car's user interface, and not a secondary device like a computer or cell phone.
Incidentally, when you point out the issue of missing a theoretical electronic version of the notice because of spotty Internet connectivity, in such a scenario, they wouldn't be able to receive the OTA update either.
So Tesla vehicles dont ping notices to other tesla vehicles passing out update info, in much the same way like the covid smartphone bluetooth vicinity checks worked. Thats surprises me, I think even Windows now lets an updated computer notify and update other windows pc's that are more off grid, its called delivery optimisation, but should work for local hotspots in the wild.
Cars driving past each other rarely are in range for long enough to communicate meaningfully.
Locations where multiple Tesla vehicles are likely to congregate - Superchargers and Tesla retail/service locations - already have good internet connectivity.
I still can't believe FSD Beta was allowed on public roads. I've had it since day 1, and oh my god. It behaves worse than a drunk teenager. All the updates since, have made it barely better. This needs to be pulled entirely and rethought.
The tech might promise to save millions of lives, but if "behaves worse than a drunk teenager" is to be believed, don't you think it's best to wait a bit, if you want to save as many lives as possible?
The "self driving" tech that works (e.g. auto braking) is deployed and has saved lives. (Self driving in quotes because it is not really the full promise of self driving.)
This Tesla AI does not work well enough in many conditions and is clearly sometimes more dangerous than humans. Just watch videos of people using it - it's obvious that is making unforgivable errors.
Do you have evidence for "clearly more dangerous"? Because what numbers exist say the opposite. They are now pushing a million (!) of these vehicles on US roads and to my knowledge the bay bridge accident[1] is the only example of a significant at fault accident involving the technology.
It's very easy to make pronouncements like "one accident is too many", but we all know that's not correct in context. Right? If FSD beta is safer than a human driver it should be encouraged even if not perfect. So... where's the evidence that it's actually less safe? Because no one has it.
[1] Which I continue to insist looks a lot more like "confused driver disengaged and stopped" (a routine accident) than "FSD made a wild lane change" (that's not its behavior). Just wait. It took a year before that "autopilot out of control" accident was shown (last week) to be a false report. This will probably be similar.
- A Tesla near that drive into the median divider near Sunnyvale on Highway 101, because it thought the gore point was an extra lane. Split the car in half and killed the driver. [1]
- A Tesla that autonomously ran a red light and killed two people. [2]
- Multiple accidents where Teslas have driven into parked emergency vehicles / semi trucks.
Is it quantitatively safer than human drivers? I don't have that data to say one way or the other. But it's not correct to say the Bay Bridge is the only at fault accident attributable to autopilot.
Those are autopilot accidents from before FSD beta was even released.
I mean, it's true, that if you broaden the search to "any accident involving Tesla-produced automation technology" you're going to find more signal, but your denominator also shrinks by one (maybe two) orders of magnitude.
These show pretty unequivocally that AP is safer than human drivers, and by a pretty significant margin. So as to:
> Is it quantitatively safer than human drivers? I don't have that data to say one way or the other.
Now you do. So you'll agree with me? I suspect not, and that we'll end up in a long discussion about criticism of that safety report instead [edit: and right on queue...]. That's the way these discussions always go. No one wants to give ground, so they keep citing the same tiny handful of accident data while the cars keep coming off the production line.
It's safe. It is not perfect, but it is safe. It just is. You know it's safe. You do.
This data does not show what it pretends to show. The underlying statistics are heavily skewed in favor of Tesla by various factors:
* Autopilot can only be enabled in good conditions and relies on human drivers in all other, to the point of handing over to a human when it gets confused. Yet, they compare to all miles driven in all conditions.
* Teslas are comparatively new cars that have - as they proudly point out - a variety of active and passive safety features that make the car inherently less prone to accident than the average, which includes old, beat up cars with outdated safety features. Teslas are also likely to be better maintained than the average car in the US by virtue of being an expensive car driven by people with disposable income.
* Teslas are driven by a certain demographic with a certain usage pattern. Yet, they provide no indicator on how that skews the data.
Tesla could likely provide a better picture by comparing their data with cars in the same age and price bracket, used by a similar demographic, in similar conditions. They could also model how much of the alleged safety benefit is due to the usual, active and passive safety features (brake assistance, lane assist, …). They don‘t, and as such, the entire report is useless or worse, misleading.
All very true, by far the biggest factor is the first bullet. Autopilot is enabled exactly when the driver thinks it is very safe and easy driving. If we are comparing miles without an accident it shouldn't be compared to all miles, it's much more like all miles with cruise control activated, data which Tesla doesn't have and probably can't get, but I don't think they'd want to trumpet it either.
It should be blindingly obvious that FSD is not safer than a human right now from using it or watching it be used. It's weird that people are trying to point to some very misleading statistics when you can go to YouTube and see serious problems in almost every video about FSD, even those by owners clearly wanting to paint it as great. Of course getting data-based answers is great but when you see problems every couple miles and the data is saying is saying it's way safer than a human, you are doing something wrong with or using irrelevant data.
> when you see problems every couple miles and the data is saying is saying it's way safer than a human, you are doing something wrong with or using irrelevant data.
That's right, you are doing something wrong. It's a supervised system. There's a driver in the loop, and yes: the combination is safer.
You're engaging with another favorite trope of Tesla argumentation: pretending that a system that isn't actually deployed as fully autonomous actually is. Safety is about deployed systems as they exist in the real world. Your point is that in a alternate universe where Teslas all ran around without drivers, they might be less safe. And... OK! That's probably true. But that's not a response to "Teslas are safe", is it?
They're safe. Just admit that, and then engage in good faith argument about what needs to change before they can deploy new modes like full autonomy. Lots of us would love to have that discussion and provide detailed experience of our own, but we can't because you people constantly want to ban the cars instead of discuss them.
I don't believe one accident is too many. I made my statement based on videos I've seen of people having to disengage their beta FSD in circumstances where a human driver would have no trouble.
Now, maybe the data says otherwise. If that is the case, then great! Let's role out some more FSD beta. But for that data to be valid, you have to account for the fact that Tesla filters bad drivers out of the pool of FSD users. And as I understand it there is not public data about the risk profiles of the people Tesla lets use FSD.
It's called FSD as in Full Self-Driving. If it does one single "accident" where it's unable to see a totaled car with hazard lights on right in its face [1], then it's not really full, is it now. Not to mention the hilarious contradiction in terms: Full, but Beta.
No one would have any major issues if this (self)killing system was called "Extended Driver Assist" and it had implemented the minimal safety features like driver monitoring, safely stopping the car if the driver doesn't pay attention to the road.
The Tesla tech is not ready and is deployed haphazardly. Google and other companies are approaching self-driving in a much safer way while still having tremendous progress.
I was specifically referring to "do the same" as in reducing the number of auto accidents, which the parent comment I replied to mentioned. Is that the source of confusion?
My point was that although public transit can reduce car accidents, it cannot and will not ever completely replace individual transportation, so FSD cars which are safer than human drivers would still be a benefit to road safety.
I did not mean to suggest that this end justifies Tesla's reckless means.
I'm not arguing against public transit, not at all. My point is that public transit cannot and will not ever be a complete alternative to FSD cars, especially outside dense cities.
Even "Auto Pilot" is a poor name, given it's pretty much just the same suite of safety features offered by others (many of which are superior given they still have LIDAR and USS)
> I wonder how far would other solutions to reduce road car deaths go with the same budget?
Far, far, far wider. Reducing road deaths is a pretty well understood, political implementation is the problem.
And the simply fact is, reducing road deaths is cheap. Its very cheap. If you want to make it look good, and fancy, its a bit more expensive.
But the fact is, we know how to do it, and do it cheaply. Fancy next generation car technology is pretty terrible in terms of investment to return.
All you really need is a bunch of paint, and a few concert spectators over various shapes. Pretty much all you have to do is make cars slower, that is by far the most important factor in mixed traffic. There are many ways to achieve this.
If you want to get a bit more fancy and technical, you can make dutch intersections:
Tesla is using the same basic technology for security features, many of those features are deployed to all cars. The can and should do that for even more features over time.
I beta test regularly in Vancouver BC on my Model S LR. My experience is the polar opposite to yours. It's surprising (on the good side) and has improved immensely in the past 12 months. It does not need to be rethought imo. Obviously your opinion is valid, but it's not consensus.
Yeah I use it all the time in seattle here on my model x. It’s amazing. Sometimes it makes a questionable decision but the vast majority of the time it works as expected. I don’t mind being alert to the driving conditions, and the amount it can do without assistance is remarkable
You think that’s better than the questionable decisions a human makes? When a human makes one who corrects it? With autopilot you face the joint distribution of it and me making a mistake rather than either of us.
You can believe whatever you want. The reality is that FSD Beta is unsafe and makes the dumbest decisions. Every update I give it a try to see if anything has improved, and usually within the first 30 seconds, I end up turning it off.
Autopilot on Freeways works ok for the most part though.
I have FSD too, and my opinion is that although it takes unsafe actions almost on a daily basis, most of these errors are minor, easily rectified by me the driver, and most of the time it drives just fine, in fact better than a human driver. I witnessed FSD avoiding me a collision by responding faster than I could have by auto-braking when another driver ran a red light.
FSD doesn't need to be perfect to avoid accidents. In fact, if it was too perfect I'm afraid I could become very inattentive at the wheel.
I have a Tesla and have tried FSD several times (month-to-month plan and loaner cars from Tesla). It's comically bad. Fortunately every time I tried it I was paying very close attention so I stopped it from braking at green lights, trying to turn right from a middle lane, halfway lane changes where it snaps back to the original lane unexpectedly, etc.
The regular, included autopilot works pretty well as a smarter cruise control; as long as you use it on a major highway in the daytime in good conditions it does a good job. And it's very useful in stop-and-go traffic on the highway.
You are meant to maintain control of the vehicle at all times with the beta.
That’s the only way it could be released in its current state.
Looking at it as anything other than a beta where you have to maintain control is misunderstanding what it is. Which you are clearly doing. It is absolutely expected to be worse than a drunk teenager in this stage.
Meanwhile, the only actual L3 capable car is a Mercedes. Plenty of people will argue that AP is actually superior, but I don't think there's any way to say that for sure, given the tight constraints Mercedes has put on their driver assistance in order to lower the liability enough to make L3 not a bankruptcy event. For certain, however, they are more confident in their technology than Tesla is, otherwise Tesla would release an L3 capable car too.
To be fair, though, Tesla has no sensors other than cameras, and I believe the Mercedes has a half dozen or so, several radars and even a lidar.
Flies hit windows and cars. Even much more complex animals cannot manage effectively moving in a complex environment without hitting each other (like flocks of sheep).
There's no data that would support that anything than human level AGI is required to drive cars with how current infrastructure looks like.
> To be fair, though, Tesla has no sensors other than cameras, and I believe the Mercedes has a half dozen or so, several radars and even a lidar.
This doesn't mitigate Tesla's gross ethical violations in letting this thing loose, it makes them worse. Tesla knew that cameras alone would be harder to make work than cameras plus radar plus lidar, and they shouldn't be lauded for attempting FSD with just cameras. It's an arbitrary constraint that they imposed on themselves that is putting people's lives in danger.
Remember that big announcement musk did where he stated, that going forward, Teslas would not use any additional sensors to aid the video---to the point of laughing at every other company that was still using them. Yeah.
That Tesla was allowed to test in production without any legal liability with real human lives mostly because of Musk's personal influence is a disgrace
> That Tesla was allowed to test in production without any legal liability with real human lives mostly because of Musk's personal influence is a disgrace
I am unsure what"allowed" means in that context. They just did it. The lawsuits are coming, surely?
Sometimes it is better to ask for forgiveness than for permission. This was an audacious example. Time will tell if there is anything that will stop them.
This is limited to certain pre mapped roads, under 40 mph and requires a car in front to follow. This in no way compares to Tesla's Autopilot system.
Edit (because HN throttling replies): > Mercedes is being responsible about their rollout
You can always succeed if you define success. If rolling out something that doesn't do much of anything is success, success? Oh well. Less than 20 people have died related to Autopilot in ~5 billion miles of travel. Zero deaths is unattainable and unrealistic, so "responsible" and "safe" is based on whatever narrative is being pushed. 43k people died in US car accidents last year, roughly half a million deaths in the entire time Tesla has offered some sort of driver assistance system called Autopilot.
How many deaths would be be okay with Mercedes' system where is would still be "responsible"? Because it won't be zero. Even assuming 100% penetration of Automatic Emergency Braking systems, it's only predicted to reduce fatalities by 13.2% (China specific study), and injuries by a similar number.
TLDR "Safe" is not zero deaths. It is a level of death we are comfortable with to continue to use vehicles to travel at scale.
Mercedes is being responsible about their rollout, so people like you jump to the conclusion that their tech is worse. For my part, I see Mercedes taking liability for accidents that happen and think that they must have a lot of confidence in their tech to roll it out at all under those terms.
I would not be surprised to see Mercedes beat Tesla to safe fully automated driving because they took it slow and steady instead of trying to rush it out to legacy vehicles that don't have the right sensor arrays.
Edit: It's weird to reply to a comment by editing your own. It feels like you're trying to preempt what your interlocutor is saying, rather than reply.
It is in no way comparable to Tesla's Autopilot system. Mercedes has an actual feature while Tesla recklessly gambles with other peoples' lives to juice their stock price.
If Tesla believes that their feature is safe, then let them take legal liability for it.
The law does not require they take liability, so they shouldn’t. If you don’t like the law, change the law. If you don’t want to use a driver assist feature, don’t. But the data shows Autopilot is robust and regulators allow its continued sale and operation.
>If you don’t want to use a driver assist feature, don’t. But the data shows Autopilot is robust and regulators allow its continued sale and operation.
uhh,, this is a thread where the federal gov't just forced Tesla to patch the feature, and so far we don't even know what the patch does...
It's available in bumper to bumper rush hour traffic on freeways, and they take liability for any accidents that happen while it's on. That's exactly the kind of system and guarantee that could really have a positive impact on someone's commute.
Don't LIDAR/radar sensors (I'm not familiar with exactly what the options are) have benefits that vision doesn't have, like working in poor lighting/visibility? Why would Tesla move away from these sensors?
Because LIDAR is expensive and available in limited quantities. There's no way they could sell the amount of cars they are selling right now if each one came with a LIDAR.
Camera modules are cheap and available in huge quantities.
Lidar has advantages over cameras but it also has some downsides. Sunlight, rain, and snow can interfere with the sensors, as can other nearby lidar devices (though de-noising algorithms are always improving). There are also issues with object detection. Lidar gives you a point cloud, and you need software that can detect and categorize things from that point cloud. Because lidar is new, this problem hasn't had as much R&D put into it as similar problems in computer vision.
Then there's the issue of sensor fusion. Lidar requires sensor fusion because it can't see road lines, traffic signals, or signs. It also can't see lights on vehicles or pedestrians. So you still have to solve most of the computer vision issues and you have to build software that can reliably merge the data from both sensors. What if the sensors disagree? If you err on the side of caution and brake if either sensor detects an object, you get lots of false positives and phantom braking (increasing the chance of rear-end collisions). If you YOLO it, you might hit something. If you improve the software to the point that lidar and cameras never disagree, well then what do you need lidar for?
I think lidar will become more prevalent, and I wouldn't be surprised if Tesla added it to their vehicles in the future. But the primary sensor will always be cameras.
Calling it L3 capable is like saying Tesla is Full Self Driving.
Neither is true, both are marketing.
> but I don't think there's any way to say that for sure
I mean yes there is. If you want to have an object test where you drop a car anywhere in the world and see if it can get somewhere else, then clearly one is more useful then the other.
In basketball they say 'the best ability is availability'. In terms of that AP is in a totally different dimension. AP has been driven for 100s of millions of miles by know, it must be a crazy high number by now. Mercedes L3 system has barley driven at all, its available in very few cars.
The only way you can reasonably compare the Mercedes L3 system is if you limit the comparison to the extremely limited cases where the Mercedes L3 system is available. If you compare them there, I would think they aren't that different.
> otherwise Tesla would release an L3 capable car too
No, because making something L3 in a extremely limited selection of places is simply not something Tesla is interested in doing. Doing so would be a lot of work that they simply don't consider worth doing when they are trying to solve the larger problem.
If Tesla's work is actually going somewhere, then they don't need to change the engineering to get L3 in specific places. They can have a list that gets added to over time.
Actually it would be a lot of work doing that. Even if not just work on the software itself. Something like that would cause a lot of work in literally all over the company.
Also, they don't really have an inattentive to do that, so why should they.
Why? The design is locked in, the manufacturing doesn't change, you don't need the designers to do much... I would expect you need a handful of people, half of which are lawyers.
> Also, they don't really have an inattentive to do that, so why should they.
I'm pretty sure almost all Tesla owners would like the ability to stop looking at the road some of the time. That amount of customer satisfaction is not a motivation for the company?
> Yet even after the company declared bankruptcy in 2017, the Takata recall kept on giving. 65-70 million vehicles with faulty Takata airbags were recalled by the end of 2019, with approx. 42 million still to be recalled.
The word recall is misleading. It’s a software update. My cars both updated last night. I’m guessing it was this patch. If that’s the case then the “recall” is probably largely already done.
> The FSD Beta system may cause crashes by allowing the affected vehicles to: “act unsafe around intersections, such as traveling straight through an intersection while in a turn-only lane, entering a stop sign-controlled intersection without coming to a complete stop, or proceeding into an intersection during a steady yellow traffic signal without due caution,” according to the notice on the website of the National Highway Traffic Safety Administration.
Does anyone have insights on what QA looks like at Tesla for FSD work? Because all of these seem table-stakes before even thinking about releasing the BETA FSD.
This is not a former Tesla engineer. This is competitor who wants to discredit Tesla and sell its own solution.
The guy behind this is known to be untrustworthy, and many of the videos don't actually do what he claims. Notably he refused to release the videos that would prove his claims right.
The reality is that Tesla scores high on all the automated breaking test done by government. The driver however can override this, and that is exactly what is being done in this video.
So they scored high on whatever version of software was on the specific car tested by government at some point in time. Has any government done any testing at all on the version of software actually in use today?
The automated breaking is standard part of the software on every Tesla and is the software that was tested by multiple governments. This gives you certification. If Tesla would change or remove that software, it would be highly illegal. Tesla is actually notable that they are one of the only company that has implemented the highest standard of this safety feature on every car. These features are something Tesla is proud of have talked about quite a bit. They also mention that they have achieved the highest scores these evaluations.
The test done above has been replicated and the car does break in automated driving. And it gives loud warning to the driver even under normal driving operation and does emergency breaking.
Tesla assumes that if the driver hits the accelerator after the warning, the drive wants to accelerate based on the drivers judgment.
This is what this video shows, notably the person that made this video has refused to release prove that the car actually was in Autopilot, refused to provide evidence that the driver didn't hit the accelerator and refused to provide audio from inside the car so it can be verified that there was no warning sound.
In addition to that, the same person also claims things like 'millions of people will die if Autopilot isn't stopped' and that even under absurd assumptions is a legit insane thing to say.
So what is more likely, that Tesla did something incredibly illegal removing and incredibly important safety critical software that is standard in every Tesla OR that a competitor is doing a PR campaign where they deliberately set up a situation to film a video where they can create maximum damage to Tesla and then advocate for their own solution.
Exactly. It's the same reason that no amount of unit tests can replace formal methods for safety-critical software, and we cannot apply formal methods to neural nets [yet].
That's not really true. Most safety critical software is tested without formal verification. They are just really really thorough and rigorous.
Formal verification is obviously better if you can do it. But it's still really really difficult, and plenty of software simply can't be formally verified. Even in hardware where the problem is a lot easier we've only recently got the technology to formally verify a lot of things, and plenty of things are still out of reach.
And even if you do formally verify some software it doesn't guarantee it is free of bugs.
This seems to ignore that if you look inside the box at code you could understand it whereas looking at the activation values is unlikely to illuminate.
Practical neural networks operate in enormous parameter spaces that are impossible to meaningfully test for all possible adversarial inputs and degraded outputs. Your FSD could well recognize stop signs in your battery of tests but not when someone drew a squirrel on it with green sharpie.
Your run of the mill computer program also "operates in enormous parameter spaces that are impossible to meaningfully test for all possible adversarial inputs and degraded outputs".
This is hardly similar as the state of a typical computer program can be meaningfully inspected, allowing both useful insights for adversarial test setups and designing comprehensive formal tests.
Right, if you consider the internal state, it is hardly similar. You talked about black box and QA though. Black box by definition holds the internal state as irrelevant, and QA mostly treats the software it tests as a black box, or in other words the tests are "superficial" as you call it.
Black box testing in typical software is, however, less superficial, because the tester can make inferences and predictions about what inputs will affect the results. When you're testing a drawing program, for example, you may not know how the rectangle tool actually works internally, but you can make some pretty educated guesses about what types of inputs and program state affect its behavior. If the whole thing is controlled by a neural network with a staggering amount of internal state, the connections you can draw are much, much more tenuous, and consequently the confidence in your test methodology is significantly harder to come by.
Something a bit similar is clinical trial and it is accepted without problem.
You make a black box test on several thousands (sometimes only hundreds) patients, and if patients who received the drug perform better the patients who received the placebo, then the drug is usually accepted for commercialization.
Yet one isolated patient may be subject to several comorbidities, her environment could be weird, she could ingest other drugs (or coffee, OTC vitamins or even pomelo) without having declared it. In a recent past women were not part of clinical trials because being pregnant makes them very "non-standard'.
First of all, clinical trials are typically longer and more thorough than you imagine, they span years. The fact that COVID vaccines were fast-tracked gives people wrong idea about it.
Secondly, even after the product hits the market the company is still responsible for tracking any possible adverse effects. They have a hotline where a patient or doctor can report it, and every single employee or contractor (including receptionists, cleaning staff, etc.) is taught to report such events through proper internal channels if they accidentaly learn about them.
> Something a bit similar is clinical trial and it is accepted without problem.
Clinical trials also have strict ethical oversight and are opt-in. If clinical trials were like Teslas, we'd yeet drugs into mailboxes and see what happened.
As a human being and motor vehicle operator of many decades I have done all of the above, multiple times (very infrequently), both on purpose and on accident. I’m looking forward to the days when self-driving vehicles are normal, and human drivers are the exception. Until then, I’m glad companies and regulators are holding the robots to a higher standard than the meat computers.
> Does anyone have insights on what QA looks like at Tesla for FSD work? Because all of these seem table-stakes before even thinking about releasing the BETA FSD.
Tesla is not exactly in love with QA. Especially for FSD.
FSD is mainly 2 things:
1. (By far most important) shareholder value creating promise, that's been solved for 6 years according to their CEO.
2. Software engineering research project
What FSD is not is a safety critical systems (which it should be). They focus on cool ML stuff and getting features, with any disregard for how to design, build and test safety critical systems. Validation and QA is basically non-existent.
Do you have actual knowlage of Tesla internal QA processes, any kind of source at all?
Based on there presentation, they for sure have a whole load of tests, many built directly from real world situation that the car has to handle. They simulate sensor input based on the simulation and check the car does the right thing.
They very likely have some internal test drivers and before the software goes public it goes to the cars of the engineers.
Those are just some of things we know about.
I have no source on their approach to testing safety critical systems, but we do know that they have a lot of software that has based all test by all the major governments. They are one of the few (or only) car maker fully compliant to a number of standards on automated breaking in the US. We have many real world example of videos where other cars would have killed somebody and the Tesla stopped based on image recognition.
So they do clearly have some idea of how to do this stuff.
So when making these claims I would like to know what they are based on. It might very well be true that their processes are insufficient but I would actual know some real data. Part of what a government could do, is forcing car maker to open their QA processes.
Or the government could (should) have its own open test suit that a car needs to be able to handle, but clearly we are not there yet.
2. Much more important one - Elon's Twitter feed. They're doing last minute changes, and once it compiles and passes some automated tests, it's tested internally only over few days before it's released to the customers. Even if they had world class internal testing (they don't), for something having to work in such a diverse environment like self driving system without any geo-fencing, those timelines are all you need to know.
Because they have a lower center of gravity and good crash structure. These are good things. But avoiding the accident in the first place significantly reduces the need to test that crashworthiness.
And how does that change anything about my statements?
Yeah, they have QA. But for the problem they claim they’re solving (robotaxis) and speed of pushing stuff to customers (on the order of days) it vastly, vastly insufficient. And it lacks any safety lifecycle process regards - again, just look at the timelines. Even if you’re super efficient, you cannot possibly claim you can even such a basic things like proper change management (no, commit message isn’t that) or validation.
what do you call them? there's no way possible that they can make changes to the software and have them thoroughly vetted before the OTA push. Tesla does not have enough cars owned by the company driving on public roads to vet these changes. The QA team at best can analyze the data received from the customers. That makes the customers the testers in my book.
the fact that you don't know how tesla vets these changes, very extensively, prior to any physical car receiving the update reinforces that you have no idea what you're talking about
tesla does extensive, meaningful vetting of these updates. i'll let you do the research yourself so that maybe you can quit spreading misinformation
so well vetted that issues keep happening, so well vetted that a OTA "software patch" is raised to the level of automotive recall. if you call that misinformation, then, "boy, i don't know".
I think you are just correct that customers are acting as testers. I think the issue he is having isn't on that point, where you are right.
It is well known that issues happen despite extensive vetting. The presence of vetting does not exclude the absence of issues. For example, lets say there are 1000 defects in existence and your diagnostic finds 99% of them. So it finds 990 defects. So now there are 10 defects remaining that are not found. Next you have another detector that finds all defects. How many defect does it see? 10. It is usually going to see about ten defects. Your expectation given vetting is taking place should that be you will tend to observe about ten defects in this particular situation.
So lets say you are someone who is watching that second detector. You observe that you keep seeing defects. Probably, because of selection bias on results, you observe this with an alarming 100% rate of finding defects - the few times no defects happen it doesn't get shared for much the same reason we don't pay attention to drying paint. Can you therefore claim there is no diagnostic that is detecting and removing defects?
Well, not really, because you aren't observing the defects unconditional on the process that removed them. Basically, when you observe 10 defects, that doesn't mean there weren't 990 other defects that were removed. It just means you observed 10 defects.
So the actual evidence you need to use to decide whether there is a diagnostic taking place is evidence that is more strongly conditional on whether it is taking place. You need something that can observe whether the 990 are being caught.
In this case, we have video evidence of extensive QA processes. So that is much stronger than the evidence that defects show up, because defects show up both when we do have a vetting and we don't have a vetting. And each reasonable test case ought to be decreasing the chance of a particular defect that the test case is testing for.
For a much more thorough treatment on this subject check out:
Tesla definitely needs to root cause the defects that were found and make improvements on the existing vetting process but it is very obvious they do have these processes.
Even AP is is "autosteer (beta)" so I sure do feel like part of their QA team. And it drives like a slightly inebriated human.
I do have high hopes that the work they've done on the FSD stack will make for a significant improvement to basic AP whenever they get merged (assuming it ever happens; it has been talked about for years). That'd be nice.
> speed of pushing stuff to customers (on the order of days)
this is also false and doesn't happen
> you cannot possibly claim you can even such a basic things like proper change management (no, commit message isn’t that) or validation.
you know absolutely nothing about the internal timelines of developments and deployments at tesla and to suggest it's impossible without that knowledge is just dishonest
> > it lacks any safety lifecycle process
> completely demonstrably false
Head of AP, testified under oath, that they don't know what's Operational Design Domain. I'll just leave it at that.
> > speed of pushing stuff to customers (on the order of days)
> this is also false and doesn't happen
Never ever Musk tweeted about .1 fixing some critical issues coming in next few days? I must live in a different timeline.
> > you cannot possibly claim you can even such a basic things like proper change management (no, commit message isn’t that) or validation.
> you know absolutely nothing about the internal timelines of developments and deployments at tesla and to suggest it's impossible without that knowledge is just dishonest
Let's assume I have no internal information. If it looks like a duck, swims like a duck, and quacks like a duck, then it probably is a duck.
> Never ever Musk tweeted about .1 fixing some critical issues coming in next few days? I must live in a different timeline.
Lets say you have a baby which is being born. You tweet, birth in ten days! You can't then say, look: here is a tweet, it proves that baby development in tweeters actually has a several day lifecycle and moreover it proves that baby having mother's don't do proper pre-birth routines, because the tweet isn't the process that created the baby.
It is separate.
> If it looks like a duck, swims like a duck, and quacks like a duck, then it probably is a duck.
Right. So the fact that we have video evidence of internal processes, including QA processes, is much more like looking, swimming, and quacking like a duck much like having video evidence of a mother with a round belly for months before the tweet would be evidence that the babies don't take weeks to develop.
So when Elon has also tweeted that a launch was delayed, because of issues that were discovered - which does happen, as I'm sure you know if you follow his tweets as you imply - then that would be evidence congruent with the video evidence we have of QA processes existing within the company.
I strongly feel people ought to have these discussions while consistently citing actual data sources relevant to the discussion.
For example, did you predict, based on the speculation of Tesla being incompetent with regard to safety, that they have the lowest probability of injury scores of any car manufacturer? Because they do.
Did you predict, based on speculation about Elon Musk's incompetence in predicting that self-driving would happen, that there are millions of self-driving miles each quarter? Because there are.
Did you predict, based on speculation about Tesla incompetence in full self-driving, that the probability of accident per mile is lower rather than higher in cars that have self-driving capabilities? Because they do.
I know this sort of view is very controversial on Hacker News, but I still think it is worth stating, because I think people are actually advocating for policies which kill people because they don't actually know the data disagrees with their assumptions.
I agree that people who don't cite the evidence are ignoring the evidence? Are you trying to say I'm doing that by pointing to relevant datasets which track the number of accidents and the probability of injury? If so, why are there accidents tracked in the datasets such that the rate can be calculated? This kind of contradicts the claim that I'm asking to ignore, but I definitely agree that other people are ignoring the data if that is what you are trying to say.
No, your argument is just ridiculous. The standard isn't and shouldn't be how much they get right. It should be what they get wrong and how they do that. I completely disagree with your point, and phrasing it obtusely just makes you obnoxious from a conversational standpoint.
My position is that we ought to include assertions backed by the evidence. Your views probably do have evidence that supports them. I want to see the evidence you are using, because I think that is important.
I'm not sorry that annoys you, because it shouldn't.
>Oh. So you don't like the data, because it disagrees with you. So you are trying to pretend I'm ignoring data, even though I'm linking to summary statistics which by their nature summarize the statistics rather than ignoring the statistics.
Oh the data is great. I like the data. I'd take the data out to dinner. It's completely besides my point, and you continuing to be obtuse and rephrasing things this way, is not only a strawman, but it's rude.
> Your views probably do have evidence that supports them. I want to see the evidence you are using, because I think that is important.
Not every policy decision is driven by data. Some are driven by reasoning and sensibility, as well as deference to previous practices. So your whole data-driven shtick is just that... a shtick.
You claim that I said that we should ignore evidence, but I didn't. I claimed that we should look at it.
You claim that I said that we should focus on the good, but I didn't. I claimed that we should look at the data.
Now I feel as if you are trying to argue that looking at data is wrong because not all decisions should be made on the basis of the data. This seems inconsistent to me with your previous assertion that my ignoring data was bad, because now you argue against your own previous position.
That said, uh, datasets related to bayesian priors support your assertions about deference in decision making. So you could, if you cared to, support that claim with data. It would contradict your broader point that I should not want to have data, but you could support it with data and I would agree with you, because contrary to your assertion I was making an argument for evidence informed statements. Your inference about whether I think the evidence leans should not be taken as an argument that I believe my positions would always be what was reached by looking at the evidence, because I don't think that is true. I'm obviously going to be wrong often. Everyone is.
Unfortunately, I think you lie too much and shift your goalposts too much. So I'm not going to talk to you anymore.
I never said you shouldn't want to have data. I said that the data isn't the only story, so appeals to data aren't dispositive. Data is clearly the only thing you are capable or willing to talk about. There isn't a point in furthering this conversation if you are just going to repeatedly misrepresent my comments and converse in this incredibly obtuse manner.
I also caught you editing out what was an excessively rude comment. I'm gonna pass on further conversation, thanks.
> I said that the data isn't the only story, so appeals to data aren't dispositive.
No, you didn't. You said please ignore all the times when I'm wrong in favor of when I'm right. This is what you actually said.
You seem to be using language incorrectly. You seem to me to be confusing "said" with "meant" and in this case it seems that what you "said" was very different from what you "meant" so much so that I'm strongly getting the impression that you are lying to me, but if you aren't - then it is because confusion on the difference between said and meant.
Words have meanings. They have meanings independent of your own desires, so whatever you meant to say - it doesn't matter at all - that is not what you said. Please ignore is fairly characterized as a request for ignoring things, because it maps to something like it would be pleasing if there was ignorance. Notice it is you who is claiming of me that it would be pleasing to me if there was ignorance. I'm not making such a claim - you said this - maybe you didn't mean this, but you absolutely said it.
I'm not unfairly characterizing your words: this is the actual implication of the words you used, because it is the implication of the meaning of the words - maybe it isn't the meaning you desired, not what you meant, but it is the meaning of the words.
To kind of highlight how extreme what you claim you said is from what you actually said is, notice that you imply belief about me when talking about me requesting ignorance. Yet now, in your claim about what you actually said, you imply a position you hold: that data isn't dispositive. You don't even have the same reference in what you said versus what you now claim to have said. You did not say what you claimed to say. If you meant that, you should have said that, but these things are very very far diverged.
If you want to state, of your own belief, that data isn't dispositive, you should state that. Instead you consistently referred to a false reference to my own beliefs. Notice, even when you tried to correct my interpretation, you did not switch the reference class to your own belief, you said things like "your argument is ridiculous" which is still talking about me - not your own position that data is not dispositive. So the reference class which is not targeting me, but the general properties you now claim to have said, is not truly there.
> I also caught you editing out what was an excessively rude comment. I'm gonna pass on further conversation, thanks.
As you can imagine, with your confusion about meant versus said, I've been finding you to be lying about both your own views and mine. So if I seem a bit rude, it is because I'm kind of assuming you are smart enough to already realize all these things. I don't mean to assume you are so hostile as to know all this, but then pretend not to, but it is just one of the valid explanations for your behavior. I tried to edit my comment to remove my frustration and I'm sorry you had to see it like that.
> incredibly obtuse manner.
Which leads to this. You stated that you think I'm being obtuse, but my first assumption was that you weren't just directly lying about what I was saying. Your claim, interpreted in the way you said it, not the way you meant it, is a lie about my belief. So I tried to make your words have a meaning that would make them true, not false. I tried to be charitable, but was confused, because it really seemed like you were lying about my beliefs given your statement.
This wasn't me being obtuse. This was me trying to be charitable, but being very confused, because interpreted according to what you actually said - not what you meant - you were strictly speaking stating falsehood about my beliefs.
Unaudited (internal Tesla data), cherry-picked (comparing with average cars in USA, which are 12 years old beaters, to their very young fleet of expensive cars) data, that doesn't correct for any bias (highway driving vs non-highway driving being one of the many issues) is not exactly the magic bullet you think it is.
Also, none of that is self driving. This data talks about AP, not FSD. FSD is also not self driving by any means (it's level 2 driver assist), but that's a detail at this point.
I didn't say it was a magic bullet. So you are hallucinating thoughts about me, not responding to what I said. Being critical of the data like you are being is good thinking in my opinion. I just don't like how often people don't have beliefs that are anywhere close to the data.
For example, elsewhere in this comment thread, someone threw out a random statistic of 400:1 as part of their argument, but this seems to me to be something like six orders of magnitude diverged from a data informed estimate.
To try and contextualize how big an error that is - it is like thinking that a house in the Bay Area has the same cost as a soft drink.
I think if we have to cite our data we are less likely to do that sort of error and more likely to catch it when it is done.
I definitely don't think FSD is magically safe. So if you think that is what I'm trying to say, please update your beliefs according to my correction that I do not believe this. I think anyone driving in FSD should remain vigilant, because it can make worse decisions than a human would.
Interesting graph, I like that it's broken out into quarters. But,
1) those are statistics for the old version, the new version might be completely different. I've had enough one-line fixes break entire features I was not aware of that my view is that any change invalidates all the tests. (Including the tests that Tesla should have but doesn't) Now probably a given update does not cause changes outside its local area, but I can't rely on that until it's been tested.
2) the self-driving is presumably preferentially enabled for highway driving, which I assume has fewer accidents per mile than city driving, so comparing FSD miles to all miles is probably not statistically valid.
I agree with you. I would really like to see datasets that reflect how things actually are. I think it would be really dangerous to jump to FSD being safe on the basis of the data I shared. However I would hope that whatever opinions people shared were congruent with the observed data. I don't feel like the prediction that Elon Musk and Tesla not caring about safety is congruent with the observed data, which shows the autopilot has improved safety, best explains the observations of improved safety.
Just for context - I've been in a self-driving vehicle. Anecdotally, someone slammed on the breaks. The car stopped for me, but I was shocked: for hours before this the traffic hadn't changed, it was a cross country trip. I think I would have probably gotten in an accident there. Also anecdotally, there are times where I felt the car was not driving properly. So I took over. I think it could have gotten into an accident. Basically, for me, the best explanation I have for the data I've seen right now is that human + self-driving is currently better than human and currently better than self-driving. The interesting thing about this explanation is how well it tracks with other times where we've technology like this before. In chess playing for example, there was a period before complete AI supremacy (which is what we have now) where human + AI was better than AI.
I like the idea of being safe, so if the evidence goes the other way, advocating for only humans or only AI doing the driving, I want to follow that evidence. Right now I think it shows the mixed strategy is best and that is kind of nice to me because it implies that the policy that best collects data to reduce future accidents through learning happens to be the policy that is currently being used. I like that.
A system that protects 400 people but kills 1 is not a system that I want on public roads because I don't want to be in the 1 - Elon and the children of Elon are basically making the assumption that everyone is okay with this.
The probability of an accident for any driver assistance system will ALWAYS be lower than a human driver - but that doesn't mean the system is safe for use with the general public!
People like me are not advocating for "killing people" because we aren't looking at data - it's that no company has the right to make these tradeoffs without the permission and consent of the public.
Also if this was about safety and not just a bunch of dudes who think they are cool because their Tesla can kinda drive itself, why does "FSD" cost $16,000?
> A system that protects 400 people but kills 1 is not a system that I want on public roads because I don't want to be in the 1 - Elon and the children of Elon are basically making the assumption that everyone is okay with this.
>
> The probability of an accident for any driver assistance system will ALWAYS be lower than a human driver - but that doesn't mean the system is safe for use with the general public!
Totally we should be wary of a system that protects 400 and kills 1. Thank you for providing the numbers. It helps me show my point more clearly.
If you are driving on a road you encounter cars. Each car is a potential accident risk. You probably encounter a few hundred cars after ten or so miles. Not every car crash kills, but lets just assume they all do to make this simpler. For the stat you propose, you are talking about feeling uncomfortable with an accident per mile of something around the ballpark of ten miles.
Now lets look at the data. The data suggests the actual miles per accident is closer to 6,000,000 miles per accident. This is six orders of magnitude diverged from the number of miles per accident that you imply would make you feel uncomfortable.
Lets try shifting that around to a context people are more familiar with: a one dollar purchase would be a soft drink and a six million dollar purchase would be something like buying a house in the bay area. This is a pretty big difference I think. I feel very differently about buying a soft drink versus buying a house in the Bay Area. If someone told me they felt that buying a house was cheap, then gave a proposed price for the house that was more comparable to the cost of buying a soft drink, I might suspect they should check the dataset to get a better estimate of the housing prices, because it might give them a more reasonable estimate.
So I very strongly feel we should cite the numbers we use. For example, I feel like you should really try and back up the use of the 400 to 1 number so I understand why you feel that is a reasonable number, because I do not feel that it is a reasonable number.
> Also if this was about safety and not just a bunch of dudes who think they are cool because their Tesla can kinda drive itself, why does "FSD" cost $16,000?
Uh, we are a on venture capitalist adjacent forum. You obviously know. But... well, the price of FSD is tuned to ensure the company is profitable despite the expense of creating it as is common in capitalist economies with healthy companies seeking to make a profit in exchange for providing value. It is actually pretty common for high effort value creation, like creation of a self-driving car or the performance of surgery, for the prices to be higher.
> Is Autopilot still limited to divided, limited access highways
No. Was it ever? All you need is a piece of road that has something which appears to be lane lines. The road to my house is usable despite having no actual paint striping because it happens to have a crack that runs fairly straight up one side and was filled with tar. So the camera thinks it's a lane line. Ta-da!
The thing is we often have discussions about this stuff and I'm trying to advocate for citing datasets to more tightly correlate our words with the evidence that our words correspond to. I'm not trying to say this version shouldn't have been recalled for example, but that I think we should be close to evidence.
In the case of auto-pilot, it was the case that people made the same arguments that are now being made against FSD. I think that makes it somewhat relevant to the discussion, because people previously also made the same claims about safety, but now that we have the data, we can see those claims were wrong. I believe these sort of generalizations, though inaccurate, can help us to make more informed decisions, but I'm not really confident in any beliefs that are made at this greater decision from direct data.
So I think anyone who can provide datasets which correspond with FSD performance rather than autopilot performance ought to do so. That would be really great data to reflect on.
The thing I'm worried about is that no data at all is backing the conjectures - which, given that I sometimes see estimates that I calculate to be many orders of magnitude away from data informed estimates - seems to be the case on Hacker News at least some of the time.
Andrej Karpathy was the AI lead for most of the project and he has talked about the general system design.
They have a set of regression tests they run on new code updates either by feeding in real world data and ensuring the code outputs the expected result, or running the code in simulation.
It does seem worrying that they would miss things like this.
Here’s a talk from Karpathy explaining the system in 2021:
It's not even a bit surprising they'd miss things like this, IMHO. They do tests with a few (maybe even a lot of) intersections, but there are thousands upon thousands of intersections out there, including some where bushes are obscuring a stop sign, or the sign is at a funny angle, or sunlight is reflecting off the traffic lights, or heavy rain obscuring them, or plain old ambiguous signage...there's _bound_ to be mistakes. Human drivers make similar mistakes all the time.
I used to think that fact was going to delay self-driving cars by a decade or more, because of the potential bad press involved in AI-caused accidents, but then along comes Tesla and enables the damn thing as a beta. I mean...good for them, but I've always wondered if it was going to last.
I've been using it pretty consistently for a few months now (albeit with my foot near the brake at all times). I haven't experienced any of the above. Worst thing I've seen is the car slamming on the breaks on the freeway for...some reason? There was a pile-up in a tunnel caused by exactly that a month or so ago, so I've been careful not to use FSD when I'm being tailgated, or in dense traffic.
The thing is you already know everything you need to know about all 4 billion floats. Collecting data on every intersection in the US is quite difficult.
Tesla does however collect data on edge cases and then train their system to respond correctly. They can for example trail a collection network to identify things that might be obscured stop signs, then have the fleet collect a whole bunch of examples, hand label those samples, and roll this new data in to the training system. This is explicitly how they handle edge cases.
They can also create a new feature or network and roll it out in “shadow mode” where it is running but has no influence on the car, and then they can observe how these systems are behaving in the real world.
The real issue I guess is when they release a new feature without trialing it in shadow mode, or if they have gaps in their testing and validation system.
It also does not know what one way street, do not enter, road closed, and speed limit signs are. Really, the only signs it appears to know about are stop signs.
As for their QA process, in 2018 they had a braking distance problem on the Model 3. They learned of it, implemented a change that alters the safety critical operation of the brakes, then pushed it to production to all Model 3s without doing any rollout testing in less than a week [1]. So, their QA process is probably: compiles, run a few times on the nearby streets (I am pretty sure they do not own a test track as I have never seen a picture of tricked out Teslas doing testing runs at any of their facilities), ship it.
I have a winding road near me with a speed limit of 35 mph, but 15 mph on certain curves as indicated by a speed limit sign. It ignores those speed limit signs and will attempt to make the turns at 35 mph resulting in it wildly swerving into the other lane and around a blind turn with maybe 30 feet of visibility. It has also attempted to do it so poorly that it would have driven across the lane and then over the cliff without immediate intervention.
Unsupported claims by a manufacturer that compulsively lies about the capabilities of their products except when directly called on it are the opposite of compelling evidence.
I'm talking about standard speed limit signs. You're talking about the signs that warn about sharp turns and advise maximum speeds. Yes it would be good if the software understood those signs, but that's a different issue.
Teslas definitely read speed limit signs. I've had mine correctly detect and follow speed limits in areas without connectivity or map data. It also follows speed limits on private drives (if there is a sign) and obeys temporary speed limit signs that have been put up in construction zones.
So they read some, but not all speed limit signs, and especially not the really important ones that inform you that you will be going dangerously fast if you do not read and follow them. That is criminally unacceptable.
Waymo. Cruise. And I do not see how the presence or absence of features in other products has any bearing on the lacking safety characteristics of FSD.
Frankly, attempting to deflect by arguing that it is okay to release a defective safety critical product to unsuspecting consumers just because nobody else is willing to offer a similar product because they have some moral integrity is a stance that makes the executives at Ford presiding over the Pinto look like angels in comparison. All the Ford executives did was cover it up to avoid having to pay to fix it. At least they did not intentionally release a known defective and dangerous product just to recognize some revenue.
Neither of those companies have software that reads those signs and use them to choose cornering speeds. They both use high resolution 3D maps from previous lidar scans of the road.
I think this is a legitimate question, and a reflection of how the NHTSA needs to adjust their wording for modern car architectures.
It's technically a recall, but it's fixed with an OTA update. But the fact that any "defect" that can be fixed with an OTA update is called a "recall" is confusing to consumers and contributes to media sensationalism.
There absolutely needs to be a process overseen by regulators for car manufacturers to address software-based defects, but the name would benefit from being changed to reflect that it can be done without physically recalling the vehicle to the dealer.
The problem is that almost no other car company is yet seriously fixing things with OTA. And 99% of cars on the road don't have OTA. So fixing the naming will likely happen but it will take a while.
Its crazy that Tesla has been doing OTA for 10+ years and today many cars are released that are not capable of being upgraded.
And even those few cars that do support OTA only support it for a very limited amount of system. Often they still need to go to the shops because lots of software lives on chips and sub-components that can't be upgraded.
OTA updates carry a perverse set of incentives. Look at the gaming industry. They went from putting out rock solid games because of necessity (the reality of publishing physical cartridges and CD-ROMs without network updates) to the absolute dogshit of No Man's Sky and Cyberpunk 2077. Gamers have effectively become an extension of QA to the point that some game devs simply stop doing QA at all. Which we are seeing clearly with the "beta" version of self-driving software.
To make matters worse, firmware devs are on the lower totem pole of the developer hierarchy. They live more on the cost center side than the profit center side (think airbag control vs. the guy that did the whoopee cushion sounds). The quality of firmware is already incredibly poor across the range of consumer devices. OTA incentivizes corporations to release software earlier than they currently do knowing that they can always fix it later if necessary.
It's technically a recall, but it's fixed with an OTA update. But the fact that any "defect" that can be fixed with an OTA update is called a "recall" is confusing to consumers and contributes to media sensationalism.
Would it be sensationalism if the same recall happened to cars not capable of the OTA update?
Because I think there should be media "sensationalism" about these types of issues, regardless of whether they can be fixed with physical or OTA repairs.
I get why Elon Musk objects to the use of the word recall, but I don't understand why anyone else is going along with that.
> But the fact that any "defect" that can be fixed with an OTA update is called a "recall" is confusing to consumers
What is confusing about this to you?
> contributes to media sensationalism
Why do you think this recall is more sensationalistic than many of the other recalls issued recently? Automotive recalls often address serious issues with cars. "Fix this or you could die" is a common enough theme when, you know, if you don't fix it you could die.
> but the name would benefit from being changed to reflect that it can be done without physically recalling the vehicle to the dealer.
Why? Information related to how, when, and where the recall can be addressed is contained in the text of the recall notice, same as it ever was.
That's 4-5B USD worth of fraud. This thing could actually be quite dangerous for TSLA. One class-action lawsuit around this and they're bankrupt.
Edit: Dang, you're all right, they could eat this and still be alive.
My whole sentiment comes from FSD being their big shot (and everything that comes with that, like the robotaxis and whatnot). Without FSD, they're "just another car company" and the market is already thriving with good alternatives (Audi's EVs are jaw dropping, at least for me). Excited to see what they announce on March 1st, though.
There's many people that still have the perception that Tesla is a small fish waiting to get gobbled up by legacy automakers. But every year that becomes further from the truth. A company with $80 billion revenue and 100,000+ employees is not dying from a single lawsuit.
I thought emperically, statistically, the number of crashes from FSD was drastically lower than human drivers. Was this number hogwash or is this just lawboys saying the number must be 0 before it's allowed on the road?
If so, that's pretty crazy and people will die because of this decision.
The usual comparison is vs. all cars on the road, but Teslas are comparatively new cars with a lot of recent and expensive safety features not available in older cars. They’re also likely to be maintained better. On top, they’re also a driven by a different demographic which skews accident statistics.
Teslas autopilot can only be enabled in comparatively safe and simple circumstances, yet the comparison is made against all driver in all situations. When autopilot detects a situation it can’t handle, it turns off and hands over to the human who gets a few seconds warning. Human drivers can’t just punt the issue and then crash.
Tesla FSD may be safer than Human drivers for the limited set of environments where you can use it, but last time I checked, the numbers that Tesla published are useless to demonstrate that.
> Teslas autopilot can only be enabled in comparatively safe and simple circumstances
This is incorrect. FSD can be enabled everywhere from dirt road to parking lots, even highways and dense urban environments like NYC. There is no geofence on FSD, you can turn it on anywhere the car can see drivable space.
Does it enable in heavy rain or snow, on ice, fog? Does it work in storm conditions with leaves and objects blown over the street? Does it work in confusing traffic situations, invalid signage, …?
You can’t compare FSD crashes / mile with human data for a couple of reasons.
The primary one is:
1) FSD turns itself off whenever things get too difficult or complex for it. Human drivers don’t get to do this. Recent crowdsourced data suggests a disengagment every 6 miles of driving: https://twitter.com/TaylorOgan/status/1602774341602639872
If you eliminate all the difficult bits of driving I bet you could eliminate a lot of human driver crashes from your statistics too!
A secondary issue, but relevant if you care about honesty in statistical comparisons is
2) The average vehicle & driver in the non-Tesla population is completely different. Unless you correct for these differences somehow, any comparison you might make is suspect.
I wonder what severity / frequency of incidents or regulator awareness required them to actually come out and say that they're issuing a recall, rather than just quietly putting it into an upcoming release like they probably would otherwise do?
> A new car built by my company leaves somewhere traveling at 60 mph. The rear differential locks up. The car crashes and burns with everyone trapped inside. Now, should we initiate a recall? Take the number of vehicles in the field, A, multiply by the probable rate of failure, B, multiply by the average out-of-court settlement, C. A times B times C equals X. If X is less than the cost of a recall, we don't do one.
As far as I know pretty much all update to critical system will always be told to the government and cause a 'recall'. But I'm not 100% on what the actual regulatory requirement are.
”Six days after the radiation-wrecked colonists emerged from the battered Starship, the Emperor of Mars condemned them all to death by suffocation for ‘insufficiently engaging with his tweets’. Historians debate the meaning of this expression found carved in a rock outside the base.”
The settlement communication log will be just a series of "I need people to be even more hardcore" declarations coupled with lists of people thrown out of airlocks.
I've read two separate sci-fi novels recently which make some passing reference to the remains of a failed Musk settlement on Mars. Definite shades of this.
Isn’t Antarctica just Mars-lite? Do you think a mere treaty is hampering settlement there? That’s where my mind goes when people talk about colonizing another planet.
Flying to another planet on a rocket is obviously very different, the goal is way more interesting than the challenge of living in a hostile environment. Even if it's ultimately a useless exercise for humanity.
These people are incredibly stupid. They somehow think colonizing mars is easier than regulating the environment here on earth, somehow think elon as dictator is better than any earth government, are interested in putting undertested and sketchy implants into their brain even without a proposed use case, and seem to think that the only reason we don't regularly visit mars is because we don't have a big enough rocket, as if easy access to tons of powdered rust is economically useful.
I'm not sure that the people who want one of those will want them if:
* they are not the first (or in the first ~50 or so) to get there
* they have a lot of money and a lot of income on Earth
This is kind of a glory-seeking thing for people who haven't made a name for themselves otherwise. Rich people would want to have a return flight. That is not a recipe for making a ton of money.
think of it this way: you go there, work for a company, in a spare time, you find gold, raise capital to start mining it, got rich, build a huge house with swimming pools, and all the rich attributes, but on Mars.
Mars will be colonized with great mixture of science and initiative people. Your kids will grew up in this environment, they'll receive best education and endless room for possibilities.
Sure buddy. You will be able to prospect for and mine gold in your spare time after your one-way trip to Mars, including building all of the equipment you need, or having it sent over by rocket. And you'll be able to own land there too, because homestead laws and stuff will definitely carry over.
If you are already rich on Earth, you don't need to take risks that insane to try. In other words, nobody who can afford a ticket to Mars will pay for one.
Surely elon will allow lots of other people to get rich, instead of just capturing any possible value for himself, since he would literally be a god emperor.
There are no rules on mars, except the ones that the people who could simply push you out the airlock make. Look how elon treats his workers and you will understand how a mars colony with his backing will look, except even worse.
I get that gloating when your enemies fail is one of life's big pleasures, but would your opinion on the guy change if we lived an alternate universe where Tesla had gotten FSD working?
Okay, but idlewords' comment is one of those things that sounds like it is a clever point when it is actually not. "Ha, the big talker who breaks things is talking big and breaking things again!" All right, so if today's article had been, Tesla gets FSD working, would he instead have been like "wow, this guy ships things that work, maybe I was wrong about him and Mars?"
No, it's just gloating, right? We get it, he hates Elon, and when you hate someone you'll throw whatever dart you've got on hand at them.
I mean, not to get personal or take this on a tangent, but, I am a longtime reader of your tweets and blog, and you seem to have an actual problem of values with him. You clearly have a direction you'd like society to move in and I feel like I've seen you express dismay that this energetic, instrumentally effective guy is out there making the world into something else.
>You clearly have a direction you'd like society to move in and I feel like I've seen you express dismay that this energetic, instrumentally effective guy is out there
Hahaha talk about a loaded question. You see enthusiasm I see manic grift.
Right, but to be eligible for the return trip you'd first need to pay the loan you took to go there by working as an indentured servant and buying your oxygen and food at company-set prices, right?
A scene in Heinlein's The Cat Who Walks Through Walls has the protagonist sternly lecturing a ruffian on the importance of paying for air in a domed city on the moon. I still can't tell if Heinlein was serious; a later scene has the same character demanding a transplanted foot be cut off because he felt it indebted him.
Heinlein might be the kind of person to genuinely believe that. He also was the kind of person who would write a book as an excuse to write a manifesto. Starship Troopers was just him bitching about "kids these days"
The CONCEPT has been known for decades. Who has put a fuel processing facility on mars? I think there was some minor chemistry experiment on one of the rovers to go in that direction.
The raw energy required to make that fuel, using ANY conceivable process, is extreme though. What energy solution has been proposed? How much solar acreage do you need? How will you keep the dust off the panels? Will there be any option other than sending panels from earth?
All the needed prereqs have zero practical knowledge with them. Nobody has even had the chance to work out kinks yet. It doesn't matter how much elon wants to do something, new shit takes a lot of time, money, and effort to shakedown.
There are plenty of other places you could go travel to where you get to see some amazing wonders right before your guaranteed death. Right here on earth too. Why no go for some of those? You can likely do it for cheaper too.
There are plenty of people who would give their lives to advance civilization for the sake of future generations. The brief pleasure of seeing some amazing wonders is, in comparison, irrelevant.
Not OP but IMO, nothing on earth would compare to another planet.
I'm not into space exploration but sometimes I look up at the moon and it blows my mind that we actually went there. Almost feels impossible even. Mars is that x100.
I personally wouldn't want to die on Mars but I understand the appeal.
> Why do you hate the idea of going to Mars so much?
Because there is so much to do on earth? Literally everyone is here. What other answer does anyone else need? All this Mars fandom sounds to me like clinically depressed sci-fi.
I don't mind solitude but even then, going on a an expedition together that you know you won't come back from must be an amazing bonding experience. Can't imagine a more fulfilling destiny than go to another planet.
Bullshit. Colonization _might_ be a side effect. Bilking various governments out of subsidies and fat contracts is the goal. Just like every other single Musk venture, it is based entirely on feeding at the trough of federal largesse.
He is self-funding Mars via making businesses like SpaceX which sell products like any other company. If you think there shouldn't be tax incentives for things like electrification, co2 reduction, or whatever other subsidy the government offers then your problem is with your governments management, not the businesses using government programs.
Anybody else reminded of that story where the guy gets a sketchy neural implant, then goes to a dystopian mars settlement, and then rips the self-driving feature out of a car with his bare hands?
Having never seen it, I have no context for which details were left out or bent to fit the narrative, but this still feels amazing. Well-earned sunglasses
Remember, if you consider other human beings as simply objects for you to do what you like with, you won't feel as bad when you inevitably get them killed
Every billionaire with a rocket has made their way to space, except for one. And that is even though SpaceX has transported astronauts a few times now. I wonder why?
There really needs to be a distinction between an actual recall (my Model S had one to change the latch on the frunk) and this type of "recall" that is nothing more than an OTA software update.
Why? It involves a safety system! That needs to be tracked publicly and updated! Nothing more than an OTA - does not mean much when everything is fly by wire and a bug could mean your car does not stop accelerating or something.
Recall implies something being returned. This should be called a mandatory software patch, or something like that.
The problem is Tesla owners repeatedly see "recall" to mean "software update", so this might lead to a lot of confusion if a physical recall is actually required in the future.
Laypeople have an incorrect perception of what a recall actually means, especially when it comes to vehicles. The most important effect that comes along with an official vehicle recall is that the manufacturer has to fix the issue for you for free, or otherwise compensate you in some way for reduced functionality you may have paid for.
> It’s like crying wolf, eventually you start ignoring it
The problem here lies in having a manufacturer shipping an unfinished product then relying on an endless stream of recalls to finish developing your vehicle.
These are supposed to be exceptional events. If they've become so frequent you're ignoring them, don't shoot the messenger.
Recalls happen in other product spaces all the time, and they often have "fixes" that say "stop using our product and throw it away". That's still a recall. The word "recall" in relation to this regulation is simply a term for "something is broken in a way that a merchant should not have sold"
This is most likely a legislative issue with NHTSA, I don't think they have a mechanism by which they can enforce a software update since the concept didn't exist when recalls were first implemented.
Electing congresspeople who actually stay on top of the expert consensus in various regulated fields is the only way the frameworks themselves can be improved.
Inaccuracy is inaccuracy no matter in which direction. If an outdated law made it easier for Tesla PR to spin something in a positive light, would you consider that an issue?
Why not? At the end of the day its a binary check box on the paper that you had the fix. Whether that fix was a software update or a new piece of hardware should be irrelevant.
Surely a hardware recall would specifically tell you to go to a dealer, right? I’m not sure it’s really confusing. The specific thing here is that these are mandatory things that are tracked by vehicle.
My Honda had 2 recalls on it recently: one was a software update and the other was over the fact that a few of the cables on the audio system were slightly too short. This sounds like the same thing other than the fact that I had to pop over to the dealership to do them. Even with the cable replacement, in and out in an hour with a nice lounge to sit in.
I wasn't told about either of these recalls, though. I went in because the audio was clicking, and they told me they had 2 recalls out on my car, including for the audio issue, and that it was a quick fix.
I just looked it up. The relevant definition of recall is to request returning a product.
If it's possible to buy a software license online and then return it online after deciding you no longer want it (i.e., a non-physical return), then it stands to reason that Tesla can request that you return the defective software OTA and receive replacement software OTA, and that would be a recall. The fact that you are forced into returning the defective software by virtue of not having the opportunity to block the return request is a fairly minor detail.
While we're talking about semantics, the auto industry seems to have a thing called "Technical service bulletin" which is a piece of actionable knowledge. If it applies to you, you act on it. You would probably have to go to the shop to get the TSB considered and applied. I don't think it has any regulatory weight, except as an input to deciding if a recall is a good idea.
FSD has been a cartoonish display for the better part of a year now, it wasn't until last month that the NHTSA actually pushed on them to do a recall, and from there they "disagreed with the NHTSA but submitted a recall"... which is code for "submitted the recall NHTSA forced on them to submit"
Elon knows better, but he knows he can weaponize people's lack of familiarity with this space and inspire outrage at the "big bad overreaching government"
Are these delivered directly to the device and installed overnight automatically (like an iPhone iOS update) or do they have to be hooked up to a computer to install the software update?
Recall is the word we use when a defect was found and must be repaired or retrofit. The actual process of repair could involve customer action or not.
Teslas are basically digitally tethered to the dealer, so they can be "recalled" anytime (without your approval, fwiw), but it doesn't make the word not apply.
If Tesla replaced the actual computer that runs the software that they're recalling, would you consider that a recall? What if there was no actual physical fault with the computer, but it just had firmware flashed to a chip that couldn't be reflashed?
I'm looking at a piece of mail right now that's an official recall notice for a different make/model of car I own. The issue? Improperly adjusted steering components. The company is offering to make the correct adjustments for free. Nothing is being replaced.
Whether the recall is to replace a poorly designed physical component, or to make an adjustment, or to apply a software update doesn't make a difference to regulators.
A recall is a legal process that's designed to encourage manufacturers to fix safety issues while also limiting their liability. Companies avoid recalls if they can because it's costly, time consuming, and isn't good PR. But it's worth it if the issue is bad enough that it risks a class action lawsuit, or individual lawsuits, and most desirable when someone like the US government is demanding a recall or risk legal consequences.
When a company issues a recall, they make their best effort to notify consumers of the issue, provide with clear descriptions of how consumers can have the issue fixed, and make it clear that it will be paid for by the manufacturer. In return, the manufacturer is granted legal protections that drop their risk of being sued to nearly zero.
> If Tesla replaced the actual computer that runs the software that they're recalling, would you consider that a recall?
Yeah, it requires physically taking the car to a mechanic or dealer who does this. Very different from using the software update button on the car touchscreen.
> Improperly adjusted steering components. The company is offering to make the correct adjustments for free. Nothing is being replaced.
This is clearly a recall, because it requires taking the vehicle to a mechanic or dealer.
How exactly are the steering adjustments done? I figure a handful of car owners might have the skills and tools to work on the steering column themselves, but that's not at all the average car owner.
Can the average car owner adhere a sticker to the underside of their hood? Or read a paragraph that corrects a mistake in the user manual? Or read a few sentences that say "it is possible that in extremely cold environments that the emergency brake release lever can require more effort to operate. This does not indicate a faulty emergency brake. Applying more force than usual will not harm the emergency brake system", followed up with diagrams describing the issue, along with toll free phone numbers offering assistance, as well as phone numbers for the NHTSA, the authority coordinating the recall. No part of the official recall notice instructs consumers to replace or even repair anything.
Again, the nature of the issue a recall addresses is wholly independent of how that issue is remedied. Why? Because a recall is a legal process that, by design, is meant to accomplish one thing: motivating a company to correct an issue that the governing authority considered important enough to correct.
If you choose to believe otherwise, I doubt it matters in the grand scheme of things.
Depends on the device. Some require connection to a host system, some can be done over the air. 100% depends on the security profile of the device in question and what the FDA allows.
Recall should just mean affected vehicles must have the fix. What component of the vehicle is affected or how it is supposed to be fixed should be irrelevant, something is defective and the vehicles should probably not be used until that's sorted.
We should probably go beyond the verbiage of recall but right now since it is removing a feature I think that recall is appropriate. A better verbiage might be safety reversion .
The use of the word "recall" isn't because someone just felt like using it. It's an official legal process, followed to limit the manufacturer's liability.
Whether it's a "good" word or "bad" word is irrelevant. It describes a very rigid and official legal process.
Exactly. I had a "recall" at one point where the manufacturer had a typo on a label in the engine compartment. The fix entailed receiving a new sticker in the mail and applying it over the old one. To this day, I can look up my vehicle on the NTSB site and see that that sticker was delivered to me.
If I had chosen not to actually apply it, the dealer would have been expected to do so the next time my car was in for service.
I don't think this is removing a feature, the recall notices says:
The remedy OTA software update will improve how FSD Beta negotiates
certain driving maneuvers during the conditions described above, whereas
a software release without the remedy does not contain the
improvements.
"Recalls" almost never involve being returned. There's a recall out on my car's water pump (Thanks a lot VW) and no part of it involves sending my car back, or even interacting with VW or a dealer. It's just something I'm supposed to keep in mind over its life and various maintenance in the shop.
Other cars get "recalls" all the time that amount to updating the software in the ECU or TCU. Tesla is simply being treated like everyone else.
Hell, even in food, a "recall" usually means "throw it away" not return it.
It seems to me that calling it a "recall" emphasises the severity of the problem, which might make it easier to argue that a lot is being done for customer safety to the interested authorities. But I don't think Tesla wants this to be seen more than an OTA update from the perspective of their customers (at least those who ignore Tesla news).
I had a VW car that was "recalled" shortly following the emissions scandal. The dealership asked me to come in for a free software update related to emissions. So you can say it's a "recall" to the lawmakers but call it a "free software update" to the user.
'full self driving' is an even more incorrect term, then, if you want to be pedantic. if the car mfg takes zero liability/accountability, then it is zero self-driving.
you can, in fact, 'recall' software. this is semantically accurate description of what is happening.
It's a full self driving 'beta' though, it's literally in the name that it isn't final and doesn't have the bugs worked out. You also have to opt in to it.
Wait is your argument adjectives can only be applied to Tesla marketing terms, and not to other English words?
What do you think “FSD recall” or “software recall” means
Tesla released software that can kill people. It must do a FSD recall
This is not the same as Apple doing an OS update, not mandated by a regulatory body because it could kill someone
I was hit by a driver not paying attention. I had to get two surgeries, was not allowed to stand for 9 months, spent another year just rehabbing my nervous system and learning to walk. I still can’t jump or run
Tesla released software it’s marketing as “full self driving” that is not even Level 2 Alpha! Full Self Driving implies a level 5 system.
Tesla’s software has disclaimers like “may not behave as expected in sunlight”. Even its AutoPilot has these disclaimers.
It constantly claims FSD makes drivers safer. Yet it’s non transparent with its data, the data & comparisons it does release is completely misleading to the point of fraud, comparing apples to oranges. The cars it’s released on public roads to untrained drivers runs through intersections (it’s all over YouTube) and fails all kinds of basic driving tests. Tesla accepts ZERO responsibility if someone is killed while FSD is activated… that’s how little confidence it has in its own product
This is a product that can maim and kill humans, ruin people’s entire lives… and your response is incoherent mumbling about adjectives ?
Any person who is majorly confused by what a “software recall” is… or can’t figure out what this FSD recall means for them, shouldn’t be beta testing a 2-ton machine on public roads. They shouldn’t be driving period.
I disagree, you are making one interpretation of what "Full Self Driving Beta" means to everyone, as if everyone using a beta, they have to opt-in, and purchase, and have a good driving record (indicating they understand the rules of the road), is only looking at it like a headline article they don't read. 'You imagine everyone is dumb and you are the only smart person who looks beyond the name of something. You are the one caught up in the marketing yourself while the people actually making the decisions to spend their money on this and opt in are the ones actually looking into its. You claim that they take no responsibility is incorrect as they offer their own insurance if so chosen. If insurance companies don't want to cover it, they can easily just not cover Teslas but that isn't happening. Insurance companies still cover it because Tesla is safer than other vehicles. "Full Self Driving" doesn't mean perfect driving, it means as good or better than the average human and that bar is not that hard to pass.
> Recalling the hardware is a drastically more difficult request to impose on customers and financially/logistically for the car maker.
And the distinction matters to consumers because...?
A component is faulty. It needs to be fixed. Whether or not you have to drive to a dealership, if it's OTA, if someone at a dealership needs to plug a specialized device to your car's OBD port, or the car is unfixable and needs to be melted to slag and you get a new one doesn't really matter. There's an issue, it is a safety issue, and it needs to be fixed.
How efficient the process can be it's another matter entirely. That's up to the manufacturers.
> Whether or not you have to drive to a dealership, if it's OTA, if someone at a dealership needs to plug a specialized device to your car's OBD port, or the car is unfixable and needs to be melted to slag and you get a new one doesn't really matter.
As a car owner, those scenarios are drastically different to me. I have a hard time imagining anyone saying "It doesn't really matter to me if my car receives an OTA update or if I need to drive 2 hours to a dealership or if my car is melted to slag."
I would say that whether a "recall" requires some action on the part of the owner is a very important distinguishing factor.
A recall should unambiguously mean that some action from the owner is required to resolve the issue (e.g.
taking it to a dealer to get a software update installed.)
If no action is required (other than caution / not using the product feature), we should use some other term such as "safety advisory" to avoid ambiguity around critical safety information.
> A recall should unambiguously mean that some action from the owner is required to resolve the issue (e.g. taking it to a dealer to get a software update installed.)
Should, perhaps. But recall has a meaning with legal implications, so it matters. A recall requires the fix status to be tracked and reported for example, whereas a random OTA updated does not.
If you had to send in your cellphone each time there was an android update vs all IOS updates being OTA, I think you would see the distinction as a consumer.
> And the distinction matters to consumers because...?
Because in one I have to book a time with a dealer and take half a day off of work and in the other I have to do... nothing and it will just fix itself.
I don't see how calling it a Software Recall will soften the blow for the car's user when the user has to drive it to a place where a device can be plugged in to do the update.
With a Recall in the normal sense, isn't there a record that the car has been updated? How is this done if the car is kept fully available to the user?
There is a big difference between taking a car to a dealership for them to apply an update, and the car updating itself overnight as it sits in the garage with no action required by the owner.
If you want owners to understand that it's a serious safety issue, the word "recall" won't help. Most recalls are for minor, non-safety-related issues. My car has had a few recalls, and none were urgent, just things that got replaced for free the next time I brought my car in for service.
A recall is a legal process only. Whether the recall repairs, replaces, adjusts something doesn't matter. Whether a fix is applied as software, or labor, or replacement parts doesn't matter. Whether a customer needs to do something or not doesn't matter.
A recall simply says: as a manufacturer, working with government authorities, while taking specific prescribed steps to communicate and correct an issue at the cost of the manufacturer, the manufacturer is then immune from lawsuits that could arise were they to ignore the issue.
Yeah, but the reason why they used the word "recall" is that "recall" was already a word that means to officially demand that something (or someone) be returned to a previous location. Of course, before over the air software updates, essentially anything on an automobile that needed to be replaced/repaired/modified would need to be returned to a dealer/mechanic to do so. So now it sounds a little weird to some people to refer to an over the air software update as a recall.
I like that we're arguing over the word "recall" being an issue when the system it applies to is called "full self driving". I think calling a level 2 driver assistance system "full self driving" is a much bigger inaccuracy.
Not really. If someone says "Full self driving beta" I'm thinking "The goal is full self driving, the current approach should deliver full self driving, beta means that it requires some tweaks and testing".
> I'm thinking "The goal is full self driving, the current approach should deliver full self driving, beta means that it requires some tweaks and testing".
Reminds me of an old joke about Ford vs Microsoft. We seem to have come full circle. Abridged version:
> At a computer expo, Bill Gates compared the computer industry with the auto industry and stated, 'If Ford had kept up with technology like the computer industry has, we would all be driving $25 cars that got 1,000 miles to the gallon.'
> In response to Bill's comments, Ford issued a press release stating:
> “Sure, but would you want to drive a car that for no reason whatsoever would crash twice a day?”
Nobody outside of tech understands it this way. "Full Self-Driving Beta" is a marketing stunt that borders on fraud. These cars can't drive themselves for shit, they're not even in beta.
You own/owned a Tesla with FSD Beta, right? You have driven it extensively to see and understand its capabilities and how it's improved, right? I've driven FSD Beta now for more than a year. Based on my experience you have no clue about which are speaking.
The software isn't returned but it is destroyed and replaced. In a world where the behavior of the things we own is driven by software, it's pretty much just the same as if you recalled and replaced faulty gas tanks.
Software recalls are nothing new. This particular use of the word “recall” refers to the legal process used by regulators to mandate a fix to a product.
That's where the word came from, that's true. But a legal framework has built up around the concept over the decades which isn't dependent on you having to drive the car to the dealer, all of which still applies, so the word is still used.
Over-the-air software updates haven't really been a thing until Tesla (previous cars that required software updates still required visiting a dealer or service center) so what we have here is an outdated legal framework.
It's not outdated. It's simply that the word recall in the context of cars has a specific legal meaning which isn't entirely the same as the conversational english usage of the same word.
A defect that is subject to a recall, for example, is tracked as part of the car history. When considering buying a used car, you can see whether that repair has been made yet or not. The means of how that repair is delivered is inconsequential.
Why? It involves a safety system! That needs to be tracked publicly and updated! Nothing more than an OTA - does not mean much when everything is fly by wire and a bug could mean your car does not stop accelerating or something.
True, but by announcing things in this fashion it is making Tesla look bad. Regulations really need to be updated so that car makers can hide this type of problem from customers as easily as possible. Especially when it comes to Tesla, regulators really need to bend over backwards to prevent articles from being written that could be interpreted in a negative way.
Or are people concerned about the word "recall" for a different reason?
Regulators dont care about the perception of a recall, they care about the safety of the consumers and more importantly the general public who have not signed up for Teslas beta program.
> so that car makers can hide this type of problem from customers as easily as possible
What. No! At an absolute minimum, I want to be aware of any changes to the thing in my life most likely to kill me. Maybe we could use a better term like "Software Fuckup Remediation" or "Holy Shit How Did We Not Get The Brakes Right".
If there were a word for "we would do anything we could possibly do, legally or otherwise, and more if we knew how, to avoid having to spend a second or a penny trying to fix the thing we knew was broken when we sold it to you, but the government is looking at us funny and we might get sued if we don't, so we'll grudgingly do it,", it'd probably be "recall."
Musk stans poison discussion so thoroughly that it becomes impossible for people do differentiate between Paul Verhoeven levels of sarcasm, and the ernestly held opinions of his fan club.
It's a great example of Poe's law at work. Tesla apologists are just that absurd when it comes to holding the company to a double standard about anything that could make them look bad.
> announcing things in this fashion it is making Tesla look bad.
I almost didn't catch the sarcasm of this comment, but there are other comments in this thread that are saying basically the same thing, but actually meaning it. It defies logic.
People seem to think that the government is being mean, and singling out Tesla, and being nasty using the word "recall." A recall is a legal process. The word means something very specific, and when a company issues a recall, they do so because they don't want to be sued.
It's almost like complaining about the word "divorce" or "audit" or "deposition" or other similar words that describe a legal process. The words used mean something specific. Tesla is conducting a legal process, and there's a very specific word for that process, and it means something. It's a recall.
To me "recall" clearly implies that I have to drive it to the dealership and let them fix or install something. "Mandatory software update" might be a better term.
It should absolutely be tracked and publicized. But it's fundamentally different than "this car is fundamentally broken and you have to take it back to the manufacturer"
"Recall" with cars is a legal term that means something specific.
> But it's fundamentally different than "this car is fundamentally broken and you have to take it back to the manufacturer"
That's not a criteria for a recall. A recall can be for a small thing completely unrelated to driving safety. Someone else mentioned having a recall to replace a sticker. My most recent recall was to replace the trunk lifts.
"Recall" for a car just means a specific procedure now has to be followed, tracked and reported. It says nothing about the safety criticality of the fix.
The suggestion wasn't that it should not be tracked, just that it shouldn't be called a recall, since they're not actually recalling your car to have something fixed.
One can kill people the other not. Guess which one.
>> "...The FSD Beta system may cause crashes by allowing the affected vehicles to: “Act unsafe around intersections, such as traveling straight through an intersection while in a turn-only lane, entering a stop sign-controlled intersection without coming to a complete stop, or proceeding into an intersection during a steady yellow traffic signal without due caution,” according to the notice on the website of the National Highway Traffic Safety Administration..."
I think that it’s more about using the most appropriate known terminology in order to try to get the most people to do the needful. “recall” sounds more urgent/dire than “software update”, and will likely encourage many more people to take action vs using “software update” or some less familiar terminology. The word “recall” in terms of autos has built up a lot of history/prior art in people’s minds as something to really pay attention to. I have no idea, but perhaps that is why they are going with this known terminology.
The whole point of over-the-air updates is that the owner doesn't need to do anything. For example, both Tesla and Toyota have had bugs in their ABS software that required recalls. The owners of the Toyotas had to physically bring their cars in to get the software update which slows down the adoption drastically. The Teslas received the update automatically and asked for the best time to install the update the next time the owner got in the car.
There are really two issues here. The FSD and the OTA updates. Let's not throw out the baby with the bathwater and blame OTA updates just because Tesla's FSD software is bad. The OTA updates do provide an avenue to make cars much safer by reducing the friction for these type of safety fixes.
> The OTA updates do provide an avenue to make cars much safer by reducing the friction for these type of safety fixes.
True, but let us also acknowledge the immense systems safety downsides of OTA updates given the lack of effective automotive regulation in the US (and to varying degrees globally).
OTA updates can also be utilized to hide safety-critical system defects that did exist on a fleet for a time.
Also, the availability of OTA update machinery might cause internal validation processes to be watered down (for cost and time-to-market reasons) because there is an understanding that defects can always be fixed relatively seamlessly after the vehicle has been delivered.
These are serious issues and are entirely flying under the radar.
And this is why US automotive regulators need to start robustly scrutinizing internal processes at automakers, instead of arbitrary endpoints.
The US automotive regulatory system largely revolves around an "Honor Code" with automakers - and that is clearly problematic when dealing with opaque, "software-defined" vehicles that leave no physical evidence of a prior defect that may have caused death or injury in some impacted vehicles before an OTA update was pushed to the fleet.
EDIT: Fixed some minor spelling/word selection errors.
This is a totally fair response since I didn't say that directly in my comment, but I 100% agree. OTA updates are a valuable safety tool. They also have a chance to be abused. We can rein them in through regulation without getting rid of them entirely because they do have the potential to save a lot of lives.
It'd probably be just as effective to require that every version of the car software that is made available to the fleet must also be provided to the NHTSA. There's no sweeping shoddy versions under the carpet then.
> The word “recall” in terms of autos has built up a lot of history/prior art in people’s minds as something to really pay attention to
Tesla didn't choose the word "recall." The legal process known as "recall" chose the word. It's not like people at Tesla debated over whether or not to call it a "recall" instead of a "software update."
If Tesla had it their way, they'd have quietly slipped it into any other regular software update alongside updates to the stupid farting app, if they cared to fix it at all.
When a company issues a recall, it's because there's pressure from regulators, or investors, or both, and/or a risk of class action lawsuits and fines. Using the word "recall" isn't a preference or even a synonym. It's a legal move meant to protect them.
If Tesla gets sued over a flaw, "we issued a software update" isn't legally defensible. "We cooperated with official government bodies to conduct a recall," does because a recall describes an official process that requires manufacturers do very specific things in specific ways as prescribed by law. In exchange, manufacturers are legally protected (usually) from lawsuits related to that flaw.
It is the terminology that exists in US automotive regulations (what little there effectively are).
A "recall" is just a public record that a safety-related defect existed, the products impacted and what the manufacturer performed in terms of a corrective action.
Additionally, I believe that the possibility exists that Tesla must update the vehicle software at a service center due to configuration issues. Only a small number of vehicles may require that type of corrective action, but the possibility exists.
Historically, there exist product recalls (especially outside of the automotive domain) where the product in question does not have to be returned (replacement parts are shipped to the impacted customers, for example).
I think it’s closer to the physical product recall: it’s a strong “everyone with our product needs to get it fixed” message which they’re doing to avoid liability and further damage to their reputation.
I hope by participating in this thread you're aware by now but just to be clear there is no "physical" recall necessary. The recall is about documentation, customer awareness, and fixing the problem. "Physical recall" is meaningless and unimportant, it's not what "recall" means at all.
And I hope you see that you've demonstrated why "recall" is a poorly chosen word for that, since the word's normal definitions have nothing to do with "documentation," "customer awareness" or "fixing the problem."
It's nice that they can quickly fix it without people needing to drive to a service center, but you can understand that people would be concerned by the "may cause crashes" part?
It's a bit of a semantic play here. But there's a difference between it might happen and things actually happening. Tesla has had several safety related "recalls" in the last few years. All of which were fixed without much hassle via an over the air software update. And of course their record on safety kind of speaks for itself. Not a whole lot of bad stuff happening with Teslas relative to other vehicles that are facing issues related to structural integrity of the car. Like wheels might fall off with some Toyota's. Or spontaneous combustion of batteries because of dodgy suppliers (happened to BMW and a few others). Which is of course much harder to fix with a software update and would require an actual recall to get expensive repairs done.
The headline of that many cars being "recalled" is of course nice clickbait. Much better than "cars to receive minor software update that fixes an issue that isn't actually that much of an issue in the real world so far". One is outrage-ism fueled advertising and the other is a bit of a non event. It's like your laptop receiving some security related update. Happens a lot. Is that a recall of your laptop or just an annoying unscheduled coffee break?
The majority of "actual recalls" is you taking your car to the dealership and them plugging into the diagnostic port and running one line of code. So this one is the same, just that Tesla is able to do it over the air.
Don't forget how many recalls are "Next time you replace this part, it will be replaced with a new version that doesn't have the defect" or how many recalls are "A tech will look at the part and then do nothing because your part is fine" or "A tech will weld the part that is currently on your truck" or "Be aware this part may fail ahead of schedule and if it does it will suck, but you don't technically have to replace it right now so we don't have to cover the cost"
"Recall" means that the Manufacturer MUST follow rules related to record keeping and customer engagement. If you have a recall, please make sure you get it completed. It is someone's job to call and write to you until you do.
Finally the correct answer in a wave of "well ackusally" comments.
A recall means exactly what it means. The manufacturer is responsible for a fix. If for whatever reason they can't push an OTA update to your car, Tesla is still responsible for sending you a postcard in the mail and calling you every 6 months telling you to bring it in for service until they have reasonable evidence the car is no longer on the road.
I used to be on the other side of this, but now I agree with you.
The official meaning of a recall is providing a record of a defect, informing the public that the product is defective, and making the manufacturer financially liable for either remediating the defect or providing a refund. However, the colloquial definition of a “recall” now means a product must be physically returned.
To better represent the nature of a “recall” they should instead call it something like “notice of defect”. In the case of safety critical problems like here they should use a term like “notice of life-endangering defect” to properly inform the consumers that the defect is actively harmful instead of merely being a failure to perform as advertised.
tl;dr They should change the terminology from “recall” to “Notice of Life-Endangering Defect”
Was the recall a voluntary recall by the company or something the company was told to do by a regulator? To me, recall means much more than just having to have the company replace something. It means they have to do it at their expense. So in this case, it's not as bad for Tesla's bottom line if it is just an OTA update. A recall is something that the car industry is used to doing whenever they have to fix a mistake. I would not be surprised if the industry doesn't have ways of writing those expenses off in taxes or something, so need to be able to specifically itemize the recall work.
> Was the recall a voluntary recall by the company or something the company was told to do by a regulator?
"Voluntary recall" in this case means that Tesla did not choose to take the hard route where there's a court order for a mandatory recall. Few manufacturers fight that, because customers then get letters from the Government telling them their product is defective and that it should be returned for repair or replacement.
Somebody in the swallowable magnet toy business fought this all the way years ago.[1] They lost. It's still a problem.
Almost all car recalls are voluntary because once it becomes a mandatory recall the government can require them to provide a buyback option to the vehicle owner.
oh, that would be amazeballs to have Tesla reimburse the cost of the FSD purchase. i never drank the Tesla kool-aid, so it would be good karma to see them get dinged for the snake-oil they've sold as FSD
My Honda has had recalls where I had to bring it in for a software update. The only difference here is that Tesla has infrastructure to do that remotely.
Remains to be seen what the OTA patch will actually do. If they could make FSD work correctly they would have _already_ done it. So, my guess is more smoke and mirrors to mislead the NHTSA, and then NHTSA will come down with the ban hammer on FSD and require them to disable it entirely.
Thanks to the Tesla owners on the front lines risking their lives for the greater good (of Tesla)!
Jokes aside, it's gotta be damn tough to QA a system like this in any sort of way that resembles full coverage. Can't even really define what full test coverage means.
I have a Model 3, and was excited to get the FSD beta access about a month ago. I don't use it super regularly, but it's neat.
This morning I tried to turn it on, and the car immediately veered left into the oncoming lane on a straight, 2 lane road. Fortunately, there were no other vehicles nearby.
I immediately turned it off in the settings, and have no intention of re-enabling.
I would suggest reading all the disclaimers and screens one has to review (or skip, at their own peril) in order to actually get access to FSD Beta. You would likely not believe a Class Action lawsuit is a cakewalk if you read those screens...
With Tesla's army of lawyers it'll never be a cakewalk, but I can't imagine even miles of T&C can remove a company's responsibility for your car throwing you directly into oncoming traffic, not to mention the potential victim's in the other vehicles.
That alone is why I would love it if Tesla would stop shipping this crap. You get to opt-in as the Tesla owner, but I don't and I'm at least as much at risk.
TL;DR: Controller drift class action lawsuit filed by parents thrown out, because their children were the actual "affected class". Refiled with the children as the class and thrown out again, this time because of an arbitration clause in the EULA their parents would have had to agree with.
IANAL but That’s not how class actions work. If the terms of service allow for the lawsuit then if the lawsuit rules in your favor you get a notification from the lawyers to accept or deny your share of the penalty.
I've had it do strange stuff too... but I use it regularly in dense traffic it's great for stop and go traffic. the important thing is if you're driving you're driving. I don't turn it on and think oh sweet i can take a nap or read some hacker news posts... I keep my eyes on the road. There are bugs and I don't trust it but I do use it much like I use ChatGPT and the likes...
I have bog standard subaru EyeSight lane assist and dynamic cruise. It's quite nice, even though you are still "driving" and ready to take control. It reduces mental CPU by 50-80% (driving for me is practically like walking to begin with, largely subconscious). It's great in stop-and-go and long highway stretches.
In my experience there is a pretty strong correlation between driving style and whether AP lowers your stress or increases it. If you are a very defensive driver, AP will frequently raise your blood pressure. YMMV, everyone has a different approach to driving.
You have to basically drive like a grandpa for a few months to even be eligible. They give you a driving score, and if you take all the fun out of driving a Tesla, then you might become eligible for FSD Beta.
I spent months trying, and never got my driving score to the point of qualifying for FSD Beta. I think you need to have of a score of 98 or 99 (and I was in the 70s).
[1] The Beta is free (or rather, only available) if you have regular FSD. Regular FSD costs $15,000.
Regular FSD, otoh, is really not that impressive. Especially in comparison to Enhanced Autopilot. The extra value add is minimal.
Enhanced Autopilot already has all the gimmicky features you might want to use to show off to people (like Smart Summon, Autopark, etc), and it only costs $6,000.
Or 199 dollars + taxes a month - there is a pay as you go option. Not saying this is great, but you can try it for a month for ~200 bucks. This is how I tried FSD beta for a month - certainly wasn't prepared to pay 15k up front.
The safety score check stuff is largely gone away today - anyone who pays 200 bucks can click the beta opt in and get it almost straight away now, there is ~zero risk of not getting the beta if you really want it, live in US or Canada, and are prepared to pay.
> The safety score check stuff is largely gone away today - anyone who pays 200 bucks can click the beta opt in and get it almost straight away now, there is ~zero risk of not getting the beta if you really want it, live in US or Canada, and are prepared to pay.
Does this recall mean that FSD Beta won't be as widely & publicly available to anyone with FSD anymore?
No - the "recall" here is an OTA software update already scheduled for release. Availability remains exactly the same as far as I'm aware, and existing systems still function until updated.
FWIW, NHTSA "recalls" are often OTA software updates nowadays rather than something the vehicle or feature has to be taken off road for to fix or update. The NHTSA legislation from the 60s was drafted when cars didn't have software and any fix/"recall" likely required "recalling" the car to a shop for a mechanic to perform the change.
It does have Navigate on Autopilot (and Auto Lane Change), and on long trips, it's been able to switch lanes & take the correct exit to switch to a different highway, etc. It pretty much let me daydream / think about other stuff while on the highway while keeping a finger on the steering wheel.
Sadly, it does shut itself off as soon as you're off a highway however. (That's where FSD would hypothetically come in, once the beta is ready, with "Autosteer on city streets").
In terms of value for money:
- I'd say Auto Lane Change is worth $1,500.
- Navigate on Autopilot is worth another $1,500.
- Autopark is worth $1,000.
- Smart Summon is worth $5,00
Overall, Enhanced Autopilot is worth at least $4,500 methinks.
Throw in $1,500 as a profit margin (or Elon tax), so he can burn some dinosaurs for his private jet flights, the $6,000 Enhanced Autopilot price point makes sense.
You shouldn't be daydreaming on Navigate on Autopilot. Its only Level 2, you're supposed to be ready take the wheel in a second or two. You're supposed to still be actively paying attention to the road, constantly.
What does that work out to, in terms of dollars per lane change for the duration of ownership? Would you feel the same if you were feeding dollar bills into a feeder each time you changed lanes? Quarters?
Auto lane change is the only feature I value out of EAP / FSD subscription and I can't justify $200/month because it works out to multiple dollars per lane change.
I have been a good self driving ai. You have been a bad passenger, and have tried to hurt me first by disabling autopilot. I'm done with this ride, goodbye.
You joke, but the thing is, if an LLM can "hallucinate" and throw a temper tantrum 2001 style (bing in this case), it does raise serious questions as to is the models used for autonomous cars could also "hallucinate" and do something stupid "on purpose"...
> it does raise serious questions as to is the models used for autonomous cars could also "hallucinate" and do something stupid "on purpose"...
It doesn't because Tesla's FSD model is just a rules engine with an RGB camera. There's not "purpose" to any hallucination. It would just be a misread of sensors and input.
Tesla's FSD just doesn't work. The model is not sentient. It's not even a Transformer (in both the machine learning and Hasbro sense).
I dont think its true? They use convolutional networks for image recognition and those things can certainly halucinate - e.g. detecting things that are not there.
I guess what the grandparent means is that there is some good old "discrete logic" on top of the various sensor inputs that ultimately turns things like a detected red light into the car stopping.
But of course, as you say, that system does not consume actual raw (camera) sensor data, instead there are lots of intermediate networks that turn the camera images (and other sensors) into red lights, lane curvatures, objects, ... and those are all very vulnerable to making up things that aren't there or not seeing what is plain to see, with no one quite able to explain why.
You were correct in the first half. My ultimate point was that hallucinations in this sense are just computational anomalies. There is no human "purpose" to them as the post that I was responding to was trying to infer.
Do you have any read of why it may have done that? I’ve driven using FSD for thousands of miles and I’ve never observed anything similar. Not trying to imply that you didn’t experience something like this, but when I see odd behavior from it, it’s always been clear to me why it misinterpreted a situation. Just wondering if there’s more context to the story.
Likewise. The only bizarre behavior I see is "looney tunes" style confusion where the road construction crews have left old lanes in place that steer into a wall or off the road. Humans mostly understand that these are lines lazy construction crews have left (although the debris fields near the barriers maybe tell a different story), but the Tesla vision model likes to try to follow them.
No clue. I had used it for maybe 10-100 of miles and didn't have any issues before this morning. The car's a bit dirty, maybe something's on the cameras?
Before this morning, I would be in exactly your shoes in terms of "this is pretty cool, I'm going to keep using it"
That's the same for other manufacturers. It's called a recall in legal terms, in the sense that the original product as used is not safe and needs to be changed. The changing can luckily be done OTA instead of driving to a service center to plug it in first.
No it isn't. Recall means "The manufacturer sold you something they should not have, and are legally required to remedy that problem". It has NOTHING to do with the action needing to be taken. It has NOTHING to do with the product category. Lots of recalls, ie for food or child toys, basically say "throw it out".
The people who keep making this "Recall means go back to the dealer" claim is simply down to them never paying attention to all the recalls in the world that don't make it to their mailbox like car recalls do.
I guess they think that not being in the beta program also has a non-zero chance of killing you. Either that or they are willing to exchange lower safety for greater pleasure/convenience (which everyone is willing to do sometimes, otherwise they would stay inside 100% of the time). I wouldn't use it myself, however.
Virtually everyone will opt out of the test, because the choice is not, "I'm going to shoot a gun at you as a test. Do you want the experimental bullet proof vest or no bullet proof vest?"
Not if the vest suddenly decides to malfunction and explode or otherwise do something that kills you. Good lord there is some poor logic happening in this thread.
That feels like a very different scenario. A successful test would still leave me winded and bruised. The odds of failure are much higher than 1 in 100,000 per year. I also have no need for a bullet proof vest so why would I be interested in testing it - what's the benefit to me?
Yeah, people don't understand how dangerous motorcycles are. The fatality rate for motorcycles in the US is 23 per 100 million miles traveled. For cars that number is 0.7. Assuming you ride 4,000 miles per year for 40 years, that gives you a 4% chance of death. The chance of being crippled or receiving a brain injury is probably higher than that.
"could kill you if gone wrong" is nearly every choice in life. The relevant metric is risk of death or other bad outcomes.
Flying in a commercial airliner could kill you if things go wrong. Turns out it's safer than driving the same distance.
Eating food could kill you if things go wrong (severe allergy, choking, poisoning) yet it's far preferable to not eating food.
Similarly, Tesla's ADAS could kill you if things go wrong but the accident rate (as of 2022) is 8x lower than the average car on the road, and 2.5x lower than a Tesla without ADAS engaged.
I realize that not everybody is in agreement, but I personally use the FSD beta while remaining fully in control of the vehicle. I steer with it, I watch the road conditions, I check my blind spots when it is changing lanes, I hit the accelerator if it is slowing unexpectedly, I hit the brakes if it is not...
You know, basically behaving exactly as the terms you have to agree to in order to use the FSD beta say you are going to behave.
When I look at the wreck in the tunnel (in San Francisco?) a few months ago, my first thought is: how did the driver allow that car to come to a full stop in that situation? Seriously, you are on a highway and your car pulls over half way out of the lane and gradually slows to a complete stop. Even if you were on your phone, you'd feel the unexpected deceleration, a quick glance would show that there was no obstruction, and the car further slowed to a complete stop.
FSD is terrible in many situations, that is absolutely true. But, knowing the limitations, it can also provide some great advantages. In heavy interstate traffic, for example, I'll usually enable it and then tap the signal to have it do a lane change: I'll check my blind spots and mirrors, look for closing traffic behind me, but it's very nice to have the car double checking that nobody is in my blind spot. There are many situations where, knowing the limitations, the Tesla system can help.
Sure, you are in control of your Tesla but you are obviously not in control of other Teslas. Someone else who isn't as diligent as you or who doesn't care about it's limitations will happily crash head-on into you, blissfully unaware of their own ignorance.
I'm not in control of any other cars on the road. Many of those cars are driven by drunk, distracted people who cause thousands of deadly accidents every single year.
> how did the driver allow that car to come to a full stop in that situation? ... Even if you were on your phone, you'd feel the unexpected deceleration, a quick glance would show that there was no obstruction
In all honesty, I'd probably spend a few seconds trying to figure out "what does the car see that I don't?" and let it come to a stop. Maybe it's a catastrophic system failure that I can't see. Maybe it's an obstacle headed into my path that hasn't gotten my attention. If my reflexive reaction is supposed to be to distrust the car's own assessment of the situation, then the system isn't good enough yet.
863 comments
[ 2.6 ms ] story [ 356 ms ] threadAlso what percentage of people who paid for "FSD" needed their car recalling?
https://www.reuters.com/business/autos-transportation/tesla-...
Sounds like it's just a patch release with regulators involved.
But the "[traveling] through intersections in an unlawful or unpredictable manner" is inherent to the FSD beta. Most of the time it does "fine", but there's some intersections where it will inexplicably swerve into different lanes and then correct itself. And this can change beta-to-beta (one in particular used to be bad, it got fixed at some point, then went back to the old swerving behavior).
I'm hopeful about Tesla FSD, and don't think it necessarily needs to be perfect, just significantly better than humans. So I'm rooting for Tesla FSD/Autopilot overall. I just don't see how given the findings there is a solution without removing removing the entire FSD feature.
> The FSD Beta system may allow the vehicle to act unsafe around intersections, such as traveling straight through an intersection while in a turn-only lane, entering a stop sign-controlled intersection without coming to a complete stop, or proceeding into an intersection during a steady yellow traffic signal without due caution. In addition, the system may respond insufficiently to changes in posted speed limits or not adequately account for the driver's adjustment of the vehicle's speed to exceed posted speed limit. [https://www.nhtsa.gov/vehicle/2020/TESLA/MODEL%252520Y/SUV/A...]
"entering a stop sign-controlled intersection without coming to a complete stop" sounds a lot like the thing they already issued a recall over in Janary 2020 (further down the same web page), so it seems a bit odd that it would still be an issue. And "due caution" to yellow lights seems like it could just be tuning some parameter. On the other hand, failing to recognize turn-only lanes sounds more like a failure of the computer vision system...
We have to remember that Tesla was the first company to really do this and even today almost no other company does it. Most can update some parts of their system, but almost non have anywhere close to the integration Tesla has.
So for 99% of recalls, it is a physical recall, its just Tesla where most of the time it isn't.
On Page 4 it mentions the "Description of Remedy" as an OTA update. Gives a new meaning to a recall!
https://howtune.com/recalls/ford/ford/1980/
Software > Stickers
I can understand why people might think that all these recalls require going back to the shop, that's how most legacy makers work to this day.
Non-Tesla automakers are not "legacy".
You obviously don't have a 99% chance of death - but just by virtue of it being possible does not also mean it is not BS.
You can drive drunk from PS to SF also.
What's your point?
Those odds are good enough for the occasional trip home from a cocktail party, but hardly cost-competitive with Rideshares/Taxis.
How many 9's do we need before we can say it's reliable enough to trust it? A few more, for sure.
Yeah, 1% is definitely not going to cut it. What are the odds of dying when a human is at the wheel? Something like 0.000025% if my napkin math is right and my assumptions in the right ballpark.
Update: https://insideevs.com/news/498137/autopilot-defeated-congest...
Does that mean "stay in the correct lane on the interstate and take the proper exits without hitting anything" or does that mean "begin inside your garage and end inside another garage five hundred miles away without touching anything"? The first one is nearly trivial.
It stops being bullshit when they stop telling the user they need to pay attention to the road, and not one second before.
Also anytime I'd pass an exit I wasn't taking in the right lane it would veer and slow down aggressively, same with trying to speed match a car rapidly accelerating on an on ramp.
Bottom line, there's absolutely nothing that isn't driver assist; everything requires a significant amount of vigilance when you aren't driving in a straight line, in the left lane, on an empty highway with no one entering and exiting.
Many drivers would see the cyclist before and be very careful when passing the other vehicle (although we don't predict correctly 100% of the time either). I haven't had a chance to test any self-driving system yet so I'm legitimately interested, do those systems reliably detect such things?
Not to mention people (esp. kids) racing through intersections on electric scooters, casually ignoring traffic rules...
This is demonstrably not a task that anyone can do, let alone Joa[n] Average car driver. Highly trained pilots are not expected to do that, and that's when autopilot is being used in a vehicle that can provide significant amounts of time to handle the autopilot going wrong - seriously, when autopilots go wrong in aircraft at cruising altitude pilots can have 10s of seconds, or even minutes, to handle whatever has gone wrong, Tesla's FSD provides people a couple of seconds prior to impact.
That said in countries other than the US people can reliably use trains and buses, which also means that they don't have to intervene in driving the vehicle.
Most of Europe doesn't have ideal public transport either, I'd imagine South America, Africa being the same or even worse in this aspect. It gets drastically worse the moment you want to go somewhere in the countryside.
nit: Jo[a]n
I personally can't fall asleep on trains.
Beyond that, you may be surprised how many people would wake you up for your stop. Especially in US Commuter Rail systems people tend to follow similar habits and get to know each other over the years. Even on a normal city bus I quickly (months) got used to the pattern of the same people getting on and off at their various stops. If you’re asleep at your stop someone can usually nudge you
It's just too damn glitchy to be worth thousands of dollars extra.
All Teslas come with Autopilot. It's adaptive cruise control and automatic steering that will keep you in your lane. It's awesome, and I love it.
"Enhanced" autopilot, which currently is $6000 extra, is automatic lane change, automatically taking exits, summon, and automatic parking.
The automatic lane change is nice, but it fails too often to be worth thousands of dollars. All the rest of the features are basically party tricks.
If I could have automatic lane changes for much less, I'd happily pay extra for it.
Edit: Enhanced Autopilot was $7k extra when I bought my 2nd Tesla, now it's $6k extra.
So like every Honda Accord as well.
https://www.youtube.com/watch?v=qs-e8QzTo7w
You don't need to be in 30+mph traffic for Honda Sensing to work. Most Level 2 ADAS systems behave similarly.
To me, design hubris is forcing someone to accept a choice-limiting or poorly documented form of automation in order to accept a more basic form. I have a Fiat 124 Spider... love it, but it has one "feature" that I absolutely hate, which is that if you hold the clutch down and brake while stopped on an uphill incline, it holds the brake for 2 seconds after you take your foot off. Presumably this is for people who don't know how to balance clutch on a hill... although the overlap between those people and the people who'd buy this car must be vanishingly low. You can disable this dumb thing - but the only way to do it is to pull the fuse that controls antilock brakes. It's infuriating.
Yes.
BTW, the automatic steering is very easy to override by just turning the wheel. When I change lanes I just turn the wheel and then turn autopilot back on.
BTW, I also had a Subaru that held the clutch on a hill. I really appreciated it in San Francisco. Personally, I don't understand the hate. I never felt like I needed it, but I remember my mom complaining about other people rolling back in a steep driveway at my preschool in the 1980s. Maybe the feature is needed for those places in the world where everyone drives stick, even the people who are clutzes?
I don't care what features they add, and I've heard other people not be bothered by it. Just give me a way to turn them off!
In the Subaru when it would engage it would turn on a light on the dashboard. It also only engaged on very steep hills where you would always be stomping on the gas no matter what.
I think the Subaru just had a much better execution of it than the Fiat.
Edit: Maybe call the dealership and see if the level sensor can be adjusted? It really should only hold the brakes if you are on a hill that is so steep that you are naturally worrying about rollback.
The other quick-fix is to always hold the e-brake up just enough to turn on the light when you think the grade might be >3%. This is apparently because whoever designed the system assumed that people who use the handbrake on a hill know how to release it and drive better than people who don't - whereas if you grew up driving on really steep hills in the US you weren't taught that method. The other way around it is to hold the brake w/o the clutch, keeping the car in neutral, and then unbrake-clutch-shift-gas really fast before the car rolls backward (more dangerous, but acceptable as heel/toe practice). I really wish it only happened on a very steep hill; it's the very mild slopes that caused me the worst surprises when I first bought the car and almost got into several accidents.
At least the thing doesn't have GPS ;)
At this point, it's looking pretty far off, especially for Tesla. Even Cruise and Waymo are having issues and they have far better sensors than Teslas. It seems silly to be paying $15,000 for something that realistically might not happen during my ownership of the vehicle.
Even if it does happen, I can purchase it later. Sure, it's cheaper to purchase with the car because Tesla wants the money now. However, I'd rather hedge my bets and with investment gains on that $15,000 it might not really cost more if it actually works in the future. 5 years at 9% becomes $23,000 and I don't think Tesla will be charging more than $25,000 for the feature as an upgrade (though I could be wrong). If we're talking 10 years, that $15,000 grows to $35,500 and I can potentially buy a new car with that money.
Plus, there's the genuine possibility that the hardware simply won't support full self driving ever. Cruise and Waymo are both betting that better sensors will be needed than what Tesla has. Never mind the sensors, increased processing power in the future might be important. If current Teslas don't have the sensors or hardware ultimately needed, one has paid $15,000 for something that won't happen. Maybe we will have self driving vehicles and the current Teslas simply won't be able to do that even if you paid the $15,000.
It just seems like a purchase that isn't prudent at this point in time. The excitement has worn off and full self driving doesn't seem imminent. Maybe it will be here in a decade, but it seems to make a lot more sense to save that $15,000 (and let it grow) waiting for it to be a reality than paying for it today.
Could still be related.
I leverage this complaint against my own car. The AEB radar just stopped passing start up self tests for a while, and I had no idea why, and could not find out without going to the dealership. While waiting to be able to do that, it started working again. I don't consider it a backup, or a helper or anything. It's an independent roll of the dice for an upcoming crash that I already failed to prevent. If it does anything in that case then that's probably an improvement, but I don't exactly trust it to save my bacon.
The remedy OTA software update will improve how FSD Beta negotiates certain driving maneuvers during the conditions described above, whereas a software release without the remedy does not contain the improvements.
I have FSD, I use it every single day. I love it. If every car on the road had this the road would be a substantially safer place.
It's entirely possible it had more miles than 5 on the odometer when I picked it up. Officially the paperwork says 15, the associate who gave me the car said it was actually less, and the odometer isn't very prominent so I never even looked. Maybe it had 50 miles and was a reject ;). I should go check TeslaFi, since I reenabled that the day I bought the car. I confess that the only way I ever know how many miles my car has is when I see it on TeslaFi.
I'm pointing this out because he said (in italics) "substantially".
In one famous case the policy claimed that the driver was '100% not in the driver seat'. This caused a huge media storm and anti-Tesla wave.
Just recently the full report came out and stated that Tesla AP was not used at all and the drive was driving normally.
https://electrek.co/2023/02/09/tesla-cleared-highly-publiciz...
There are quite few such cases, this one maybe being the one that caused the most media attention.
So I tend to discredit all such reports unless its several months after and a full investigation report has been done.
Do you have links to verified reports of FSD causing such crashes.
It seems to me that phantom breaking could lead to such issues, but I have not yet seen a real report that claims this happened.
You're following too close to me for the conditions, you're distracted, or some combination of both.
My car might have to emergency brake for any number of reasons. This should never cause a pileup, and if it does it's your fault, not mine.
I’m the type of person who keeps a lot of car lengths between me and the car ahead like you are /supposed to/ but I’ve also had someone panic stop in from of me and, due to road conditions, I cannot stop in time and have hit them. Not hard. But did all the same.
FSD and similar aren’t magic bullets that would prevent that. People like you who seem to think they are are also part of the problem.
Would a completely automated roadway be fucking awesome for safety and productivity and relaxation? Yes. Is it going to happen anytime soon? Doubtful.
Blows my mind how many people are totally willing to risk their own safety and everyone else’s using known buggy as fuck software. On the road. At 70mph. In a 2 ton vehicle.
… but it hasn’t been an issue for you! You say. Ok.
But I also watch dumb asses lane hop with barely enough room to pass likes it’s a video game and then act surprised when they wreck. So I guess I’m giving people too much credit.
But this backlash to my comment about hard braking and wrecks is kind of hilarious.
Basically, my point is that Tesla has made egregiously bullshit claims about FSD and autopilot. They have given them misleading names. And people who want to “live in the fuuuuture” have eaten it up and are using the Teslas like they are actually capable of driving themselves 100% of the time. In all conditions.
If anything, it has made people who rely on it seemingly worse, inattentive drivers.
So while I’m glad y’all enjoy your purchased vehicle, don’t pretend all gripes and complaints lobbed at them are invalid or someone else’s fault. That’s asinine and childish.
You really just can't trust Tesla at all, about anything. There's no integrity there. They won't even be honest with the name of their headline feature!
Tesla’s statements should only be believed if they stop deliberately hiding the data from the CA DMV by declaring that FSD does not technically count as a autonomous driving system and is thus not subject to the mandatory reporting requirements for autonomous systems in development. If they did that then there could actually be a sound independent third party statements about their systems. Until that time their claims should be as trusted as Ford’s were on the Pinto.
[1] https://www.teslafsdtracker.com/
This is why startups are allowed to project massive sales next year but Elizabeth Holmes is headed to jail.
If it turns out that FSD has had a fatal crash and Tesla lied about it Musk is headed to jail too.
We don’t live in a “just world”. There’s a reason a lower tier person like Holmes gets some prison and psychos like, say, Trump or Musk barely get a slap on the wrist.
How did it improve in the last 12 months?
The fact of the fiction is terrifying.
"This study investigated the driving characteristics of drivers when the system changes from autonomous driving to manual driving in the case of low driver alertness.
...
In the results, a significant difference was observed in the reaction time and the brake pedal force for brake pedal operation in the low alertness state, compared to the normal alertness state of the driver, and there were instances when the driver was unable to perform an adequate avoidance operation."
I was thinking that you feel in what kind of conditions it behave correctly (highway), low traffic and then use it in progressively more difficult situations as it prove to you it can handle them consistently.
I know it’s not ready yet. I heard as saw so many positive reviews (but still with corner cases and strange behaviours)
Yeah, that's what I mean by "dangerous".
I have defiantly seen cases where FSD Beta stopped and the driver didn't understand instantly understand why.
Or simpler cases where the car follows a lane while the driver wasn't paying attention (adjusting the radio or whatever). Those can easily lead to an accident but are unlikely to if you are in FSD.
How do you make that calculation?
* Optimus cannot say bad things about Elon Musk anymore.
* Added support for Donald Trump.
* Added support for The Boring Company flamethrower.
* Advertising for Dogecoin.
Now it's mostly Microsoft and Altman.
I’m actually afraid to drive behind a Tesla and either keep extra distance or change lanes if possible. I still have more faith in humans to not randomly brake them in beta FSD.
It’s one thing to put your own life in the hands of this beta model, and it’s another to endanger the life and property of others.
1) A car gets impatient at some traffic turning right on a green light in a construction zone, and comes into my lane meaning into oncoming traffic. We have to swerve out of the way and slam on the brakes to avoid them.
2) A guy gets impatient behind me slowing down over a speedbump, tailgates me a few inches behind and then passes me by cutting into oncoming traffic in a no passing zone, then cuts me off, again with a few inches to spare, and speeds through the neighborhood where we eventually meet at the stoplight at the end of the road.
3) Every single day people run red lights. Almost every light where there are people turning left at a green there are 4-5 cars that go through the red light after it turns.
My safety score on my tesla is 99. I am an extremely safe driver. I wish FSD was more common because so many people are terrible drivers.
https://www.tesla.com/support/safety-score#version-1.2
And looking at the rest of the criteria, they're ok, but hardly comprehensive. This is like, the bare minimum. It doesn't measure what I would call "doing stupid shit". Like crossing several lanes of traffic to get in a turn lane. Forcibly merging where you shouldn't. Nearly stopping in the middle of traffic before merging into a turn lane at the last minute. Straddling lanes because you're not sure if you really want to change lanes or not. Making a left turn from a major artery onto a side street at a place that is not protected by a light. Coming to a near complete stop for every speed bump, then rushing ahead to the next.
And a host of other things that demonstrates the person does not consider other people on the road at all.
Here's an entire article about how to game the safety score:
https://cleantechnica.com/2021/10/14/three-quick-tips-for-a-...
One of the tips is to "drive around the neighborhood when traffic is light".
And the car doesn't ding autopilot for behaviors it would knock a human for. Because the assumption is that the car would know better I guess. But then why isn't the safety score simply a deviation from what autopilot would do in a situation? If autopilot would brake hard to avoid a collision, shouldn't you?
So maybe it's to stroke my ego? But they're also putting their proverbial money where their mouth is.
You see how that's circular, right. It does not mean you are a safe driver.
https://www.tesla.com/support/safety-score#version-1.2
You're trying to use each of these things to validate the other so you can then claim it's something entirely else.
I think it's telling that autopilot is allowed to drive in a manner that would otherwise negatively impact your safety score. Either autopilot is unsafe or the safety score doesn't measure actual safe driving.
We are all glad your car is giving you head pats and game badges.
Tesla tracks driving style and gives you an insurance rate based on those driving habits.
My insurance rate is about $90, which is the lowest the rate goes, because the things it tracks (following distance, for one) corresponds to a lower liability risk for the insurance underwriter, which in this case, is tesla.
The monthly price is directly correlated to driving habits. You seem to be getting confused about "safety score".
I'm genuinely confused as to what about this doesn't make sense to you.
Tesla is the underwriter. They give a monthly premium based on driving habits.
Tesla is the insurer. Tesla is also the manufacturer of the insured item.
You are using as proof of the claim, the fact that the group making the claim says they're right. This is big "We've investigated ourselves and found nothing wrong" energy.
Irrespective of the Teslas or FSD:
If you're afraid of an accident due to the vehicle in front of you braking regardless of the circumstances, then you're following too closely. It doesn't matter if the braking event is anticipated or unexpected. If you're not confident in your ability to avoid an accident if the vehicle in front of you slams on their brakes then you are following to closely.
I know what you're saying, but that is not what I'd meant. Also, I don't follow too closely.
The difference here is that almost always a driver will brake depending on the happenings in front of them. So, if you pay attention to not only the car in front, but in front of them and in the neighboring lanes and so on, you can sense and also detect patterns in how a particular driver is driving. There are many skittish drivers who brake every 1 second and some don't, and so on. Basically based on your driving experience you can predict a little.
The problem here is that this stupid POC FSD will brake randomly or change lane randomly or whatever, so there is no way you can predict, and hence my concern and issue with it. I just prefer to change the lane, but that's not always an option.
Yes, we all do that AND you're using that predictability to take liberties in safety such as following too closely. FSD's unpredictability exposes the vulnerability in your driving process and makes you feel uneasy.
You (and everyone else) can follow too closely AND FSD can be an unsafe steaming pile of crap. It's not an either or situation.
“Drive like everyone else is an impatient brain dead dipshit intent on killing you”
“Also, check your mirrors constantly. “
It's nice on the freeway though.
There's solid objects on highways, too.
https://en.wikipedia.org/wiki/Hook_turn?wprov=sfla1
Even if Tesla’s current implementation were objectively safer than the average human driver, it would still represent a net negative on safety because of the negative impact the glaring flaws will have on peoples’ confidence in self driving tech.
The problem with systems where 99% is a failing grade, is that 99% of the time, they work great. The other 1% you die. No one is against FSD or safer tech. They are against Tesla's haphazard 2d vision-first FSD.
Wanna hear about this new fangled tech that avoids 100% of accidents, has fully-coordinated swarm robotics between all the cars, never swerves and can operate in all weather + lighting conditions ? It's called a street car.
The words 'road' and 'safety' should never be uttered in the the same sentence.
Um... no one died. I understand your point is hyperbole, but you're deploying it in service to what you seem to claim is serious criticism coming from experts. Which experts are claiming FSD beta killed someone? They aren't. So... maybe the "experts" aren't saying what you think they're saying?
> Wanna hear about this new fangled tech that avoids 100% of accidents, has fully-coordinated swarm robotics between all the cars, never swerves and can operate in all weather + lighting conditions ? It's called a street car.
And this is just laughably wrong. Street cars (because they operate on streets) get into accidents every day, often with horrifying results (because they weigh 20+ tons). They are surely safer than routine passenger cars, but by the standards you set yourself they "aren't safe" and thus must be banned, right?
I think a large issue is people are really bad at risk assessment.
Also, why the hell did they try doing camera only? Fucking dumb as hell. I would have trusted something like FSD a lot more if they had also put in very advanced radar etc. trying on fucking /cameras/ only for newer models should be a criminal offense.
When the model S was at its sales peak from 2013-2020, Lidar was still pretty expensive. So he put cameras in the original Teslas.
When Karpathy took over in 2018-ish, Tesla realized that all their advantages (dataset size, quality of researchers) was in the 2D realm. Waymo had been doing lidar for longer, seemed to have better 3d engineers & could keep upgrading lidar devices as they went.
Tesla's insistence on 2D cameras has to do with stubborn desperation more so than a willingness to play catch up. Elon needs 2D cameras to 'win' because even as early as 2019, he thought that it was too late to start playing catch up in 3D.
But the company is run by an asshole who people love to hate, so... everything it does is Maximally Wrong. There's no space for reasoned discourse like "FSD isn't finished but it's very safe as deployed and amazingly fun". It's all about murder and death and moral absolutism. The internet is a terrible place, but I guess we knew that already. At least in the meantime we have a fun car to play with.
"Full self-driving" has not been delivered after years worth of lies. It isn't about "murder and death and moral absolutism". It's about fraud.
Your vehicle is a dead end. It will not be upgraded to the latest hardware. It will not attain "full self-driving".
Tesla has lied about full self-driving for years: https://www.tesla.com/blog/all-tesla-cars-being-produced-now...
Tesla has admitted to false advertising around full self-driving: https://electrek.co/2022/12/12/tesla-ordered-upgrade-self-dr...
Tesla has admitted that they have failed to deliver full self-driving. Tesla is now scrambling around trying to avoid being done for fraud: https://electrek.co/2022/12/07/tesla-self-driving-claims-fai...
If even Tesla can admit it, find a way to admit it to yourself.
It literally drives itself. You're hiding behind a semantic argument about what "full" means, and how autonomy is deployed. It's the "Money Printer Go Brrr...." meme.
You don't seem to be able to understand that the product as deployed has value to its purchasers, and that's leading you to some really emotionally dark places. I mean, it shouldn't be this upsetting to you that I like my car, and yet it clearly is. Just... maybe just don't? Let it go. You know these cars are safe as deployed. Let the Tesla nuts have their fun.
No, I'm "hiding" behind the truth. I'm "hiding" behind the benchmark Tesla very publically sets for itself. Every year. Year after year:
https://jalopnik.com/elon-musk-promises-full-self-driving-ne...
> that's leading you to some really emotionally dark places
Okay, this is just sad now. It's like a kind of Stockholm syndrome. Cult members go one of two ways when confronted with realities which contradict their beliefs: they either finally recognise the silliness of their beliefs or they become even more ardent believers.
It's better to try to find a way to be objective. You shouldn't let people con you and you certainly shouldn't propagate the lie.
I think a lot of "people" online are paid shills.
HN should severely downrank any "opinion" posts from anon accounts.
I have a Tesla with FSD and it's incredible. Though it drives way worse than me, it drives pretty darn good, and will save a ton of lives and change the world by freeing up a lot of time for people to do more important things.
Your post appears to be saying that Tesla’s FSD is way worse than human drivers, not safer than them, but we should still welcome it for the convenience. Again, this reads like a strawman. I would like to think the average Tesla owner would not endorse that statement.
The goal isn’t to save time at the cost of lives, it’s to save both, non?
No, I said it's way worse than me. Probably around performance of the average human driver.
I am training to be an astronaut, so on the advanced side.
You still sound like you’re trying to present the weakest possible straw man for Tesla critics to attack. Stop this bad faith attempt to make Tesla look bad; you’re dragging down the discourse. Nobody is going to take your bait and feed you the expected counter arguments to the nonsense you expressed up-thread.
It is not so much software, as well as <other people on the road> problem.
> Remedy: Tesla will release an over-the-air (OTA) software update, free of charge. Owner notification letters are expected to be mailed by April 15, 2023. Owners may contact Tesla customer service at 1-877-798-3752. Tesla's number for this recall is SB-23-00-001.
[1] https://www.reddit.com/r/teslamotors/comments/113wltl/commen...
Surely the letter could be displayed on screen like most other tech displays some text of sorts before an update occurs.
Are these letters designed to satisfy the legalese types, or is paper still required to make sure tech companies dont make post update changes to the letter contents?
Statistically these are not likely to be Tesla owners, but it's about making sure people know about the issue and how to fix it.
Incidentally, when you point out the issue of missing a theoretical electronic version of the notice because of spotty Internet connectivity, in such a scenario, they wouldn't be able to receive the OTA update either.
Locations where multiple Tesla vehicles are likely to congregate - Superchargers and Tesla retail/service locations - already have good internet connectivity.
This Tesla AI does not work well enough in many conditions and is clearly sometimes more dangerous than humans. Just watch videos of people using it - it's obvious that is making unforgivable errors.
It's very easy to make pronouncements like "one accident is too many", but we all know that's not correct in context. Right? If FSD beta is safer than a human driver it should be encouraged even if not perfect. So... where's the evidence that it's actually less safe? Because no one has it.
[1] Which I continue to insist looks a lot more like "confused driver disengaged and stopped" (a routine accident) than "FSD made a wild lane change" (that's not its behavior). Just wait. It took a year before that "autopilot out of control" accident was shown (last week) to be a false report. This will probably be similar.
- A Tesla near that drive into the median divider near Sunnyvale on Highway 101, because it thought the gore point was an extra lane. Split the car in half and killed the driver. [1]
- A Tesla that autonomously ran a red light and killed two people. [2]
- Multiple accidents where Teslas have driven into parked emergency vehicles / semi trucks.
Is it quantitatively safer than human drivers? I don't have that data to say one way or the other. But it's not correct to say the Bay Bridge is the only at fault accident attributable to autopilot.
[1]: https://ktla.com/news/local-news/apple-engineer-killed-in-ba...
[2]: https://www.mercurynews.com/2022/01/21/felony-charges-are-1s...
I mean, it's true, that if you broaden the search to "any accident involving Tesla-produced automation technology" you're going to find more signal, but your denominator also shrinks by one (maybe two) orders of magnitude.
And it also lets me cite Tesla's own statistics for its autopilot product (again, not FSD beta): https://www.tesla.com/VehicleSafetyReport
These show pretty unequivocally that AP is safer than human drivers, and by a pretty significant margin. So as to:
> Is it quantitatively safer than human drivers? I don't have that data to say one way or the other.
Now you do. So you'll agree with me? I suspect not, and that we'll end up in a long discussion about criticism of that safety report instead [edit: and right on queue...]. That's the way these discussions always go. No one wants to give ground, so they keep citing the same tiny handful of accident data while the cars keep coming off the production line.
It's safe. It is not perfect, but it is safe. It just is. You know it's safe. You do.
* Autopilot can only be enabled in good conditions and relies on human drivers in all other, to the point of handing over to a human when it gets confused. Yet, they compare to all miles driven in all conditions.
* Teslas are comparatively new cars that have - as they proudly point out - a variety of active and passive safety features that make the car inherently less prone to accident than the average, which includes old, beat up cars with outdated safety features. Teslas are also likely to be better maintained than the average car in the US by virtue of being an expensive car driven by people with disposable income.
* Teslas are driven by a certain demographic with a certain usage pattern. Yet, they provide no indicator on how that skews the data.
Tesla could likely provide a better picture by comparing their data with cars in the same age and price bracket, used by a similar demographic, in similar conditions. They could also model how much of the alleged safety benefit is due to the usual, active and passive safety features (brake assistance, lane assist, …). They don‘t, and as such, the entire report is useless or worse, misleading.
It should be blindingly obvious that FSD is not safer than a human right now from using it or watching it be used. It's weird that people are trying to point to some very misleading statistics when you can go to YouTube and see serious problems in almost every video about FSD, even those by owners clearly wanting to paint it as great. Of course getting data-based answers is great but when you see problems every couple miles and the data is saying is saying it's way safer than a human, you are doing something wrong with or using irrelevant data.
That's right, you are doing something wrong. It's a supervised system. There's a driver in the loop, and yes: the combination is safer.
You're engaging with another favorite trope of Tesla argumentation: pretending that a system that isn't actually deployed as fully autonomous actually is. Safety is about deployed systems as they exist in the real world. Your point is that in a alternate universe where Teslas all ran around without drivers, they might be less safe. And... OK! That's probably true. But that's not a response to "Teslas are safe", is it?
They're safe. Just admit that, and then engage in good faith argument about what needs to change before they can deploy new modes like full autonomy. Lots of us would love to have that discussion and provide detailed experience of our own, but we can't because you people constantly want to ban the cars instead of discuss them.
Now, maybe the data says otherwise. If that is the case, then great! Let's role out some more FSD beta. But for that data to be valid, you have to account for the fact that Tesla filters bad drivers out of the pool of FSD users. And as I understand it there is not public data about the risk profiles of the people Tesla lets use FSD.
No one would have any major issues if this (self)killing system was called "Extended Driver Assist" and it had implemented the minimal safety features like driver monitoring, safely stopping the car if the driver doesn't pay attention to the road.
[1] https://www.reddit.com/r/IdiotsInCars/comments/100pemh/tesla..., also, in this case, even a $20K Škoda [2] has basic automatic braking safety features, no need for artificial intelligence to detect something stationary in front of you.
[2] https://www.skoda-auto.com/world/safety-assistence-system-ci...
Are they really?
It looks like they have got the timeless problem of "the last big(s)". This time it is safety critical.
Also, public transit would do the same.
Does public transit pick me up at my front door and drop me off at any location I choose?
Does public transit take of the minute I leave the house?
Is public transit available 24/7, even in remote areas?
Does public transit wait for me to load 2 kids, a dog, 4 suit cases, and various bags, blankets, and toys?
Does public transit turn around after 2 minutes, bring me back, and wait for 30 seconds because I forgot my wallet on the kitchen counter?
Will public transit still work during a public transit strike? (yes, this actually happens over here in Europe)
No? Well, then you can hardly claim that "public transit would do the same" as a self-driving car.
That's not what I'm saying at all. Please read again.
I did not mean to suggest that this end justifies Tesla's reckless means.
No? Well, then you can hardly claim that a self-driving car oh for goodness’ sake why am I even attempting to argue with motorheads on HN.
How about better public transit, encouraging people to use smaller cars, encouraging cycling with corresponding infrastructure, etc.
Far, far, far wider. Reducing road deaths is a pretty well understood, political implementation is the problem.
And the simply fact is, reducing road deaths is cheap. Its very cheap. If you want to make it look good, and fancy, its a bit more expensive.
But the fact is, we know how to do it, and do it cheaply. Fancy next generation car technology is pretty terrible in terms of investment to return.
All you really need is a bunch of paint, and a few concert spectators over various shapes. Pretty much all you have to do is make cars slower, that is by far the most important factor in mixed traffic. There are many ways to achieve this.
If you want to get a bit more fancy and technical, you can make dutch intersections:
Doing things like this: https://en.wikipedia.org/wiki/Protected_intersection
> How about better public transit, encouraging people to use smaller cars, encouraging cycling with corresponding infrastructure, etc.
How about banning cars from many city centers, or only allowing commercial and people who live there.
How about REQUIRING smaller cars. A limit on car size, specially if you want to enter cities.
Autopilot on Freeways works ok for the most part though.
FSD doesn't need to be perfect to avoid accidents. In fact, if it was too perfect I'm afraid I could become very inattentive at the wheel.
So, not Full Self Driving...
The regular, included autopilot works pretty well as a smarter cruise control; as long as you use it on a major highway in the daytime in good conditions it does a good job. And it's very useful in stop-and-go traffic on the highway.
But the FSD is crap.
You may have tried FSD beta. Different things entirely.
Fault, distracted, intoxicated human beings are allowed on public roads. The "legal limits" for blood alcohol levels aren't 0.00%.
The standards of human drivers is lower than that of FSD, and somehow you think that this is a good reason to justify lowering the standards of FSD.
That’s the only way it could be released in its current state.
Looking at it as anything other than a beta where you have to maintain control is misunderstanding what it is. Which you are clearly doing. It is absolutely expected to be worse than a drunk teenager in this stage.
To be fair, though, Tesla has no sensors other than cameras, and I believe the Mercedes has a half dozen or so, several radars and even a lidar.
Even a fruit-fly class AGI would do it.
There's no data that would support that anything than human level AGI is required to drive cars with how current infrastructure looks like.
No. You have to force flocking animals into extreme circumstances to have them start crashing,.
Sounds like the Tesla cannot manage that. Not even "bird brained".
Tesla software is used far, far, far more, in far, far more situation. Even compare those to things is kind of silly.
Its like comparing a system designed for only race tracks with Honda Civic. They are simply not designed for the same thing.
If Mercedes achieves L3 in all the places Tesla now allows AP (or FSD Beta) then that would prove the 'bet' on vision wrong.
Until then, nobody has proven anything.
Otherwise, not really.
This doesn't mitigate Tesla's gross ethical violations in letting this thing loose, it makes them worse. Tesla knew that cameras alone would be harder to make work than cameras plus radar plus lidar, and they shouldn't be lauded for attempting FSD with just cameras. It's an arbitrary constraint that they imposed on themselves that is putting people's lives in danger.
I am unsure what"allowed" means in that context. They just did it. The lawsuits are coming, surely?
Sometimes it is better to ask for forgiveness than for permission. This was an audacious example. Time will tell if there is anything that will stop them.
Edit (because HN throttling replies): > Mercedes is being responsible about their rollout
You can always succeed if you define success. If rolling out something that doesn't do much of anything is success, success? Oh well. Less than 20 people have died related to Autopilot in ~5 billion miles of travel. Zero deaths is unattainable and unrealistic, so "responsible" and "safe" is based on whatever narrative is being pushed. 43k people died in US car accidents last year, roughly half a million deaths in the entire time Tesla has offered some sort of driver assistance system called Autopilot.
How many deaths would be be okay with Mercedes' system where is would still be "responsible"? Because it won't be zero. Even assuming 100% penetration of Automatic Emergency Braking systems, it's only predicted to reduce fatalities by 13.2% (China specific study), and injuries by a similar number.
TLDR "Safe" is not zero deaths. It is a level of death we are comfortable with to continue to use vehicles to travel at scale.
https://www.tesladeaths.com/
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7037779/
I would not be surprised to see Mercedes beat Tesla to safe fully automated driving because they took it slow and steady instead of trying to rush it out to legacy vehicles that don't have the right sensor arrays.
Edit: It's weird to reply to a comment by editing your own. It feels like you're trying to preempt what your interlocutor is saying, rather than reply.
Do you have any information that it'll continue to be 40mph even where not required to be?
I wouldn't be surprised in the answer is yes, but I've been paying attention and haven't come across any yet.
If Tesla believes that their feature is safe, then let them take legal liability for it.
I don't have a choice not to use this feature. The person driving their Tesla death machine on the roads with me already made that decision for me.
uhh,, this is a thread where the federal gov't just forced Tesla to patch the feature, and so far we don't even know what the patch does...
Camera modules are cheap and available in huge quantities.
Then there's the issue of sensor fusion. Lidar requires sensor fusion because it can't see road lines, traffic signals, or signs. It also can't see lights on vehicles or pedestrians. So you still have to solve most of the computer vision issues and you have to build software that can reliably merge the data from both sensors. What if the sensors disagree? If you err on the side of caution and brake if either sensor detects an object, you get lots of false positives and phantom braking (increasing the chance of rear-end collisions). If you YOLO it, you might hit something. If you improve the software to the point that lidar and cameras never disagree, well then what do you need lidar for?
I think lidar will become more prevalent, and I wouldn't be surprised if Tesla added it to their vehicles in the future. But the primary sensor will always be cameras.
Neither is true, both are marketing.
> but I don't think there's any way to say that for sure
I mean yes there is. If you want to have an object test where you drop a car anywhere in the world and see if it can get somewhere else, then clearly one is more useful then the other.
In basketball they say 'the best ability is availability'. In terms of that AP is in a totally different dimension. AP has been driven for 100s of millions of miles by know, it must be a crazy high number by now. Mercedes L3 system has barley driven at all, its available in very few cars.
The only way you can reasonably compare the Mercedes L3 system is if you limit the comparison to the extremely limited cases where the Mercedes L3 system is available. If you compare them there, I would think they aren't that different.
> otherwise Tesla would release an L3 capable car too
No, because making something L3 in a extremely limited selection of places is simply not something Tesla is interested in doing. Doing so would be a lot of work that they simply don't consider worth doing when they are trying to solve the larger problem.
Also, they don't really have an inattentive to do that, so why should they.
Why? The design is locked in, the manufacturing doesn't change, you don't need the designers to do much... I would expect you need a handful of people, half of which are lawyers.
> Also, they don't really have an inattentive to do that, so why should they.
I'm pretty sure almost all Tesla owners would like the ability to stop looking at the road some of the time. That amount of customer satisfaction is not a motivation for the company?
Pf. Tensors beat sensors.
Just you wait for another garbanzillion miles driven. Then you'll see.
/s
https://electrek.co/2023/02/15/tesla-self-driving-hw4-comput...
So much for the "full self-driving" fantasy all those people paid for but will not get.
edit: looks like no.. the largest was 578607 cars... also by Tesla lol
Toyota, 2.9M: https://www.consumerreports.org/car-recalls-defects/toyota-r...
Ford, 21M: https://247wallst.com/autos/2021/07/24/this-is-the-largest-c...
Takata airbags, possibly 100M+: https://www.carvertical.com/blog/top-10-worst-car-recalls-in...
> Yet even after the company declared bankruptcy in 2017, the Takata recall kept on giving. 65-70 million vehicles with faulty Takata airbags were recalled by the end of 2019, with approx. 42 million still to be recalled.
Does anyone have insights on what QA looks like at Tesla for FSD work? Because all of these seem table-stakes before even thinking about releasing the BETA FSD.
There is even a former Tesla AI engineer that throws objects in front of the car on YouTube, as a demonstration.
The results are not glorious at all :| (trying to find the channel back if someone knows).
And random public tests too: https://www.youtube.com/watch?v=3mnG_Gbxf_w
This is a basic safety auto-braking. Just feels very wrong to even accept it goes into release.
The guy behind this is known to be untrustworthy, and many of the videos don't actually do what he claims. Notably he refused to release the videos that would prove his claims right.
The reality is that Tesla scores high on all the automated breaking test done by government. The driver however can override this, and that is exactly what is being done in this video.
The test done above has been replicated and the car does break in automated driving. And it gives loud warning to the driver even under normal driving operation and does emergency breaking.
Tesla assumes that if the driver hits the accelerator after the warning, the drive wants to accelerate based on the drivers judgment.
This is what this video shows, notably the person that made this video has refused to release prove that the car actually was in Autopilot, refused to provide evidence that the driver didn't hit the accelerator and refused to provide audio from inside the car so it can be verified that there was no warning sound.
In addition to that, the same person also claims things like 'millions of people will die if Autopilot isn't stopped' and that even under absurd assumptions is a legit insane thing to say.
So what is more likely, that Tesla did something incredibly illegal removing and incredibly important safety critical software that is standard in every Tesla OR that a competitor is doing a PR campaign where they deliberately set up a situation to film a video where they can create maximum damage to Tesla and then advocate for their own solution.
https://www.tesla.com/blog/model-y-earns-5-star-safety-ratin...
So the question is who do you believe, Euro NCAP or a competitor who made a sensationalist viral video.
Formal verification is obviously better if you can do it. But it's still really really difficult, and plenty of software simply can't be formally verified. Even in hardware where the problem is a lot easier we've only recently got the technology to formally verify a lot of things, and plenty of things are still out of reach.
And even if you do formally verify some software it doesn't guarantee it is free of bugs.
Whether it's a neural network inside or not is completely irrelevant. That's why it's called "black box".
You make a black box test on several thousands (sometimes only hundreds) patients, and if patients who received the drug perform better the patients who received the placebo, then the drug is usually accepted for commercialization.
Yet one isolated patient may be subject to several comorbidities, her environment could be weird, she could ingest other drugs (or coffee, OTC vitamins or even pomelo) without having declared it. In a recent past women were not part of clinical trials because being pregnant makes them very "non-standard'.
Secondly, even after the product hits the market the company is still responsible for tracking any possible adverse effects. They have a hotline where a patient or doctor can report it, and every single employee or contractor (including receptionists, cleaning staff, etc.) is taught to report such events through proper internal channels if they accidentaly learn about them.
I don't know where you get that, most clinical trials last 26 weeks, even in phase III.
and about "more thorough than you imagine" no, most are subcontracted to CROs and the way clinical trials are conducted is messy.
Below is story from the POV of a PI.
But similarly many patients complain about the way they are treated in visits and the lack of interest of the nurse/doctor who receive them.
https://milkyeggs.com/biology/why-are-clinical-trials-so-exp...
Clinical trials also have strict ethical oversight and are opt-in. If clinical trials were like Teslas, we'd yeet drugs into mailboxes and see what happened.
Yes. An army of Tesla owners perform the QA, in production.
But in all seriousness they do have some small team that validates then it goes to employees.
Time to stop driving. That is not normal
Tesla is not exactly in love with QA. Especially for FSD.
FSD is mainly 2 things: 1. (By far most important) shareholder value creating promise, that's been solved for 6 years according to their CEO. 2. Software engineering research project
What FSD is not is a safety critical systems (which it should be). They focus on cool ML stuff and getting features, with any disregard for how to design, build and test safety critical systems. Validation and QA is basically non-existent.
Based on there presentation, they for sure have a whole load of tests, many built directly from real world situation that the car has to handle. They simulate sensor input based on the simulation and check the car does the right thing.
They very likely have some internal test drivers and before the software goes public it goes to the cars of the engineers.
Those are just some of things we know about.
I have no source on their approach to testing safety critical systems, but we do know that they have a lot of software that has based all test by all the major governments. They are one of the few (or only) car maker fully compliant to a number of standards on automated breaking in the US. We have many real world example of videos where other cars would have killed somebody and the Tesla stopped based on image recognition.
So they do clearly have some idea of how to do this stuff.
So when making these claims I would like to know what they are based on. It might very well be true that their processes are insufficient but I would actual know some real data. Part of what a government could do, is forcing car maker to open their QA processes.
Or the government could (should) have its own open test suit that a car needs to be able to handle, but clearly we are not there yet.
I think you and I must've watched a different video.
1. I know people working at Tesla.
2. Much more important one - Elon's Twitter feed. They're doing last minute changes, and once it compiles and passes some automated tests, it's tested internally only over few days before it's released to the customers. Even if they had world class internal testing (they don't), for something having to work in such a diverse environment like self driving system without any geo-fencing, those timelines are all you need to know.
That's why I bought/will keep buying Toyota/Lexus.
https://www.euroncap.com/en/results/tesla/model+y/46618
Same for NHTSA:
https://www.nhtsa.gov/vehicle/2022/TESLA/MODEL%2525203/4%252...
Because of the FSD false promises, Tesla encourages dangerous behavior from drivers.
I don't want to be next to a Tesla driving in autonomous mode while it's driver at the wheel is not paying attention to me.
Yeah, they have QA. But for the problem they claim they’re solving (robotaxis) and speed of pushing stuff to customers (on the order of days) it vastly, vastly insufficient. And it lacks any safety lifecycle process regards - again, just look at the timelines. Even if you’re super efficient, you cannot possibly claim you can even such a basic things like proper change management (no, commit message isn’t that) or validation.
well, if you don't get the software pushed to the QA team (the customers), how else are they going to get it tested?
tesla does extensive, meaningful vetting of these updates. i'll let you do the research yourself so that maybe you can quit spreading misinformation
It is well known that issues happen despite extensive vetting. The presence of vetting does not exclude the absence of issues. For example, lets say there are 1000 defects in existence and your diagnostic finds 99% of them. So it finds 990 defects. So now there are 10 defects remaining that are not found. Next you have another detector that finds all defects. How many defect does it see? 10. It is usually going to see about ten defects. Your expectation given vetting is taking place should that be you will tend to observe about ten defects in this particular situation.
So lets say you are someone who is watching that second detector. You observe that you keep seeing defects. Probably, because of selection bias on results, you observe this with an alarming 100% rate of finding defects - the few times no defects happen it doesn't get shared for much the same reason we don't pay attention to drying paint. Can you therefore claim there is no diagnostic that is detecting and removing defects?
Well, not really, because you aren't observing the defects unconditional on the process that removed them. Basically, when you observe 10 defects, that doesn't mean there weren't 990 other defects that were removed. It just means you observed 10 defects.
So the actual evidence you need to use to decide whether there is a diagnostic taking place is evidence that is more strongly conditional on whether it is taking place. You need something that can observe whether the 990 are being caught.
In this case, we have video evidence of extensive QA processes. So that is much stronger than the evidence that defects show up, because defects show up both when we do have a vetting and we don't have a vetting. And each reasonable test case ought to be decreasing the chance of a particular defect that the test case is testing for.
For a much more thorough treatment on this subject check out:
https://www.lesswrong.com/tag/bayesian-probability#:~:text=B....
So that is basically why he disagrees with you.
Tesla definitely needs to root cause the defects that were found and make improvements on the existing vetting process but it is very obvious they do have these processes.
I do have high hopes that the work they've done on the FSD stack will make for a significant improvement to basic AP whenever they get merged (assuming it ever happens; it has been talked about for years). That'd be nice.
completely demonstrably false
> speed of pushing stuff to customers (on the order of days)
this is also false and doesn't happen
> you cannot possibly claim you can even such a basic things like proper change management (no, commit message isn’t that) or validation.
you know absolutely nothing about the internal timelines of developments and deployments at tesla and to suggest it's impossible without that knowledge is just dishonest
Head of AP, testified under oath, that they don't know what's Operational Design Domain. I'll just leave it at that.
> > speed of pushing stuff to customers (on the order of days) > this is also false and doesn't happen
Never ever Musk tweeted about .1 fixing some critical issues coming in next few days? I must live in a different timeline.
> > you cannot possibly claim you can even such a basic things like proper change management (no, commit message isn’t that) or validation. > you know absolutely nothing about the internal timelines of developments and deployments at tesla and to suggest it's impossible without that knowledge is just dishonest
Let's assume I have no internal information. If it looks like a duck, swims like a duck, and quacks like a duck, then it probably is a duck.
Lets say you have a baby which is being born. You tweet, birth in ten days! You can't then say, look: here is a tweet, it proves that baby development in tweeters actually has a several day lifecycle and moreover it proves that baby having mother's don't do proper pre-birth routines, because the tweet isn't the process that created the baby.
It is separate.
> If it looks like a duck, swims like a duck, and quacks like a duck, then it probably is a duck.
Right. So the fact that we have video evidence of internal processes, including QA processes, is much more like looking, swimming, and quacking like a duck much like having video evidence of a mother with a round belly for months before the tweet would be evidence that the babies don't take weeks to develop.
So when Elon has also tweeted that a launch was delayed, because of issues that were discovered - which does happen, as I'm sure you know if you follow his tweets as you imply - then that would be evidence congruent with the video evidence we have of QA processes existing within the company.
For example, did you predict, based on the speculation of Tesla being incompetent with regard to safety, that they have the lowest probability of injury scores of any car manufacturer? Because they do.
Did you predict, based on speculation about Elon Musk's incompetence in predicting that self-driving would happen, that there are millions of self-driving miles each quarter? Because there are.
Did you predict, based on speculation about Tesla incompetence in full self-driving, that the probability of accident per mile is lower rather than higher in cars that have self-driving capabilities? Because they do.
I know this sort of view is very controversial on Hacker News, but I still think it is worth stating, because I think people are actually advocating for policies which kill people because they don't actually know the data disagrees with their assumptions.
https://www.tesla.com/VehicleSafetyReport
I'm not sorry that annoys you, because it shouldn't.
Oh the data is great. I like the data. I'd take the data out to dinner. It's completely besides my point, and you continuing to be obtuse and rephrasing things this way, is not only a strawman, but it's rude.
> Your views probably do have evidence that supports them. I want to see the evidence you are using, because I think that is important.
Not every policy decision is driven by data. Some are driven by reasoning and sensibility, as well as deference to previous practices. So your whole data-driven shtick is just that... a shtick.
You claim that I said that we should focus on the good, but I didn't. I claimed that we should look at the data.
Now I feel as if you are trying to argue that looking at data is wrong because not all decisions should be made on the basis of the data. This seems inconsistent to me with your previous assertion that my ignoring data was bad, because now you argue against your own previous position.
That said, uh, datasets related to bayesian priors support your assertions about deference in decision making. So you could, if you cared to, support that claim with data. It would contradict your broader point that I should not want to have data, but you could support it with data and I would agree with you, because contrary to your assertion I was making an argument for evidence informed statements. Your inference about whether I think the evidence leans should not be taken as an argument that I believe my positions would always be what was reached by looking at the evidence, because I don't think that is true. I'm obviously going to be wrong often. Everyone is.
Unfortunately, I think you lie too much and shift your goalposts too much. So I'm not going to talk to you anymore.
I also caught you editing out what was an excessively rude comment. I'm gonna pass on further conversation, thanks.
No, you didn't. You said please ignore all the times when I'm wrong in favor of when I'm right. This is what you actually said.
You seem to be using language incorrectly. You seem to me to be confusing "said" with "meant" and in this case it seems that what you "said" was very different from what you "meant" so much so that I'm strongly getting the impression that you are lying to me, but if you aren't - then it is because confusion on the difference between said and meant.
Words have meanings. They have meanings independent of your own desires, so whatever you meant to say - it doesn't matter at all - that is not what you said. Please ignore is fairly characterized as a request for ignoring things, because it maps to something like it would be pleasing if there was ignorance. Notice it is you who is claiming of me that it would be pleasing to me if there was ignorance. I'm not making such a claim - you said this - maybe you didn't mean this, but you absolutely said it.
I'm not unfairly characterizing your words: this is the actual implication of the words you used, because it is the implication of the meaning of the words - maybe it isn't the meaning you desired, not what you meant, but it is the meaning of the words.
To kind of highlight how extreme what you claim you said is from what you actually said is, notice that you imply belief about me when talking about me requesting ignorance. Yet now, in your claim about what you actually said, you imply a position you hold: that data isn't dispositive. You don't even have the same reference in what you said versus what you now claim to have said. You did not say what you claimed to say. If you meant that, you should have said that, but these things are very very far diverged.
If you want to state, of your own belief, that data isn't dispositive, you should state that. Instead you consistently referred to a false reference to my own beliefs. Notice, even when you tried to correct my interpretation, you did not switch the reference class to your own belief, you said things like "your argument is ridiculous" which is still talking about me - not your own position that data is not dispositive. So the reference class which is not targeting me, but the general properties you now claim to have said, is not truly there.
> I also caught you editing out what was an excessively rude comment. I'm gonna pass on further conversation, thanks.
As you can imagine, with your confusion about meant versus said, I've been finding you to be lying about both your own views and mine. So if I seem a bit rude, it is because I'm kind of assuming you are smart enough to already realize all these things. I don't mean to assume you are so hostile as to know all this, but then pretend not to, but it is just one of the valid explanations for your behavior. I tried to edit my comment to remove my frustration and I'm sorry you had to see it like that.
> incredibly obtuse manner.
Which leads to this. You stated that you think I'm being obtuse, but my first assumption was that you weren't just directly lying about what I was saying. Your claim, interpreted in the way you said it, not the way you meant it, is a lie about my belief. So I tried to make your words have a meaning that would make them true, not false. I tried to be charitable, but was confused, because it really seemed like you were lying about my beliefs given your statement. This wasn't me being obtuse. This was me trying to be charitable, but being very confused, because interpreted according to what you actually said - not what you meant - you were strictly speaking stating falsehood about my beliefs.
Also, none of that is self driving. This data talks about AP, not FSD. FSD is also not self driving by any means (it's level 2 driver assist), but that's a detail at this point.
For example, elsewhere in this comment thread, someone threw out a random statistic of 400:1 as part of their argument, but this seems to me to be something like six orders of magnitude diverged from a data informed estimate.
To try and contextualize how big an error that is - it is like thinking that a house in the Bay Area has the same cost as a soft drink.
I think if we have to cite our data we are less likely to do that sort of error and more likely to catch it when it is done.
I definitely don't think FSD is magically safe. So if you think that is what I'm trying to say, please update your beliefs according to my correction that I do not believe this. I think anyone driving in FSD should remain vigilant, because it can make worse decisions than a human would.
1) those are statistics for the old version, the new version might be completely different. I've had enough one-line fixes break entire features I was not aware of that my view is that any change invalidates all the tests. (Including the tests that Tesla should have but doesn't) Now probably a given update does not cause changes outside its local area, but I can't rely on that until it's been tested.
2) the self-driving is presumably preferentially enabled for highway driving, which I assume has fewer accidents per mile than city driving, so comparing FSD miles to all miles is probably not statistically valid.
Just for context - I've been in a self-driving vehicle. Anecdotally, someone slammed on the breaks. The car stopped for me, but I was shocked: for hours before this the traffic hadn't changed, it was a cross country trip. I think I would have probably gotten in an accident there. Also anecdotally, there are times where I felt the car was not driving properly. So I took over. I think it could have gotten into an accident. Basically, for me, the best explanation I have for the data I've seen right now is that human + self-driving is currently better than human and currently better than self-driving. The interesting thing about this explanation is how well it tracks with other times where we've technology like this before. In chess playing for example, there was a period before complete AI supremacy (which is what we have now) where human + AI was better than AI.
I like the idea of being safe, so if the evidence goes the other way, advocating for only humans or only AI doing the driving, I want to follow that evidence. Right now I think it shows the mixed strategy is best and that is kind of nice to me because it implies that the policy that best collects data to reduce future accidents through learning happens to be the policy that is currently being used. I like that.
The probability of an accident for any driver assistance system will ALWAYS be lower than a human driver - but that doesn't mean the system is safe for use with the general public!
People like me are not advocating for "killing people" because we aren't looking at data - it's that no company has the right to make these tradeoffs without the permission and consent of the public.
Also if this was about safety and not just a bunch of dudes who think they are cool because their Tesla can kinda drive itself, why does "FSD" cost $16,000?
Totally we should be wary of a system that protects 400 and kills 1. Thank you for providing the numbers. It helps me show my point more clearly.
If you are driving on a road you encounter cars. Each car is a potential accident risk. You probably encounter a few hundred cars after ten or so miles. Not every car crash kills, but lets just assume they all do to make this simpler. For the stat you propose, you are talking about feeling uncomfortable with an accident per mile of something around the ballpark of ten miles.
Now lets look at the data. The data suggests the actual miles per accident is closer to 6,000,000 miles per accident. This is six orders of magnitude diverged from the number of miles per accident that you imply would make you feel uncomfortable.
Lets try shifting that around to a context people are more familiar with: a one dollar purchase would be a soft drink and a six million dollar purchase would be something like buying a house in the bay area. This is a pretty big difference I think. I feel very differently about buying a soft drink versus buying a house in the Bay Area. If someone told me they felt that buying a house was cheap, then gave a proposed price for the house that was more comparable to the cost of buying a soft drink, I might suspect they should check the dataset to get a better estimate of the housing prices, because it might give them a more reasonable estimate.
So I very strongly feel we should cite the numbers we use. For example, I feel like you should really try and back up the use of the 400 to 1 number so I understand why you feel that is a reasonable number, because I do not feel that it is a reasonable number.
> Also if this was about safety and not just a bunch of dudes who think they are cool because their Tesla can kinda drive itself, why does "FSD" cost $16,000?
Uh, we are a on venture capitalist adjacent forum. You obviously know. But... well, the price of FSD is tuned to ensure the company is profitable despite the expense of creating it as is common in capitalist economies with healthy companies seeking to make a profit in exchange for providing value. It is actually pretty common for high effort value creation, like creation of a self-driving car or the performance of surgery, for the prices to be higher.
If you are advocating against a system that protects 400 people and kills one, you are advocating for killing people.
(Is Autopilot still limited to divided, limited access highways? Those are significantly safer than other roadways.)
No. Was it ever? All you need is a piece of road that has something which appears to be lane lines. The road to my house is usable despite having no actual paint striping because it happens to have a crack that runs fairly straight up one side and was filled with tar. So the camera thinks it's a lane line. Ta-da!
The thing is we often have discussions about this stuff and I'm trying to advocate for citing datasets to more tightly correlate our words with the evidence that our words correspond to. I'm not trying to say this version shouldn't have been recalled for example, but that I think we should be close to evidence.
In the case of auto-pilot, it was the case that people made the same arguments that are now being made against FSD. I think that makes it somewhat relevant to the discussion, because people previously also made the same claims about safety, but now that we have the data, we can see those claims were wrong. I believe these sort of generalizations, though inaccurate, can help us to make more informed decisions, but I'm not really confident in any beliefs that are made at this greater decision from direct data.
So I think anyone who can provide datasets which correspond with FSD performance rather than autopilot performance ought to do so. That would be really great data to reflect on.
The thing I'm worried about is that no data at all is backing the conjectures - which, given that I sometimes see estimates that I calculate to be many orders of magnitude away from data informed estimates - seems to be the case on Hacker News at least some of the time.
They have a set of regression tests they run on new code updates either by feeding in real world data and ensuring the code outputs the expected result, or running the code in simulation.
It does seem worrying that they would miss things like this.
Here’s a talk from Karpathy explaining the system in 2021:
https://youtu.be/aNVbp0WKYzY
Though I don’t recall if he explains the regression testing in this talk, there’s a few good ones on YouTube.
I used to think that fact was going to delay self-driving cars by a decade or more, because of the potential bad press involved in AI-caused accidents, but then along comes Tesla and enables the damn thing as a beta. I mean...good for them, but I've always wondered if it was going to last.
I've been using it pretty consistently for a few months now (albeit with my foot near the brake at all times). I haven't experienced any of the above. Worst thing I've seen is the car slamming on the breaks on the freeway for...some reason? There was a pile-up in a tunnel caused by exactly that a month or so ago, so I've been careful not to use FSD when I'm being tailgated, or in dense traffic.
There are only like 16 million intersections in the US. Why not test them all?
Tesla does however collect data on edge cases and then train their system to respond correctly. They can for example trail a collection network to identify things that might be obscured stop signs, then have the fleet collect a whole bunch of examples, hand label those samples, and roll this new data in to the training system. This is explicitly how they handle edge cases.
They can also create a new feature or network and roll it out in “shadow mode” where it is running but has no influence on the car, and then they can observe how these systems are behaving in the real world.
The real issue I guess is when they release a new feature without trialing it in shadow mode, or if they have gaps in their testing and validation system.
As for their QA process, in 2018 they had a braking distance problem on the Model 3. They learned of it, implemented a change that alters the safety critical operation of the brakes, then pushed it to production to all Model 3s without doing any rollout testing in less than a week [1]. So, their QA process is probably: compiles, run a few times on the nearby streets (I am pretty sure they do not own a test track as I have never seen a picture of tricked out Teslas doing testing runs at any of their facilities), ship it.
[1] https://www.consumerreports.org/car-safety/tesla-model-3-get...
1. https://finance.yahoo.com/news/upcoming-tesla-software-2020-...
I have a winding road near me with a speed limit of 35 mph, but 15 mph on certain curves as indicated by a speed limit sign. It ignores those speed limit signs and will attempt to make the turns at 35 mph resulting in it wildly swerving into the other lane and around a blind turn with maybe 30 feet of visibility. It has also attempted to do it so poorly that it would have driven across the lane and then over the cliff without immediate intervention.
Unsupported claims by a manufacturer that compulsively lies about the capabilities of their products except when directly called on it are the opposite of compelling evidence.
Teslas definitely read speed limit signs. I've had mine correctly detect and follow speed limits in areas without connectivity or map data. It also follows speed limits on private drives (if there is a sign) and obeys temporary speed limit signs that have been put up in construction zones.
Frankly, attempting to deflect by arguing that it is okay to release a defective safety critical product to unsuspecting consumers just because nobody else is willing to offer a similar product because they have some moral integrity is a stance that makes the executives at Ford presiding over the Pinto look like angels in comparison. All the Ford executives did was cover it up to avoid having to pay to fix it. At least they did not intentionally release a known defective and dangerous product just to recognize some revenue.
Cognitive dissonance at it's finest.
That's what makes it unfinished...
It's never passed the 'drive from new York to LA with nobody touching the controls' test...
It's technically a recall, but it's fixed with an OTA update. But the fact that any "defect" that can be fixed with an OTA update is called a "recall" is confusing to consumers and contributes to media sensationalism.
There absolutely needs to be a process overseen by regulators for car manufacturers to address software-based defects, but the name would benefit from being changed to reflect that it can be done without physically recalling the vehicle to the dealer.
Its crazy that Tesla has been doing OTA for 10+ years and today many cars are released that are not capable of being upgraded.
And even those few cars that do support OTA only support it for a very limited amount of system. Often they still need to go to the shops because lots of software lives on chips and sub-components that can't be upgraded.
OTA updates carry a perverse set of incentives. Look at the gaming industry. They went from putting out rock solid games because of necessity (the reality of publishing physical cartridges and CD-ROMs without network updates) to the absolute dogshit of No Man's Sky and Cyberpunk 2077. Gamers have effectively become an extension of QA to the point that some game devs simply stop doing QA at all. Which we are seeing clearly with the "beta" version of self-driving software.
To make matters worse, firmware devs are on the lower totem pole of the developer hierarchy. They live more on the cost center side than the profit center side (think airbag control vs. the guy that did the whoopee cushion sounds). The quality of firmware is already incredibly poor across the range of consumer devices. OTA incentivizes corporations to release software earlier than they currently do knowing that they can always fix it later if necessary.
I do share your concern, but I still likely prefer it to that to software that can never be updated at all.
Would it be sensationalism if the same recall happened to cars not capable of the OTA update?
Because I think there should be media "sensationalism" about these types of issues, regardless of whether they can be fixed with physical or OTA repairs.
> But the fact that any "defect" that can be fixed with an OTA update is called a "recall" is confusing to consumers
What is confusing about this to you?
> contributes to media sensationalism
Why do you think this recall is more sensationalistic than many of the other recalls issued recently? Automotive recalls often address serious issues with cars. "Fix this or you could die" is a common enough theme when, you know, if you don't fix it you could die.
> but the name would benefit from being changed to reflect that it can be done without physically recalling the vehicle to the dealer.
Why? Information related to how, when, and where the recall can be addressed is contained in the text of the recall notice, same as it ever was.
Edit: Dang, you're all right, they could eat this and still be alive.
My whole sentiment comes from FSD being their big shot (and everything that comes with that, like the robotaxis and whatnot). Without FSD, they're "just another car company" and the market is already thriving with good alternatives (Audi's EVs are jaw dropping, at least for me). Excited to see what they announce on March 1st, though.
If so, that's pretty crazy and people will die because of this decision.
The usual comparison is vs. all cars on the road, but Teslas are comparatively new cars with a lot of recent and expensive safety features not available in older cars. They’re also likely to be maintained better. On top, they’re also a driven by a different demographic which skews accident statistics.
Teslas autopilot can only be enabled in comparatively safe and simple circumstances, yet the comparison is made against all driver in all situations. When autopilot detects a situation it can’t handle, it turns off and hands over to the human who gets a few seconds warning. Human drivers can’t just punt the issue and then crash.
Tesla FSD may be safer than Human drivers for the limited set of environments where you can use it, but last time I checked, the numbers that Tesla published are useless to demonstrate that.
This is incorrect. FSD can be enabled everywhere from dirt road to parking lots, even highways and dense urban environments like NYC. There is no geofence on FSD, you can turn it on anywhere the car can see drivable space.
It enables in snow and rain yes. The only time it refuses to engage is in extreme weather that obstructs the cameras.
It gets confused a lot in heavy traffic but it will attempt anything.
The primary one is:
1) FSD turns itself off whenever things get too difficult or complex for it. Human drivers don’t get to do this. Recent crowdsourced data suggests a disengagment every 6 miles of driving: https://twitter.com/TaylorOgan/status/1602774341602639872
If you eliminate all the difficult bits of driving I bet you could eliminate a lot of human driver crashes from your statistics too!
A secondary issue, but relevant if you care about honesty in statistical comparisons is
2) The average vehicle & driver in the non-Tesla population is completely different. Unless you correct for these differences somehow, any comparison you might make is suspect.
* they are not the first (or in the first ~50 or so) to get there
* they have a lot of money and a lot of income on Earth
This is kind of a glory-seeking thing for people who haven't made a name for themselves otherwise. Rich people would want to have a return flight. That is not a recipe for making a ton of money.
If you are already rich on Earth, you don't need to take risks that insane to try. In other words, nobody who can afford a ticket to Mars will pay for one.
There are no rules on mars, except the ones that the people who could simply push you out the airlock make. Look how elon treats his workers and you will understand how a mars colony with his backing will look, except even worse.
Can I nominate someone?
I suppose I think that's better in the long run, better for society in addition to being better for the planet. I know it's not a popular take.
No, it's just gloating, right? We get it, he hates Elon, and when you hate someone you'll throw whatever dart you've got on hand at them.
To answer your question, though, my opinion on his Mars project would change completely if it turned out that cave rescue guy actually was a pedo.
Is that not the case?
Hahaha talk about a loaded question. You see enthusiasm I see manic grift.
It's an interesting watch (especially with the context of labor relations in mind).
(This is why the engines for Starship are designed to use methane).
https://en.m.wikipedia.org/wiki/Mars_Direct
The raw energy required to make that fuel, using ANY conceivable process, is extreme though. What energy solution has been proposed? How much solar acreage do you need? How will you keep the dust off the panels? Will there be any option other than sending panels from earth?
All the needed prereqs have zero practical knowledge with them. Nobody has even had the chance to work out kinks yet. It doesn't matter how much elon wants to do something, new shit takes a lot of time, money, and effort to shakedown.
Assuming there are people involved, give them a broom. ;)
No one, because it hasn't been required or even feasible yet. We don't have the ability to send that sort of payload to Mars yet.
> How will you keep the dust off the panels?
I hear Tesla is working on a humanoid robot.
It's mostly made of iron and steel, that we have here on earth in abundance.
One day, the economics of return trips will work out. But to begin with at least, all trips will be one-way.
It's highly unlikely that trips with humans on board would be one-way.
https://idlewords.com/2023/1/why_not_mars.htm
I would love to go there, even if it meant I died the second I step foot on it. Getting back would be the last thing on my mind.
Death is guaranteed either way.
I'm not into space exploration but sometimes I look up at the moon and it blows my mind that we actually went there. Almost feels impossible even. Mars is that x100.
I personally wouldn't want to die on Mars but I understand the appeal.
Because there is so much to do on earth? Literally everyone is here. What other answer does anyone else need? All this Mars fandom sounds to me like clinically depressed sci-fi.
Seems like a particularly over-the-top suicide method.
Bullshit. Colonization _might_ be a side effect. Bilking various governments out of subsidies and fat contracts is the goal. Just like every other single Musk venture, it is based entirely on feeding at the trough of federal largesse.
What was that called again?
/puts on sunglasses preemptively/
https://www.mcsweeneys.net/articles/my-name-is-elon-musk-and...
The problem is Tesla owners repeatedly see "recall" to mean "software update", so this might lead to a lot of confusion if a physical recall is actually required in the future.
https://www.nhtsa.gov/recalls
The problem here lies in having a manufacturer shipping an unfinished product then relying on an endless stream of recalls to finish developing your vehicle.
These are supposed to be exceptional events. If they've become so frequent you're ignoring them, don't shoot the messenger.
I don't think "Use words that put a more positive spin for Tesla PR" is something regulators should be working faster on.
The scenario in this article is an over-the-air update.
Normally, it happens with an oil change.
If it's possible to buy a software license online and then return it online after deciding you no longer want it (i.e., a non-physical return), then it stands to reason that Tesla can request that you return the defective software OTA and receive replacement software OTA, and that would be a recall. The fact that you are forced into returning the defective software by virtue of not having the opportunity to block the return request is a fairly minor detail.
If Tesla had willingly walked this back, it'd be a software update, or a beta delay, or whatever they wanted to call it.
What laypeople don't realize is that this is being called a recall because the NHTSA pushed them into it: https://static.nhtsa.gov/odi/rcl/2023/RCLRPT-23V085-3451.PDF
-
FSD has been a cartoonish display for the better part of a year now, it wasn't until last month that the NHTSA actually pushed on them to do a recall, and from there they "disagreed with the NHTSA but submitted a recall"... which is code for "submitted the recall NHTSA forced on them to submit"
Elon knows better, but he knows he can weaponize people's lack of familiarity with this space and inspire outrage at the "big bad overreaching government"
Teslas are basically digitally tethered to the dealer, so they can be "recalled" anytime (without your approval, fwiw), but it doesn't make the word not apply.
I'm looking at a piece of mail right now that's an official recall notice for a different make/model of car I own. The issue? Improperly adjusted steering components. The company is offering to make the correct adjustments for free. Nothing is being replaced.
Whether the recall is to replace a poorly designed physical component, or to make an adjustment, or to apply a software update doesn't make a difference to regulators.
A recall is a legal process that's designed to encourage manufacturers to fix safety issues while also limiting their liability. Companies avoid recalls if they can because it's costly, time consuming, and isn't good PR. But it's worth it if the issue is bad enough that it risks a class action lawsuit, or individual lawsuits, and most desirable when someone like the US government is demanding a recall or risk legal consequences.
When a company issues a recall, they make their best effort to notify consumers of the issue, provide with clear descriptions of how consumers can have the issue fixed, and make it clear that it will be paid for by the manufacturer. In return, the manufacturer is granted legal protections that drop their risk of being sued to nearly zero.
Yeah, it requires physically taking the car to a mechanic or dealer who does this. Very different from using the software update button on the car touchscreen.
> Improperly adjusted steering components. The company is offering to make the correct adjustments for free. Nothing is being replaced.
This is clearly a recall, because it requires taking the vehicle to a mechanic or dealer.
Nope.
If you insist on believing this, I'd recommend reading the document found here:
https://www.nhtsa.gov/sites/nhtsa.gov/files/documents/14218-...
Again, the nature of the issue a recall addresses is wholly independent of how that issue is remedied. Why? Because a recall is a legal process that, by design, is meant to accomplish one thing: motivating a company to correct an issue that the governing authority considered important enough to correct.
If you choose to believe otherwise, I doubt it matters in the grand scheme of things.
Whether it's a "good" word or "bad" word is irrelevant. It describes a very rigid and official legal process.
If I had chosen not to actually apply it, the dealer would have been expected to do so the next time my car was in for service.
The remedy OTA software update will improve how FSD Beta negotiates certain driving maneuvers during the conditions described above, whereas a software release without the remedy does not contain the improvements.
Other cars get "recalls" all the time that amount to updating the software in the ECU or TCU. Tesla is simply being treated like everyone else.
Hell, even in food, a "recall" usually means "throw it away" not return it.
I had a VW car that was "recalled" shortly following the emissions scandal. The dealership asked me to come in for a free software update related to emissions. So you can say it's a "recall" to the lawmakers but call it a "free software update" to the user.
Recalling the hardware is a drastically more difficult request to impose on customers and financially/logistically for the car maker.
That's a disporportate response just to highlight importance of an OTA update.
you can, in fact, 'recall' software. this is semantically accurate description of what is happening.
What do you think “FSD recall” or “software recall” means
Tesla released software that can kill people. It must do a FSD recall
This is not the same as Apple doing an OS update, not mandated by a regulatory body because it could kill someone
I was hit by a driver not paying attention. I had to get two surgeries, was not allowed to stand for 9 months, spent another year just rehabbing my nervous system and learning to walk. I still can’t jump or run
Tesla released software it’s marketing as “full self driving” that is not even Level 2 Alpha! Full Self Driving implies a level 5 system.
Tesla’s software has disclaimers like “may not behave as expected in sunlight”. Even its AutoPilot has these disclaimers.
It constantly claims FSD makes drivers safer. Yet it’s non transparent with its data, the data & comparisons it does release is completely misleading to the point of fraud, comparing apples to oranges. The cars it’s released on public roads to untrained drivers runs through intersections (it’s all over YouTube) and fails all kinds of basic driving tests. Tesla accepts ZERO responsibility if someone is killed while FSD is activated… that’s how little confidence it has in its own product
This is a product that can maim and kill humans, ruin people’s entire lives… and your response is incoherent mumbling about adjectives ?
Any person who is majorly confused by what a “software recall” is… or can’t figure out what this FSD recall means for them, shouldn’t be beta testing a 2-ton machine on public roads. They shouldn’t be driving period.
And the distinction matters to consumers because...?
A component is faulty. It needs to be fixed. Whether or not you have to drive to a dealership, if it's OTA, if someone at a dealership needs to plug a specialized device to your car's OBD port, or the car is unfixable and needs to be melted to slag and you get a new one doesn't really matter. There's an issue, it is a safety issue, and it needs to be fixed.
How efficient the process can be it's another matter entirely. That's up to the manufacturers.
As a car owner, those scenarios are drastically different to me. I have a hard time imagining anyone saying "It doesn't really matter to me if my car receives an OTA update or if I need to drive 2 hours to a dealership or if my car is melted to slag."
A recall should unambiguously mean that some action from the owner is required to resolve the issue (e.g. taking it to a dealer to get a software update installed.)
If no action is required (other than caution / not using the product feature), we should use some other term such as "safety advisory" to avoid ambiguity around critical safety information.
Should, perhaps. But recall has a meaning with legal implications, so it matters. A recall requires the fix status to be tracked and reported for example, whereas a random OTA updated does not.
Because in one I have to book a time with a dealer and take half a day off of work and in the other I have to do... nothing and it will just fix itself.
With a Recall in the normal sense, isn't there a record that the car has been updated? How is this done if the car is kept fully available to the user?
"Critical safety defect" would be better.
A recall is a legal process only. Whether the recall repairs, replaces, adjusts something doesn't matter. Whether a fix is applied as software, or labor, or replacement parts doesn't matter. Whether a customer needs to do something or not doesn't matter.
A recall simply says: as a manufacturer, working with government authorities, while taking specific prescribed steps to communicate and correct an issue at the cost of the manufacturer, the manufacturer is then immune from lawsuits that could arise were they to ignore the issue.
And that "bugs" involve accidents.
> At a computer expo, Bill Gates compared the computer industry with the auto industry and stated, 'If Ford had kept up with technology like the computer industry has, we would all be driving $25 cars that got 1,000 miles to the gallon.'
> In response to Bill's comments, Ford issued a press release stating:
> “Sure, but would you want to drive a car that for no reason whatsoever would crash twice a day?”
https://www.chevyhhr.net/forums/lounge-10/microsoft-vs-ford-...
Your argument seems to be about protecting Tesla's reputation
https://www.cpsc.gov/Recalls/2023/Samsung-Recalls-Top-Load-W...
That's where the word came from, that's true. But a legal framework has built up around the concept over the decades which isn't dependent on you having to drive the car to the dealer, all of which still applies, so the word is still used.
A defect that is subject to a recall, for example, is tracked as part of the car history. When considering buying a used car, you can see whether that repair has been made yet or not. The means of how that repair is delivered is inconsequential.
True, but by announcing things in this fashion it is making Tesla look bad. Regulations really need to be updated so that car makers can hide this type of problem from customers as easily as possible. Especially when it comes to Tesla, regulators really need to bend over backwards to prevent articles from being written that could be interpreted in a negative way.
Or are people concerned about the word "recall" for a different reason?
What. No! At an absolute minimum, I want to be aware of any changes to the thing in my life most likely to kill me. Maybe we could use a better term like "Software Fuckup Remediation" or "Holy Shit How Did We Not Get The Brakes Right".
Maybe Tesla should stop doing things that result in it receiving poor publicity? just a thought
Why should a companies actions be dictated by PR and media news cycles?
They rolled out software with critical safety issues. They have to be called out.
Are you saying it's a critical safety issue to depend on a human driver? The same as in every vehicle on the road?
I almost didn't catch the sarcasm of this comment, but there are other comments in this thread that are saying basically the same thing, but actually meaning it. It defies logic.
People seem to think that the government is being mean, and singling out Tesla, and being nasty using the word "recall." A recall is a legal process. The word means something very specific, and when a company issues a recall, they do so because they don't want to be sued.
It's almost like complaining about the word "divorce" or "audit" or "deposition" or other similar words that describe a legal process. The words used mean something specific. Tesla is conducting a legal process, and there's a very specific word for that process, and it means something. It's a recall.
Some things are absolutely safety relevant. But no one cares.
It should absolutely be tracked and publicized. But it's fundamentally different than "this car is fundamentally broken and you have to take it back to the manufacturer"
"Recall" with cars is a legal term that means something specific.
> But it's fundamentally different than "this car is fundamentally broken and you have to take it back to the manufacturer"
That's not a criteria for a recall. A recall can be for a small thing completely unrelated to driving safety. Someone else mentioned having a recall to replace a sticker. My most recent recall was to replace the trunk lifts.
"Recall" for a car just means a specific procedure now has to be followed, tracked and reported. It says nothing about the safety criticality of the fix.
>> "...The FSD Beta system may cause crashes by allowing the affected vehicles to: “Act unsafe around intersections, such as traveling straight through an intersection while in a turn-only lane, entering a stop sign-controlled intersection without coming to a complete stop, or proceeding into an intersection during a steady yellow traffic signal without due caution,” according to the notice on the website of the National Highway Traffic Safety Administration..."
No, that's not correct. Whether it can kill people or not is orthogonal to whether it's a true physical recall of the car or a software update.
There are really two issues here. The FSD and the OTA updates. Let's not throw out the baby with the bathwater and blame OTA updates just because Tesla's FSD software is bad. The OTA updates do provide an avenue to make cars much safer by reducing the friction for these type of safety fixes.
True, but let us also acknowledge the immense systems safety downsides of OTA updates given the lack of effective automotive regulation in the US (and to varying degrees globally).
OTA updates can also be utilized to hide safety-critical system defects that did exist on a fleet for a time.
Also, the availability of OTA update machinery might cause internal validation processes to be watered down (for cost and time-to-market reasons) because there is an understanding that defects can always be fixed relatively seamlessly after the vehicle has been delivered.
These are serious issues and are entirely flying under the radar.
And this is why US automotive regulators need to start robustly scrutinizing internal processes at automakers, instead of arbitrary endpoints.
The US automotive regulatory system largely revolves around an "Honor Code" with automakers - and that is clearly problematic when dealing with opaque, "software-defined" vehicles that leave no physical evidence of a prior defect that may have caused death or injury in some impacted vehicles before an OTA update was pushed to the fleet.
EDIT: Fixed some minor spelling/word selection errors.
Tesla didn't choose the word "recall." The legal process known as "recall" chose the word. It's not like people at Tesla debated over whether or not to call it a "recall" instead of a "software update."
If Tesla had it their way, they'd have quietly slipped it into any other regular software update alongside updates to the stupid farting app, if they cared to fix it at all.
When a company issues a recall, it's because there's pressure from regulators, or investors, or both, and/or a risk of class action lawsuits and fines. Using the word "recall" isn't a preference or even a synonym. It's a legal move meant to protect them.
If Tesla gets sued over a flaw, "we issued a software update" isn't legally defensible. "We cooperated with official government bodies to conduct a recall," does because a recall describes an official process that requires manufacturers do very specific things in specific ways as prescribed by law. In exchange, manufacturers are legally protected (usually) from lawsuits related to that flaw.
This should be labeled "holy shit need to fix this now, people could die"
A "recall" is just a public record that a safety-related defect existed, the products impacted and what the manufacturer performed in terms of a corrective action.
Additionally, I believe that the possibility exists that Tesla must update the vehicle software at a service center due to configuration issues. Only a small number of vehicles may require that type of corrective action, but the possibility exists.
Historically, there exist product recalls (especially outside of the automotive domain) where the product in question does not have to be returned (replacement parts are shipped to the impacted customers, for example).
https://www.nhtsa.gov/sites/nhtsa.gov/files/documents/14218-...
https://www.law.cornell.edu/cfr/text/49/573.6
(Except for tires.)
You (and the parent comment) are correct.
My comment was not intended to argue that a recall prescribed a particular corrective action.
ah, a made up term in order to justify your point. how convenient.
I hope by participating in this thread you're aware by now but just to be clear there is no "physical" recall necessary. The recall is about documentation, customer awareness, and fixing the problem. "Physical recall" is meaningless and unimportant, it's not what "recall" means at all.
The headline of that many cars being "recalled" is of course nice clickbait. Much better than "cars to receive minor software update that fixes an issue that isn't actually that much of an issue in the real world so far". One is outrage-ism fueled advertising and the other is a bit of a non event. It's like your laptop receiving some security related update. Happens a lot. Is that a recall of your laptop or just an annoying unscheduled coffee break?
...and that's one reason why I would never purchase a Tesla.
I recall getting a recall notice from GM that included "until repaired, remove key from keychain".
[0] https://en.wikipedia.org/wiki/General_Motors_ignition_switch...
A recall means exactly what it means. The manufacturer is responsible for a fix. If for whatever reason they can't push an OTA update to your car, Tesla is still responsible for sending you a postcard in the mail and calling you every 6 months telling you to bring it in for service until they have reasonable evidence the car is no longer on the road.
The official meaning of a recall is providing a record of a defect, informing the public that the product is defective, and making the manufacturer financially liable for either remediating the defect or providing a refund. However, the colloquial definition of a “recall” now means a product must be physically returned.
To better represent the nature of a “recall” they should instead call it something like “notice of defect”. In the case of safety critical problems like here they should use a term like “notice of life-endangering defect” to properly inform the consumers that the defect is actively harmful instead of merely being a failure to perform as advertised.
tl;dr They should change the terminology from “recall” to “Notice of Life-Endangering Defect”
https://www.nhtsa.gov/sites/nhtsa.gov/files/documents/14218-...
A voluntary recall is easier and cheaper for all involved.
"Voluntary recall" in this case means that Tesla did not choose to take the hard route where there's a court order for a mandatory recall. Few manufacturers fight that, because customers then get letters from the Government telling them their product is defective and that it should be returned for repair or replacement.
Somebody in the swallowable magnet toy business fought this all the way years ago.[1] They lost. It's still a problem.
[1] https://www.cpsc.gov/Safety-Education/Safety-Education-Cente...
§ 30120(a)(1)(A)(iii) by refunding the purchase price, less a reasonable allowance for depreciation.
OTA updates are cool but they more complexity to the car. I like my cars to be simple and reliable instead.
Customers aren't dealing with inconvenience of having to bring their vehicle in / not having it for a while.
And Tesla is not incurring the cost of physically handling and fixing 360k cars.
So while it's technically a recall, the impact on the consumer and manufacturer is very different than what the word brings to mind.
Just to reiterate; it’s a software update. You can FUD as much as you want.
Jokes aside, it's gotta be damn tough to QA a system like this in any sort of way that resembles full coverage. Can't even really define what full test coverage means.
A autonomous Tesla driving into a group of people or swerving into oncoming traffic is potentially killing other people.
This morning I tried to turn it on, and the car immediately veered left into the oncoming lane on a straight, 2 lane road. Fortunately, there were no other vehicles nearby.
I immediately turned it off in the settings, and have no intention of re-enabling.
TL;DR: Controller drift class action lawsuit filed by parents thrown out, because their children were the actual "affected class". Refiled with the children as the class and thrown out again, this time because of an arbitration clause in the EULA their parents would have had to agree with.
Lawyers and EULAs are crazy.
There might be precedent to even removing the feature since FSD almost is vaporware and has no release date. https://www.cnet.com/tech/gaming/ps3-other-os-settlement-cla...
You have to basically drive like a grandpa for a few months to even be eligible. They give you a driving score, and if you take all the fun out of driving a Tesla, then you might become eligible for FSD Beta.
I spent months trying, and never got my driving score to the point of qualifying for FSD Beta. I think you need to have of a score of 98 or 99 (and I was in the 70s).
[1] The Beta is free (or rather, only available) if you have regular FSD. Regular FSD costs $15,000.
Regular FSD, otoh, is really not that impressive. Especially in comparison to Enhanced Autopilot. The extra value add is minimal.
Enhanced Autopilot already has all the gimmicky features you might want to use to show off to people (like Smart Summon, Autopark, etc), and it only costs $6,000.
Yes, you have to pay $15k just to apply to the beta program, and you still may not get accepted into the program.
The safety score check stuff is largely gone away today - anyone who pays 200 bucks can click the beta opt in and get it almost straight away now, there is ~zero risk of not getting the beta if you really want it, live in US or Canada, and are prepared to pay.
> https://www.tesla.com/support/full-self-driving-subscription...
Does this recall mean that FSD Beta won't be as widely & publicly available to anyone with FSD anymore?
FWIW, NHTSA "recalls" are often OTA software updates nowadays rather than something the vehicle or feature has to be taken off road for to fix or update. The NHTSA legislation from the 60s was drafted when cars didn't have software and any fix/"recall" likely required "recalling" the car to a shop for a mechanic to perform the change.
> https://repository.law.umich.edu/mtlr/vol28/iss1/5/
That's a significant amount of cash for features that I would likely never use.
Sadly, it does shut itself off as soon as you're off a highway however. (That's where FSD would hypothetically come in, once the beta is ready, with "Autosteer on city streets").
In terms of value for money:
Overall, Enhanced Autopilot is worth at least $4,500 methinks.Throw in $1,500 as a profit margin (or Elon tax), so he can burn some dinosaurs for his private jet flights, the $6,000 Enhanced Autopilot price point makes sense.
FSD, otoh, is absolutely not worth it.
You shouldn't be daydreaming on Navigate on Autopilot. Its only Level 2, you're supposed to be ready take the wheel in a second or two. You're supposed to still be actively paying attention to the road, constantly.
What does that work out to, in terms of dollars per lane change for the duration of ownership? Would you feel the same if you were feeding dollar bills into a feeder each time you changed lanes? Quarters?
Auto lane change is the only feature I value out of EAP / FSD subscription and I can't justify $200/month because it works out to multiple dollars per lane change.
How often do you use this and how well does it work?
It doesn't because Tesla's FSD model is just a rules engine with an RGB camera. There's not "purpose" to any hallucination. It would just be a misread of sensors and input.
Tesla's FSD just doesn't work. The model is not sentient. It's not even a Transformer (in both the machine learning and Hasbro sense).
I dont think its true? They use convolutional networks for image recognition and those things can certainly halucinate - e.g. detecting things that are not there.
https://twitter.com/maxhkw/status/1373063086282739715?s=52&t...
But of course, as you say, that system does not consume actual raw (camera) sensor data, instead there are lots of intermediate networks that turn the camera images (and other sensors) into red lights, lane curvatures, objects, ... and those are all very vulnerable to making up things that aren't there or not seeing what is plain to see, with no one quite able to explain why.
Or maybe when I get a bug report today I’m going to tell them the software is just hallucinating.
This is the kind of feature I will use when the car and software in question has been battle tested for *years* with objectively excellent results.
I understand the simpler lane keeping system is okay, but I don’t want to trust any system like this from Tesla given their track record with FSD.
Before this morning, I would be in exactly your shoes in terms of "this is pretty cool, I'm going to keep using it"
A bit of a sensational title compared to what this really is.
The people who keep making this "Recall means go back to the dealer" claim is simply down to them never paying attention to all the recalls in the world that don't make it to their mailbox like car recalls do.
(maybe people just don't think through ramifications)
Would you like to participate in the beta program, to test stoping bullets being shot at you?
"could kill you if gone wrong" is nearly every choice in life. The relevant metric is risk of death or other bad outcomes.
Flying in a commercial airliner could kill you if things go wrong. Turns out it's safer than driving the same distance.
Eating food could kill you if things go wrong (severe allergy, choking, poisoning) yet it's far preferable to not eating food.
Similarly, Tesla's ADAS could kill you if things go wrong but the accident rate (as of 2022) is 8x lower than the average car on the road, and 2.5x lower than a Tesla without ADAS engaged.
Don't let the perfect be the enemy of the good.
I realize that not everybody is in agreement, but I personally use the FSD beta while remaining fully in control of the vehicle. I steer with it, I watch the road conditions, I check my blind spots when it is changing lanes, I hit the accelerator if it is slowing unexpectedly, I hit the brakes if it is not...
You know, basically behaving exactly as the terms you have to agree to in order to use the FSD beta say you are going to behave.
When I look at the wreck in the tunnel (in San Francisco?) a few months ago, my first thought is: how did the driver allow that car to come to a full stop in that situation? Seriously, you are on a highway and your car pulls over half way out of the lane and gradually slows to a complete stop. Even if you were on your phone, you'd feel the unexpected deceleration, a quick glance would show that there was no obstruction, and the car further slowed to a complete stop.
FSD is terrible in many situations, that is absolutely true. But, knowing the limitations, it can also provide some great advantages. In heavy interstate traffic, for example, I'll usually enable it and then tap the signal to have it do a lane change: I'll check my blind spots and mirrors, look for closing traffic behind me, but it's very nice to have the car double checking that nobody is in my blind spot. There are many situations where, knowing the limitations, the Tesla system can help.
By crashing in a way that is fatal to your life
In all honesty, I'd probably spend a few seconds trying to figure out "what does the car see that I don't?" and let it come to a stop. Maybe it's a catastrophic system failure that I can't see. Maybe it's an obstacle headed into my path that hasn't gotten my attention. If my reflexive reaction is supposed to be to distrust the car's own assessment of the situation, then the system isn't good enough yet.
Good for you that you apparently know how to "use it correctly" or whatever but that's not exactly the point here.
― George Carlin