Normally I'd expect the fact that they aren't losing these suits regularly to factor as a prior in people's opinions about the cars' safety. But I've also been in enough of these threads to know that it's actually all about people's opinions of the CEO and not the product.
Unless you can show in court that their system was defective. It is designed to not be able to do the full driving task, to need interventions. It performs as designed.
Except, notably, the one you want to yell about the most. I'm just saying that should maybe be part of your analysis.
Look. The cars are safe. We've been having the same argument, again and again, for years and years now. People show up in these threads just taking as a given that these systems are unsafe, and refuse to adjust that prior as these cars become some of the most popular products as a market.
[1] What were perceived as -- IIRC it's since been argued that the problem in the suit didn't actually exist. And they settled anyway!
I think the fact that it has supporters is because of the CEO, not the fact that it has detractors, though I'm sure some are more vehement because of the CEO.
It's amazing. The most fun I've had in a vehicle in my whole life, and absolutely worth the purchase price. I'm sorry you aren't impressed, really, and I genuinely can't understand why there's so much hate about something so great.
But none of that is relevant here. The question at hand is "is it safe?", and the clear answer is "yes, quite".
It may be fun for you but its not fun for all the other people on the road put at risk by Tesla owners that dont know the limitations of their vehicles.
No thanks, that shouldn't be allowed on public roads.
FSD monitors the driver for indicators of inattention, e.g. not looking at the road. First you get a visual alert, then a shrill alarm, and if you're still not paying attention the vehicle slows to a stop. Drivers with a history of inattention are locked out of FSD.
Contrast with traditional cruise control where you can fall asleep at the wheel going 100mph and it'll hold that speed until you crash. Of course no one wrings their hands about that.
FSD (Fools Self Driving) feature is a complete scam with dangerous and deceptive advertising putting the lives of the driver and others at risk. The Tesla vehicles themselves are not.
Isn’t it more realistic that FSD is one of those classic engineering problems where the last 10% takes 90% of the effort?
Also, the “move fast and break things” is symptomatic of the broader tech industry. Despite being glorified for decades, it’s become a toxic mentality that’s hurt us with social media, and is now exponentially more pronounced with AI. Sure, maybe Elon should be “made an example of”, but collectively we need to sober up and realize this is a systematic issue in tech rather than scapegoating one eccentric billionaire.
I'm appalled the NTSB hasn't reigned in Tesla. I'm cynical enough to start wondering what officials they've "bought off."
1) Tesla hasn't demonstrated that future hypothetical, so it's a speculative argument at best.
2) It's still unethical if they could have achieved that more safely by following industry norms and best practices, which they consistently fail to implement. Tesla could hire safety drivers for testing before public releases, but they don't. They could implement more redundancy to mitigate SPOF, but they don't. They could file the same regulatory paperwork others do (insufficient as that is), but they don't do that either. They don't even use industry standard terminology and safety concepts like ODDs.
But technically, Tesla is “fully self driving” in the sense that it can autonomously navigate, steer, etc. That doesn’t mean it won’t crash. It’s also still in Beta.
In the long run, I think FSD will become better than the majority of human drivers. Seemingly, the only way these remaining edge cases can be solved is by collecting more training data from real world driving scenarios. So the NTSB is faced with the difficult decision of allowing this Beta version to crash and even cause fatalities in the short run, betting on the prediction it will drastically improve road safety in the long run. It’s a complicated dilemma, in my opinion.
You fell for a very common misuse of statistics there. It may become better than the average human driver, but the sample for that average includes the 5-10% of drivers who cause almost all accidents. Like for other things where power laws apply using averages (or medians) on that data essentially removes any information.
FSD likely requires GAI.
The GAI requirement is an interesting theory but why? I find it dubious because, at least personally, driving seems to occur semi automatically in my motor cortex. When a car pulls out in front of me I’m not using my frontal cortex to slam on the brakes. It feels involuntary.
From a few anecdotal experiences in this thread, you would be able to tell subconsciously that that Burger King sign is not a stop sign, not even for a brief second, that floating plastic bag is not an object that one has to avoid collision with and there are just millions of edge cases humans properly handle.
Of course I might be wrong and it might not require GAI, but there is still no human construction that can even close on a human brain.
I grew up in the US, in the 1970s-80s, when we had [optional] seatbelts, that would cut you in half, and snap your spine, exploding [leaded] gas tanks, glass that would slice you to ribbons, and steering wheels that would impale the drivers.
It's always been retroactive. Ralph Nader made his name on Unsafe at Any Speed[0].
Money speaks very loudly in the US, and we have always been willing to sacrifice quite a few things (as long as the non-money-makers are making the sacrifices), and take huge risks (as long as the non-money-makers are taking the risks), for money. It's just the way we are.
No, "controlled conditions" don't do a very good job of training or proving an AI. Supervised cars on actual roads is the way it has to be.
That's the most sound policy, provide the public with alternative means of transportation that are convenient and more environment-friendly, leave room on the streets for the ones that actually require a car for their transportation. If you leave in the middle of bumfuck nowhere and can't be served by public transportation, great, you probably need a car. If you drive your car in cities for trips of 5-10km because it's convenient then we, as a society, should find ways to disincentivise this usage.
Modern car crash test into older car.
The linked article is literally about the data Tesla provides showing that their vehicles are safe, and it doesn't even refute the conclusions. It's about nitpicking at the margins.
To be clear: Tesla's data says autopilot use is about an order of magnitude safer than human driving. The author complains that this is confounded by the fact that AP is used in different conditions where the baseline accident rate is lower.
You need a lot of confounding effect to make up for an order of magnitude delta! Seems like everyone wants to skip that part and yell about the CEO instead.
The cars are safe. They are extremely common vehicles now. If they were unsafe we would know. We know they are safe. Any argument needs to treat with this truth, or it's basically just a conspiracy theory.
The article is not nitpicking at the margins. It's showing how Tesla is making misleading claims out of their insufficient data by not accounting for operating conditions. That's literally how Tesla is claiming Autopilot is "an order of magnitude safer than human driving".
> You need a lot of confounding effect to make up for an order of magnitude delta!
Or you could just make misleading comparisons to claim that. That's pretty easy too.
Which, fine. Everyone does bad marketing, there's lots to yell about. Maybe consider yelling about the deceptive marketing practices of a product you do use? There are lots to choose from.
"Show us unbiased data proving it"
"If it were unsafe, we would know"
Sigh.
I own a Tesla and use FSD, but thanks for assuming.
What the linked article is doing is just excuse-making, basically.
I get it. You want to take the data at face value without questioning anything.
I'm not saying take it at face value at all. I'm happy to discuss methodology and to admit that the numbers on Tesla's site amount to something closer to marketing than science. But nonetheless, the cars remain safe. This methodology would have to be outrageously confounded to produce a result like this for an unsafe car, and you know it.
Dude, you couldn't even get the date of the article right.
I would assume medicines have similar or higher levels of complexity involved. However, the medicine analogy is poor. We often release medicines for use even when they have incredibly harmful side-effects. We do the testing to quantify and understand those side-effects, not prove that each medicine does exactly what it should with no side effects.
A good general overview from two minute papers: https://www.youtube.com/watch?v=6hkiTejoyms Tesla's AI Day 2022 video, timestamped around labeling and simulation https://youtu.be/ODSJsviD_SU?t=6037
There are human trials for drugs as well, but those are only the latest phase, and with explicit permission.
Wouldn't that also be true of your medicine analogy? If someone takes an experimental medicine but experiences an unknown side effect and passes out while driving, isn't the risk profile equivalent?
That hypothetical certainly is far less probable than the Autopilot case, but for any experiment that takes place outside controlled conditions, there can always be unknown and unforeseen consequences that affect people who have not given explicit consent to being part of said experiment.
To argue that we cannot test things unless all potentially-involved parties have given consent is to argue for an impossibility. There is always some small amount of risk of higher-order consequences.
I would even go further and say that arguing for such is an unnecessarily conservative approach that hamstrings any attempt at improving the world. See, for example, this commentary on the current state of IRBs [1]
[1] https://astralcodexten.substack.com/p/book-review-from-overs...
Also, the “danger to others” aspect is even then negligible as opposed to some faulty software controlling a ton of steel/battery in a system that depends heavily on predictable behavior of each other.
I guess I don't understand your argument. We put faulty humans in control of a ton of steel/battery in the same system all the time; humans have the disadvantage that when lessons are learned, they aren't transferrable without great amounts of effort. Self-driving systems can share learnings across the whole fleet with software updates, which seems like a strictly better solution in the long term.
It seems like you're making perfect the enemy of "better".
If it showed things were better than average it would fit their message and I’d expect them to publish it.
If it showed things were average it would not fit their narrative. If it showed things were worse it would contradict their narrative.
Seems like there is only one reason to withhold numbers or publish them in a misleading way. They prove your claims that your tech is way safer wrong in some way.
Like...maybe it's fine. But even at 0 fatal crashes in 5 billion miles they won't have a strong argument that a more widely deployed FSD would have similar success without a lot more data to explain how the thing they measured directly connects to the thing they'll want to claim it represents.
That’s the problem. We don’t know the safety record because Tesla won’t release it.
But the anecdotal evidence we do have is very worrying. And the development model doesn’t inspire confidence.
So the calls to ban it seem very reasonable to me.
Musk boasts that Teslas have the highest crash safety ratings ever, despite this being contradicted by the testing bodies (because they don't distinguish between cars with 5 star ratings). Another bold claim, contradicted by the data!
Teslas collect tons of data while driving and when crashing, which Tesla processes. You believe that this data demonstrates that FSD is safer than regular driving but that musk is too cautious to release it?? How can you square that with Musk's behavior generally?
But we don’t know that because Tesla won’t tell us.
Nobody, even Tesla, is currently arguing that Tesla FSD is ready for driverless.
It's true that within the small geofences in which Waymo operates, their cars are dramatically less likely to come unstuck than a Tesla running FSD beta. On the other hand FSD beta drives on pretty much any public road and not a curated set of pre-driven, pre-mapped, pre-validated streets. Yes there are many geographies where it struggles; there's a cottage industry of people obsessively showing these on YouTube. There are also some areas of the country where FSD beta is utterly rock solid and people routinely drive substantial distances with zero interventions.
It's also not entirely accurate to describe Waymo cars as "fully driverless" as they have a team of human employees remotely monitoring the fleet and can remotely operate any vehicle.
This is absolutely not true. Waymo remote agents cannot operate a vehicle. All they can do is answer questions from the vehicle or plot a path to go around if it gets stuck. They cannot make any safety critical interventions. Waymo has talked about this many times.
It’s fully driverless because the vehicle is responsible for attaining minimal risk condition all on its own. This is how other L4 companies like Cruise works too.
Zoox has a good explanation on how their teleguidance works: https://youtube.com/watch?v=NKQHuutVx78
Fact is, a Waymo operator can take enough control of the vehicle to be considered in control of it. It doesn't matter if control comes in the form of a steering wheel or a joystick or drawing vectors on a map. Control is control, whether it's direct, indirect, or funnelled through a governor. Waymo operators can intervene and direct the car as needed.
Terrible analogy. You are sitting in the driver's seat with your hands on the wheel. If your car is about to hit someone, you can prevent it because you are the driver. This is what FSD is — ultimately driver is in control. No such luxury for Waymo/Cruise as their remote operators cannot do any safety critical interventions like preventing crashes.
All they can do is non-safety critical interventions by way of passing indirect commands. A Waymo is in control of itself at all times and is fully responsible for its own safety. You can play with words like "control" all you want, but this is the definition of SAE Level 4 autonomy.
If anything, calling FSD "autonomous" is highly misleading because drivers are preventing mistakes and thereby hiding its real safety outcome.
I agree, it's a terrible analogy if I was attempting to make the argument you thought I was trying to make, per your explanation for why it's a terrible analogy.
My point is that it's invalid to say that Waymo operators don't have "control" of the vehicle because it can refuse to perform some tasks. Because my car can refuse to perform some tasks too.
The deeper point is that the difference between "in control" and "not in control" isn't binary. It's all points along a continuum. If a Waymo operator controls a vehicle remotely, it's not the same level of control as a human holding the steering wheel. But it's still control in the larger sense. If the operator wants the car to stop, it will stop. Period. If they want the car to perform some action that isn't stupid or dangerous (such as colliding with a pedestrian) it will perform that action.
> If anything, calling FSD "autonomous" is highly misleading
I agree. Which is why I didn't. In most situations FSD beta can handle the "full" task of driving. By "full", they mean it's not intended to be limited to a strict subset of task, e.g. only on highways. It can "fully self drive" the car competently at least ~99.9% of the time, as evidenced by the fact that it can routinely drive most roads reliably. Drives shown on YouTube are mostly owners intentionally looking for roads which flummox it, as that's how they show improvements as software updates are released.
The challenges which preclude FSD from being a system capable of autonomy — the 0.1% of remaining error — isn't errors in spatial awareness or control surface failure. It's stuff like path planning and road sign interpretation, neither of which involve sensors that Tesla vehicles lack.
It’s a bit like stress testing two frameworks, one tested only for a day, but at the frontpage of HN, while the other was up for a months, with barely any clicks.
Look, I get that Musk isn’t popular here. But please, give me some hope for humanity and at least pretend to care about facts and evidence.
vote often, vote early
Come to Australia, the driving age here is 18. (And for your first three years as a licensed driver, you have to show plates to indicate your lack of experience because, unlike in the stupid USA, we don’t assume that passing one driving test is enough to qualify you as a “good driver”.)
I think they both have a ceiling and that the ceiling on AI is much higher.
Which is my point — the road is full of hideously bad/dangerously inexperienced human drivers holding driver licenses. Given this, it's hard for me to interpret the pearl-clutching around humans co-driving with FSD beta as anything other than confected paranoia.
Of course if every car would be self driving we wouldn’t have crazy drivers, and accidents would diminish, but so would if we simply didn’t have crazy drivers. Neither is feasible in the near future, and in the short term, humans are much better at handling tricky situations (like.. not emergency braking at the sight of a plastic bag in the air)
I wonder why there's no modified version of driving tests for AVs that each software iteration takes.
It is a perfectly valid policy to want that proof ahead of time. I’m just pointing out that this requirement would effectively kill development of this technology for the foreseeable future.
https://teslamotorsclub.com/tmc/threads/tesla-lays-off-over-...
Don’t get me wrong, I irrationally hate the robots, used to see them all the time around Phoenix, was actively gunning for them and only had one time I could have rammed one[0] with it not being my fault. Which says a lot because it was (well, still is) a life goal to take out a robocar so I was really trying.
I used to see them doing a bunch of really stupid stuff but never anything that was outright dangerous, relatively speaking. One did stop in the fast lane when everyone was zipping by at 70+ because it couldn’t figure out how to merge into the left hand exit ramp but plenty of snow birds do that too so it’s not completely unheard of.
[0] (un)fortunately I was behind on my rent and couldn’t afford to spend who knows how long dealing with an accident investigation so the robot got to go about its day unmolested.
At least they hold liability if anything happens, though, which is an important step.
The onus is on Tesla to provide honest and accurate data, and they have shown time and time again that they are not capable of doing so.