Tesla really trying to engineer good will, and unsurprisingly only using their own data, which paves over things like "driver intervention prevented FSD from crashing in this instance" or "FSD disengaged 2s before crash, therefore driver error".
Rather than play statistics games with self-reported dressed up supervised driving data to try and trick people into rolling the mortality dice with a robotaxi, just let the cars drive around empty. But he can't do that, because these cars are not FSD.
Even worse, the government could easily mandate LIDAR for autonomous cars, and that would basically kill Tesla overnight.
How do you meaningfully count interventions, though?
I intervene on FSD all the time, but only due to differences of opinion, like I want to take a different route, or I think it's being too cautious, or I want to change lanes earlier / later than it decided to. In a year of use I can't remember a time when I intervened to prevent an accident.
So if you just looked at number of interventions in my data, it doesn't really tell you anything except that I'm too much of a control freak to just let it do its thing...
Tesla has the data from Austin, they closely track when the supervisor pulls the e-stop. But that data, the most critical data, is unsurprisingly absent here.
The dude has got six weeks to get those supervisors out of the car, and he isn't even giving the most critical data around them. Not a good sign.
Elon will probably find a near closed loop around an empty development in Austin, put two robotaxis there without a driver, and let people do a loop. Tesla contractors will walk the area an stop people from crossing when a robotaxi is near. Then he will proclaim there are driverless taxis in Austin and that they will expand to the rest of the US by mid 2026, and then talk about how actually optimus robots in 2027 are the real thing to focus on.
Tesla Autopilot/FSD is by all accounts an impressive piece of tech, and the stats about accident prevention are pretty convincing. The problem IMO is that Tesla & Musk have muddled the conversation so much over the years and made so many promises that it is impossible for the average consumer to know what is or isn't real at this point.
Do I need to have my hands on the steering wheel or is the car actually self driving? Can my Tesla drop me off where I need to be and go park itself? And then can I summon it when I'm done? Can I turn my Tesla into a taxi that picks up other people and earns me money? Is it even possible for a Tesla to drive itself unsupervised?
All of these have been promised over and over again for the past decade, and at this point it is impossible for anyone to set the right expectations for themselves.
I think they made one delivery they claimed was unsupervised. It might have had lead cars and chase cars like early Waymo, and a custom build with HD mapping, I'm not sure.
I use it every single day and it is amazing! I have had my car for 6+ years now and it has only gotten better. Starting with "Navigate on Autopilot" when I originally got the car in 2019 which was pretty janky, it has only gotten better and better (with the rare regression in some cases).
As of the last year or so, I don't even have to touch the steering wheel anymore!
It is a miracle consumer technology but it does mistakes. I have 40k miles in last 2 years mostly using fsd (since v13). And my anecdotal experience is that it drives itself well until it doesn’t. So far I was lucky enough to pay attention when it was critical. Which is hard to do because car doing good most of the time.
It is legal. I still have to pay attention to the road. There is attention monitoring using the in-cabin camera which falls back to steering wheel torque detection if the cabin illumination is too low.
On ford and I believe Chevy's systems you can let go of the wheel, however their systems only work on pre-mapped roads, and those roads are 99% interstates and highways.
The problem is that if it drives well for 30,000 miles (unsupervised) on residential roads before steam rolling little billy on his bike, you will get a deluge of people who swear the system is excellent.
But when you incorporate that tech into a fleet doing 100k residential miles a week with no supervisor, your mowing down 12 kids a month.
So... that was exactly the point upthread. You're making a semantic argument over the proper definition for the word "real" when applied to autonomous vehicle systems. Nothing in this argument is actionable in any way. You can't conjure real dead kids, so you need to describe hypothetical ones. That's... yeah.
Nonetheless, our cars drive us around anyway. Neither they, nor us, actually care about hypothetical steamrollered kids.
The argument is that Tesla needs to be doing hundreds of thousands of miles without intervention to be trusted for robotaxis.
Most people using FSD don't come close to the mileage needed to get an idea of the safety level needed. If a Tesla robotaxi kills a kid, Tesla is done, and there won't be a coming back.
So Tesla actually needs millions of miles without critical intervention before they can confidently let these things en masse out on the streets.
A whole tesla fanboy meetup collectively will not have enough FSD miles to see something like that, but a robotaxi fleet will encounter it within a year.
> Tesla actually needs millions of miles without critical intervention
So... agreed. I think that sounds like it's in the right ballpark.
Here's the thing though: your whole argument (sort of a para-freakout, really) hinges on this evidence not existing.
It's true that citing numbers from supervised cars isn't the same thing. It's not true to argue that it not being the same thing is (0) not at least a somewhat reasonable proxy for the evidence you want to see, (1) evidence that it doesn't or can't exist, (2) evidence for the opposite case (you seem to be claiming that the fact that it's supervised means that it must be), and in particular (3) evidence for suppression of contrary evidence, as some of your more conspiracy-leaning comments seem to imply.
Isn't the Occam's explanation here that, yeah, the car looks pretty damn safe as shown by billions of miles of travel? Why must you be going to the mattresses to argue against something that seems pretty common sense to me?
Occam's Razor is a principle for comparing explanations, not for making predictions.
> It's not true to argue that it not being the same thing is (0) not at least a somewhat reasonable proxy for the evidence you want to see, (1) evidence that it doesn't or can't exist, (2) evidence for the opposite case (you seem to be claiming that the fact that it's supervised means that it must be), and in particular (3) evidence for suppression of contrary evidence, as some of your more conspiracy-leaning comments seem to imply.
Actually, (0) is true. Numbers from supervised FSD are not a reasonable proxy for unsupervised FSD, especially if accidents may be occurring immediately after FSD disengages.
(1) is also true, because if the evidence did exist, then Tesla would have already have publicized it.
(2) is also true, because if Tesla thought they could employ unsupervised cars like Waymo, then they would have already have done that.
As for (3), if Tesla had a reputation for transparency & honesty, then they could have provided additional data on accidents to show that accidents are not occurring right after FSD disengages at significant rates.
I'm in Los Angeles, which can be a challenging place to drive. Each time they give me a free trial of FSD for a month, I enable it and test it with excitement and optimism. Each time I only use it for a day or two before it does something dangerous enough to scare me.
Yeah, I'll be honest, that sounds awful. I like public transit and I like driving myself. Hybridizing with me in the drivers seat but not touching the wheel sounds annoying and tiresome.
I share this opinion. I've used FSD quite a bit during the free trial periods and each time come away with the sense that it's like driving with a newly licensed teenager at the wheel. If I have to be as alert and ready to avoid an accident as when I'm in command of the car, then this offers no improvement to the experience, just an added layer of stress trying to anticipate the actions of yet another actor in the environment.
My experience has been different. I find that not needing to have hyper-focus for extended periods of time with constant micro-adjustments has a big effect on fatigue - especially on long trips. Not needing to touch the wheel while using gaze-detection for attention tracking just reduces the annoyance IMO.
I find it very similar to operating an airplane with a reliable autopilot. The GFC-700 is super good at what it does. But it is still on me to monitor what it is doing, while at the same time significantly reducing my workload.
It would be more meaningful if we could see the data controlled for all the factors auto insurance is priced on (driver age, zip code, credit score, vehicle age, etc.)
Does Tesla offer massively lower insurance premiums for drivers that do the majority of their driving with FSD?
When I bought my Tesla, I had State Farm coverage for my existing (8 yo) car.
State Farm quoted me a rate of $240/mo for switching insurance to the Tesla ( up from $130/mo ). This is in California, Bay Area.
On a whim, I fired up the Tesla app and requested a quote for insurance through them.
They quoted me $134/mo.
I know, anecdata, size = 1. But I was surprised how low it was. I sent the coverage information to State Farm, to see if they would match it, but they just shrugged and said no, we can't match that.
Have you had to file a claim with them yet? I go with State Farm not because I can get a cheaper rate but because for their price they provide all of the services I expect from my insurance company when I need them.
Second this, after one incident (at fault) and having to pay nothing for the other persons $4k bill (no hassle, no deductible), SF is a good choice and paid itself back
That makes sense, and Tesla should be pricing lower for liability insurance if they have private data showing their cars are safer. But the "put your money where you mouth is" evidence would be if they offered a large discount for their own insureds that used FSD for most of their driving. I strongly believe FSD is safer than human drivers, but not sure about the magnitude. I think the FSD discount Tesla offers is a good clue, and I expect this discount to increase in the future.
Marketing page with data from the brand. Conclusion: those numbers aren't worth anything, unless they come from independant organism they are just advertising and no more believable than old cigarettes ads.
But how many of those miles are spent hanging out in the left lane of a highway while other drivers, having to pass on the right, send death stares to the person in the passenger seat?
In my experience FSD doesn't like staying in the left lane. It'll use it to pass but then gets out of it as soon as it can. Often too soon as it just has to shift left again to pass the next car...
If we take this at face value then the only remaining explanation for why the insurance industry reports that Tesla drivers have more accidents than drivers of any other brand is that Tesla drivers are a self-selected group of extremely bad drivers, which is itself notable.
I have a 2018 Model 3 with basic autopilot. I have an FSD question. Does FSD have different driving behaviors from basic autopilot? I'm not saying "capabilities"; obviously it has more capabilities, but I mean the actual driving choices.
I get that FSD (maybe) has/requires better hardware than my car. But what I hate about autopilot is all around basic driving:
* Lane centering. It's extremely aggressive about lane centering, if you're in the right lane and an onramp joins from the right, the car aggressively drives to the right as soon as it perceives that the lane is wider.
* Throttle/brake behavior. It waits too long to brake (despite having radar in my car, which can supposedly "see" more than one car ahead), and when it does apply the brakes it doesn't do so smoothly. It tips in somewhat aggressively, and you can feel the discrete steps in brake force application change. Ditto for acceleration when the traffic in front of me moves.
There's no reason to think that any of this has anything to do with compute power, it all seems to be programming decisions that have been made for whatever reasons, so I can't see why FSD would be different.
And yet, if FSD drives like this, I don't get how anyone can think it's good? On the other hand, I've also heard people say they think autopilot is good, which it's clearly not, so it makes me judge their driving skill rather than the different models. But perhaps there's some matrix of hardware revisions and software/decision models out there that I'm unaware of, that explains differences in driving behavior, if they exist?
A 2018 vehicle is 2 major version behind on the FSD it can use. Tesla is on v14 and 2018 vehicles are stuck on v12.
There's been rumors of a v14 lite coming out because tesla REALLY doesn't want to deal with the fact that they promised the 2018s could be fully autonomous.
Interesting, thanks. I haven't followed the hardware revisions on my car very closely; my father has a 2017 S, and I'm so used to thinking I have the "newer revision", since my MCU was newer than his, and he only recently upgraded it. I guess it should have been fairly obvious that a lot has changed in the past 7 years, but I've been hearing a nonstop trickle of autopilot/FSD news on these cars for so long that I guess I just assumed I had recent-enough support.
Honestly, I just dismiss all of the promos they throw at me.
Just yesterday I got an ad in the app "refer a friend and try FSD (supervised), and get $15!"
So I opened up the app just now and I suppose I got my answer that proves my initial premise was incorrect -- I need a hardware upgrade for FSD. Womp womp.
(That said, it still seems like the lane centering and brake/throttle behavior should have been easily fixed without a FSD hardware upgrade).
It strikes me as incredibly weird to maintain two separate stacks, especially because FSD was "available" when I bought my car. In retrospect, though, I guess it wasn't actually available to use, just for purchase.
So I suppose they had a parallel development process for FSD while building autopilot features?
But since there is a hardware support overlap, it seems like at some point you'd migrate autopilot cars to the FSD software stack, with limitations added in via feature flags.
Autopilot basically hasn't gotten any new features since FSD came out.
> But since there is a hardware support overlap, it seems like at some point you'd migrate autopilot cars to the FSD software stack, with limitations added in via feature flags.
This is exactly what a lot of people speculate will happen soon.
All behaviors I've observed with FSD in my 2018 model 3.
It's possible v14 is better. v12 was certainly better than v11 in all those regards. But there are still issues with the car making dumb lane choices in v12.
You won't catch me defending Tesla often, but at least they actually do regular software updates and committed to needed hardware updates.
Ford and Chevy otoh are doing exactly the self sabotaging greedy bullshit you expect. Chevy already told previous gen super cruise owners that they are no longer getting updates or more mapped areas, and ford just segregated their line into before and after 2025. 2025 cars getting the latest bluecruise version, earlier cars don't have the hardware (but you still need to pay annualy for that deprecated older version!)
Ford also has never exteneded the mapped area in 4 years, and releases minor updates maybe once a year, where you will wait another 8 months to get the OTA.
It is supposed to include hardware upgrades too. Chevy and others upcharge for semi autonomous safety features too, but I'm not sure to what extent that is pure software or added hardware.
> So Tesla is charging $8000 to activate full safety software features in their vehicles?
I think you're being obtuse, but to be clear, many car manufacturers offer trims don't include features that would qualify making the car 'safer' - blind spot detection, back up cameras (I think these are legally required now but were a premium feature for over a decade), parking assistance, crash detection, etc.
I have a Tesla and use FSD every day, and while it is a safety feature, it is _the_ pinnacle 'luxury' feature that a car can have today and they honestly do not charge enough for it.
Well, I think that is also morally reprehensible in all other cases where it's also a matter of activating software safety features.
Most of the things you mentioned aren't software locked behind a paywall, hopefully, you don't swipe your credit card and get those features added via OTA in minutes. If your car doesn't have back seat airbags it's hopefully not because you haven't paid for the back seat airbag in-app purchase.
Okay you are correct. If they are going to market FSD as a safety feature and it's just a software update, they need to include it by default and adjust their price.
I didn't realize how much they market it as a safety feature.
They have a huge store of data on accidents in teslas per mile driven. Why don't they compare their actual data on accidents? Well, they would, but it probably is worse with FSD.
For the claims they are making the only peer is Waymo.
If they want to put themselves in a peer group that only has driver assistance systems, then the comparison should be to other similarly-priced and new vehicles, no U.S. average 12-year-old Corollas.
Anyway what is most amusing about this promotional material is that from the very first frame it inadvertently highlights how much worse it is than Waymo. The "Avoiding T-bones" scenario only seems like the car came out of nowhere because Tesla's camera system is so limited. Waymo would have seen it coming a mile away.
That is old news. Today we learn that Optimus will be a world class surgeon and "democratize" medical care, that Grok 5 has a 10% chance of achieving AGI, that Grokipedia will be named Encyclopedia Galactica and that Elon will put 100GW of solar powered "AI" satellites into orbit.
Doesn't matter I and almost everyone in my circle won't touch a Tesla.
The cybertruck was a huge boondoggle and only doctors and true beleivers bought them but over time I don't see them in the parking lot when I see the doctors in the building hmmm.
Also Elon "skipping around like a dipshit" (Tim Waltz) and now claiming he's worth a trillion dollars.
By his own admission his identity is entertwined with Tesla's.
Given how famously the tesla switches self driving off moments before an accident, these statistics are impossible to trust (but also because musk is famously a liar)
> If FSD (Supervised) was active at any point within five seconds leading up to a collision event, Tesla considers the collision to have occurred with FSD (Supervised) engaged for purposes of calculating collision rates for the Vehicle Safety Report.
Sure, it's very possible, and I'm sure has happened (an autopilot-instigated crash that happened >5sec after disengagement), but I do think 5 seconds is a reasonable threshold for this data, at the very least.
Tesla appears to have an impressive history of automatically disengaging FSD immediately before a collision, and this page is quite light on details of what they’re comparing to what, so I would take it with a large grain of salt.
> and this page is quite light on details of what they’re comparing to what
In the page:
"If FSD (Supervised) was active at any point within five seconds leading up to a collision event, Tesla considers the collision to have occurred with FSD (Supervised) engaged for purposes of calculating collision rates for the Vehicle Safety Report."
They are pretty open about how the stats are reported.
The driver can at any time. And they do if it seems like it is going to do something stupid - which is getting rarer and rarer as time goes on. As a Level 2 system, the driver is always supposed to supervise the operation and stay alert.
Musk has proven time and again that things his critics say are impossible/unrealistic ends up being achieved late. See anything from reusing rocket stages to the goals from Tesla's 2018 Compensation Plan[1] ("If Mr. Musk were somehow to increase the value of Tesla to $650 billion — a figure many experts would contend is laughably impossible and would make Tesla one of the five largest companies in the United States ...")
Or the Arianespace guy saying SpaceX is "selling a dream". To quote: "I think a $5 million launch or a $15 million launch is a bit of a dream. Personally, I think reusability is a dream."
True, we need to see the details of how they compute all this data. I do remember reading that they include crash data up to like a minute after FSD has been disengaged so these types of crashes should be included.
Nevertheless, the latest version of FSD I can certainly believe is seven times less likely to get into an accident than the average driver. I experienced it daily.
Are they saving lives? The latest NHTSA FARS analysis I saw (iSeeCars, report is through 2022) shows Tesla at 5.6 fatalities per billion miles vs the 2.8 average.
You realize that almost every politician has done a similar type of pose right? So if there’s a picture or a video of your favorite politician doing that pose you’re automatically going to assume they are Nazi?
Just a minimal knowledge of Musk‘s ties to Israel, and friendships with Netanyahu, Jared Kushner, and others you should realize that there’s no way he’s a Nazi or antisemitic. If anything, he is highly racist against Muslims.
Yes, if I see a video (not a still) of someone doing a nazi salute two times in a row, I'm gonna assume they're a nazi. Absolutey no disagreements with you there.
Doubly so if they also happen to tune into an AfD and a Tommy Robinson rally.
Maybe we don’t have the same definition of what a nazi is, because Tommy Robinson is a huge Israeli schill, Elon musk is friends with Netanyahu, Jared kushner, and many other Zionists.
There is a difference between waving and doing a Nazi salute. There is a difference between seeing a single frame of a wave out of context and a full video where he does it twice.
“Next year for sure, we will have over a million robotaxis on the road,” said Musk on October 21, 2019. “The fleet wakes up with an over-the-air update. That’s all it takes.”
Yes, we all know that when Elon makes promises about the future, he's always late delivering on them. But this is about the capabilities of something that's already here, and those kind of claims are way more reliable.
Interventions Elon, interventions.
Tesla really trying to engineer good will, and unsurprisingly only using their own data, which paves over things like "driver intervention prevented FSD from crashing in this instance" or "FSD disengaged 2s before crash, therefore driver error".
Rather than play statistics games with self-reported dressed up supervised driving data to try and trick people into rolling the mortality dice with a robotaxi, just let the cars drive around empty. But he can't do that, because these cars are not FSD.
Even worse, the government could easily mandate LIDAR for autonomous cars, and that would basically kill Tesla overnight.
How do you meaningfully count interventions, though?
I intervene on FSD all the time, but only due to differences of opinion, like I want to take a different route, or I think it's being too cautious, or I want to change lanes earlier / later than it decided to. In a year of use I can't remember a time when I intervened to prevent an accident.
So if you just looked at number of interventions in my data, it doesn't really tell you anything except that I'm too much of a control freak to just let it do its thing...
Tesla has the data from Austin, they closely track when the supervisor pulls the e-stop. But that data, the most critical data, is unsurprisingly absent here.
The dude has got six weeks to get those supervisors out of the car, and he isn't even giving the most critical data around them. Not a good sign.
Elon will probably find a near closed loop around an empty development in Austin, put two robotaxis there without a driver, and let people do a loop. Tesla contractors will walk the area an stop people from crossing when a robotaxi is near. Then he will proclaim there are driverless taxis in Austin and that they will expand to the rest of the US by mid 2026, and then talk about how actually optimus robots in 2027 are the real thing to focus on.
Tesla Autopilot/FSD is by all accounts an impressive piece of tech, and the stats about accident prevention are pretty convincing. The problem IMO is that Tesla & Musk have muddled the conversation so much over the years and made so many promises that it is impossible for the average consumer to know what is or isn't real at this point.
Do I need to have my hands on the steering wheel or is the car actually self driving? Can my Tesla drop me off where I need to be and go park itself? And then can I summon it when I'm done? Can I turn my Tesla into a taxi that picks up other people and earns me money? Is it even possible for a Tesla to drive itself unsupervised?
All of these have been promised over and over again for the past decade, and at this point it is impossible for anyone to set the right expectations for themselves.
Those questions all have very clear answers.
> Do I need to have my hands on the steering wheel or is the car actually self driving?
You don't need to have your hands on the steering wheel, but you do need to be paying attention to the road.
> Can my Tesla drop me off where I need to be and go park itself?
Not yet.
> And then can I summon it when I'm done?
Yes.
> Can I turn my Tesla into a taxi that picks up other people and earns me money?
Not yet.
> Is it even possible for a Tesla to drive itself unsupervised?
They can (e.g., https://youtu.be/BO1XXRwp3mc), but it's not enabled for consumers yet.
Good progress, but important to keep in mind that the number of unsupervised miles is zero.
I think they made one delivery they claimed was unsupervised. It might have had lead cars and chase cars like early Waymo, and a custom build with HD mapping, I'm not sure.
Yes, and likely also a live teleoperator ready to "grab the wheel".
The problem with this is the source. Tesla lies constantly and it has from the very first FSD demo [1, 2] if not earlier.
1: https://www.cbsnews.com/news/tesla-autopilot-staged-engineer...
2: https://www.cnn.com/2025/08/21/business/tesla-nhtsa-self-dri...
I use it every single day and it is amazing! I have had my car for 6+ years now and it has only gotten better. Starting with "Navigate on Autopilot" when I originally got the car in 2019 which was pretty janky, it has only gotten better and better (with the rare regression in some cases).
As of the last year or so, I don't even have to touch the steering wheel anymore!
It is a miracle consumer technology but it does mistakes. I have 40k miles in last 2 years mostly using fsd (since v13). And my anecdotal experience is that it drives itself well until it doesn’t. So far I was lucky enough to pay attention when it was critical. Which is hard to do because car doing good most of the time.
I don't even have to touch the steering wheel anymore!
Is that legal where you live?
It is legal. I still have to pay attention to the road. There is attention monitoring using the in-cabin camera which falls back to steering wheel torque detection if the cabin illumination is too low.
On ford and I believe Chevy's systems you can let go of the wheel, however their systems only work on pre-mapped roads, and those roads are 99% interstates and highways.
Even if it's legal, it's pretty stupid.
> I don't even have to touch the steering wheel anymore!
What's the usefulness of this if you still have to pay attention at all times?
The level of engagement during the FSD (I only oversee the FSD) is about 10% of the engagement during “I drive”.
Nobody can be told what FSD is. You have to see it for yourself.
Ditto. The volume of knee-jerk hatred these posts engender here really just fails in the face of the capabilities of the actual system.
Like, it's OK to shout and scream about LIDAR and supervision and disengagement and all. But... it still drives itself! Really well.
The problem is that if it drives well for 30,000 miles (unsupervised) on residential roads before steam rolling little billy on his bike, you will get a deluge of people who swear the system is excellent.
But when you incorporate that tech into a fleet doing 100k residential miles a week with no supervisor, your mowing down 12 kids a month.
How many kids has it steamrollered? Obviously not 12 a month! Seems like this is an argument to be had with real numbers, no?
Zero, because there isn't, and hasn't been, a single unsupervised Tesla on the road.
There are no real numbers, because there are no real self-driving Teslas.
So... that was exactly the point upthread. You're making a semantic argument over the proper definition for the word "real" when applied to autonomous vehicle systems. Nothing in this argument is actionable in any way. You can't conjure real dead kids, so you need to describe hypothetical ones. That's... yeah.
Nonetheless, our cars drive us around anyway. Neither they, nor us, actually care about hypothetical steamrollered kids.
The argument is that Tesla needs to be doing hundreds of thousands of miles without intervention to be trusted for robotaxis.
Most people using FSD don't come close to the mileage needed to get an idea of the safety level needed. If a Tesla robotaxi kills a kid, Tesla is done, and there won't be a coming back.
So Tesla actually needs millions of miles without critical intervention before they can confidently let these things en masse out on the streets.
A whole tesla fanboy meetup collectively will not have enough FSD miles to see something like that, but a robotaxi fleet will encounter it within a year.
I don't think Tesla will be done when they kill a kid. FSD already has killed people: https://en.wikipedia.org/wiki/List_of_Tesla_Autopilot_crashe...
> Tesla actually needs millions of miles without critical intervention
So... agreed. I think that sounds like it's in the right ballpark.
Here's the thing though: your whole argument (sort of a para-freakout, really) hinges on this evidence not existing.
It's true that citing numbers from supervised cars isn't the same thing. It's not true to argue that it not being the same thing is (0) not at least a somewhat reasonable proxy for the evidence you want to see, (1) evidence that it doesn't or can't exist, (2) evidence for the opposite case (you seem to be claiming that the fact that it's supervised means that it must be), and in particular (3) evidence for suppression of contrary evidence, as some of your more conspiracy-leaning comments seem to imply.
Isn't the Occam's explanation here that, yeah, the car looks pretty damn safe as shown by billions of miles of travel? Why must you be going to the mattresses to argue against something that seems pretty common sense to me?
Occam's Razor is a principle for comparing explanations, not for making predictions.
> It's not true to argue that it not being the same thing is (0) not at least a somewhat reasonable proxy for the evidence you want to see, (1) evidence that it doesn't or can't exist, (2) evidence for the opposite case (you seem to be claiming that the fact that it's supervised means that it must be), and in particular (3) evidence for suppression of contrary evidence, as some of your more conspiracy-leaning comments seem to imply.
Actually, (0) is true. Numbers from supervised FSD are not a reasonable proxy for unsupervised FSD, especially if accidents may be occurring immediately after FSD disengages.
(1) is also true, because if the evidence did exist, then Tesla would have already have publicized it.
(2) is also true, because if Tesla thought they could employ unsupervised cars like Waymo, then they would have already have done that.
As for (3), if Tesla had a reputation for transparency & honesty, then they could have provided additional data on accidents to show that accidents are not occurring right after FSD disengages at significant rates.
What kind of setting do you live in? I live in the outskirts of Denver (Cherry Creek) and commute downtown and I have to intervene all the time.
I'm in Los Angeles, which can be a challenging place to drive. Each time they give me a free trial of FSD for a month, I enable it and test it with excitement and optimism. Each time I only use it for a day or two before it does something dangerous enough to scare me.
Los Angeles suburb - my commute is either highways or city streets depending on the traffic. It works really well here.
Yeah, I'll be honest, that sounds awful. I like public transit and I like driving myself. Hybridizing with me in the drivers seat but not touching the wheel sounds annoying and tiresome.
I share this opinion. I've used FSD quite a bit during the free trial periods and each time come away with the sense that it's like driving with a newly licensed teenager at the wheel. If I have to be as alert and ready to avoid an accident as when I'm in command of the car, then this offers no improvement to the experience, just an added layer of stress trying to anticipate the actions of yet another actor in the environment.
My experience has been different. I find that not needing to have hyper-focus for extended periods of time with constant micro-adjustments has a big effect on fatigue - especially on long trips. Not needing to touch the wheel while using gaze-detection for attention tracking just reduces the annoyance IMO.
I find it very similar to operating an airplane with a reliable autopilot. The GFC-700 is super good at what it does. But it is still on me to monitor what it is doing, while at the same time significantly reducing my workload.
Waymo's Safety page for comparison
https://waymo.com/safety/impact/
And how many miles have other manufacturers done with their own advanced adaptive cruise control systems?
It would be more meaningful if we could see the data controlled for all the factors auto insurance is priced on (driver age, zip code, credit score, vehicle age, etc.)
Does Tesla offer massively lower insurance premiums for drivers that do the majority of their driving with FSD?
> Does Tesla offer massively lower insurance premiums for drivers that do the majority of their driving with FSD?
Yes: https://www.tesla.com/support/insurance/fsd-discount
The discount is "up to 10%". I would call that a modest discount.
When I bought my Tesla, I had State Farm coverage for my existing (8 yo) car.
State Farm quoted me a rate of $240/mo for switching insurance to the Tesla ( up from $130/mo ). This is in California, Bay Area.
On a whim, I fired up the Tesla app and requested a quote for insurance through them.
They quoted me $134/mo.
I know, anecdata, size = 1. But I was surprised how low it was. I sent the coverage information to State Farm, to see if they would match it, but they just shrugged and said no, we can't match that.
Have you had to file a claim with them yet? I go with State Farm not because I can get a cheaper rate but because for their price they provide all of the services I expect from my insurance company when I need them.
[1] https://www.reddit.com/r/Insurance/comments/1kji0bm/just_fil...
Second this, after one incident (at fault) and having to pay nothing for the other persons $4k bill (no hassle, no deductible), SF is a good choice and paid itself back
That makes sense, and Tesla should be pricing lower for liability insurance if they have private data showing their cars are safer. But the "put your money where you mouth is" evidence would be if they offered a large discount for their own insureds that used FSD for most of their driving. I strongly believe FSD is safer than human drivers, but not sure about the magnitude. I think the FSD discount Tesla offers is a good clue, and I expect this discount to increase in the future.
Marketing page with data from the brand. Conclusion: those numbers aren't worth anything, unless they come from independant organism they are just advertising and no more believable than old cigarettes ads.
But how many of those miles are spent hanging out in the left lane of a highway while other drivers, having to pass on the right, send death stares to the person in the passenger seat?
In my experience FSD doesn't like staying in the left lane. It'll use it to pass but then gets out of it as soon as it can. Often too soon as it just has to shift left again to pass the next car...
If we take this at face value then the only remaining explanation for why the insurance industry reports that Tesla drivers have more accidents than drivers of any other brand is that Tesla drivers are a self-selected group of extremely bad drivers, which is itself notable.
Voyager 1 is 15B+ miles away. Kind of mind blowing.
Your mind will be really blown when you consider how far all people in the world walk in a week.
They do good marketing for sure. I would like to see same from Toyotas & Hondas of the world.
I have a 2018 Model 3 with basic autopilot. I have an FSD question. Does FSD have different driving behaviors from basic autopilot? I'm not saying "capabilities"; obviously it has more capabilities, but I mean the actual driving choices.
I get that FSD (maybe) has/requires better hardware than my car. But what I hate about autopilot is all around basic driving:
* Lane centering. It's extremely aggressive about lane centering, if you're in the right lane and an onramp joins from the right, the car aggressively drives to the right as soon as it perceives that the lane is wider.
* Throttle/brake behavior. It waits too long to brake (despite having radar in my car, which can supposedly "see" more than one car ahead), and when it does apply the brakes it doesn't do so smoothly. It tips in somewhat aggressively, and you can feel the discrete steps in brake force application change. Ditto for acceleration when the traffic in front of me moves.
There's no reason to think that any of this has anything to do with compute power, it all seems to be programming decisions that have been made for whatever reasons, so I can't see why FSD would be different.
And yet, if FSD drives like this, I don't get how anyone can think it's good? On the other hand, I've also heard people say they think autopilot is good, which it's clearly not, so it makes me judge their driving skill rather than the different models. But perhaps there's some matrix of hardware revisions and software/decision models out there that I'm unaware of, that explains differences in driving behavior, if they exist?
Basic autopilot is unrelated to FSD these days. Every aspect is improved.
Isn't there a trial you can try?
A 2018 vehicle is 2 major version behind on the FSD it can use. Tesla is on v14 and 2018 vehicles are stuck on v12.
There's been rumors of a v14 lite coming out because tesla REALLY doesn't want to deal with the fact that they promised the 2018s could be fully autonomous.
Interesting, thanks. I haven't followed the hardware revisions on my car very closely; my father has a 2017 S, and I'm so used to thinking I have the "newer revision", since my MCU was newer than his, and he only recently upgraded it. I guess it should have been fairly obvious that a lot has changed in the past 7 years, but I've been hearing a nonstop trickle of autopilot/FSD news on these cars for so long that I guess I just assumed I had recent-enough support.
Honestly, I just dismiss all of the promos they throw at me.
Just yesterday I got an ad in the app "refer a friend and try FSD (supervised), and get $15!"
So I opened up the app just now and I suppose I got my answer that proves my initial premise was incorrect -- I need a hardware upgrade for FSD. Womp womp.
(That said, it still seems like the lane centering and brake/throttle behavior should have been easily fixed without a FSD hardware upgrade).
FSD doesn't drive like that. It's a completely different software stack, not just Autopilot with extra capabilities.
It strikes me as incredibly weird to maintain two separate stacks, especially because FSD was "available" when I bought my car. In retrospect, though, I guess it wasn't actually available to use, just for purchase.
So I suppose they had a parallel development process for FSD while building autopilot features?
But since there is a hardware support overlap, it seems like at some point you'd migrate autopilot cars to the FSD software stack, with limitations added in via feature flags.
Autopilot basically hasn't gotten any new features since FSD came out.
> But since there is a hardware support overlap, it seems like at some point you'd migrate autopilot cars to the FSD software stack, with limitations added in via feature flags.
This is exactly what a lot of people speculate will happen soon.
All behaviors I've observed with FSD in my 2018 model 3.
It's possible v14 is better. v12 was certainly better than v11 in all those regards. But there are still issues with the car making dumb lane choices in v12.
So Tesla is charging $8000 to activate full safety software features in their vehicles?
How is this not way more controversial than having to pay extra to activate features like heated steering wheels in other brands?
You won't catch me defending Tesla often, but at least they actually do regular software updates and committed to needed hardware updates.
Ford and Chevy otoh are doing exactly the self sabotaging greedy bullshit you expect. Chevy already told previous gen super cruise owners that they are no longer getting updates or more mapped areas, and ford just segregated their line into before and after 2025. 2025 cars getting the latest bluecruise version, earlier cars don't have the hardware (but you still need to pay annualy for that deprecated older version!)
Ford also has never exteneded the mapped area in 4 years, and releases minor updates maybe once a year, where you will wait another 8 months to get the OTA.
All while having the nerve to charge $500/yr.
It is supposed to include hardware upgrades too. Chevy and others upcharge for semi autonomous safety features too, but I'm not sure to what extent that is pure software or added hardware.
No, only FSD is $8000. All of the safety features are free in every Tesla.
“Free” in what sense? After paying for a luxury product at premium prices, you’re getting exactly what you’re paying for, right?
In the sense that they're all always included in the base price of the vehicle, and never in optional packages that cost extra.
A $40K car is no longer a "luxury product at premium prices". That's basically Honda Accord prices.
Didn't you read the link? According to Tesla FSD is a safety feature
> So Tesla is charging $8000 to activate full safety software features in their vehicles?
I think you're being obtuse, but to be clear, many car manufacturers offer trims don't include features that would qualify making the car 'safer' - blind spot detection, back up cameras (I think these are legally required now but were a premium feature for over a decade), parking assistance, crash detection, etc.
I have a Tesla and use FSD every day, and while it is a safety feature, it is _the_ pinnacle 'luxury' feature that a car can have today and they honestly do not charge enough for it.
Well, I think that is also morally reprehensible in all other cases where it's also a matter of activating software safety features.
Most of the things you mentioned aren't software locked behind a paywall, hopefully, you don't swipe your credit card and get those features added via OTA in minutes. If your car doesn't have back seat airbags it's hopefully not because you haven't paid for the back seat airbag in-app purchase.
Okay you are correct. If they are going to market FSD as a safety feature and it's just a software update, they need to include it by default and adjust their price.
I didn't realize how much they market it as a safety feature.
7x fewer major and minor collisions!!!!*
* Compared to the estimated U.S. average.
They have a huge store of data on accidents in teslas per mile driven. Why don't they compare their actual data on accidents? Well, they would, but it probably is worse with FSD.
For the claims they are making the only peer is Waymo.
If they want to put themselves in a peer group that only has driver assistance systems, then the comparison should be to other similarly-priced and new vehicles, no U.S. average 12-year-old Corollas.
Anyway what is most amusing about this promotional material is that from the very first frame it inadvertently highlights how much worse it is than Waymo. The "Avoiding T-bones" scenario only seems like the car came out of nowhere because Tesla's camera system is so limited. Waymo would have seen it coming a mile away.
I think the parent is suggesting that they could compare the FSD accident rate to human-driven Teslas that they also have data on.
That is old news. Today we learn that Optimus will be a world class surgeon and "democratize" medical care, that Grok 5 has a 10% chance of achieving AGI, that Grokipedia will be named Encyclopedia Galactica and that Elon will put 100GW of solar powered "AI" satellites into orbit.
https://xcancel.com/XFreeze/status/1989453462770237688#m
https://xcancel.com/XFreeze/status/1989453462770237688#m
Doesn't matter I and almost everyone in my circle won't touch a Tesla.
The cybertruck was a huge boondoggle and only doctors and true beleivers bought them but over time I don't see them in the parking lot when I see the doctors in the building hmmm.
Also Elon "skipping around like a dipshit" (Tim Waltz) and now claiming he's worth a trillion dollars.
By his own admission his identity is entertwined with Tesla's.
He poisened the brand for me and many others.
Given how famously the tesla switches self driving off moments before an accident, these statistics are impossible to trust (but also because musk is famously a liar)
From the article:
> If FSD (Supervised) was active at any point within five seconds leading up to a collision event, Tesla considers the collision to have occurred with FSD (Supervised) engaged for purposes of calculating collision rates for the Vehicle Safety Report.
If it takes you off the road into the grass with no traction at 80mph it could reasonably be 5 seconds before you hit a tree.
Sure, it's very possible, and I'm sure has happened (an autopilot-instigated crash that happened >5sec after disengagement), but I do think 5 seconds is a reasonable threshold for this data, at the very least.
Pretty amazing to have driven six billion miles with 7x less major accidents. How many lives have been saved?
Tesla appears to have an impressive history of automatically disengaging FSD immediately before a collision, and this page is quite light on details of what they’re comparing to what, so I would take it with a large grain of salt.
> and this page is quite light on details of what they’re comparing to what
In the page:
"If FSD (Supervised) was active at any point within five seconds leading up to a collision event, Tesla considers the collision to have occurred with FSD (Supervised) engaged for purposes of calculating collision rates for the Vehicle Safety Report."
They are pretty open about how the stats are reported.
Who chooses to turn off FSD?
You really can't trust almost anything Musk says, he's proven this time and again, and Tesla will reflect that in its culture.
> Who chooses to turn off FSD?
The driver can at any time. And they do if it seems like it is going to do something stupid - which is getting rarer and rarer as time goes on. As a Level 2 system, the driver is always supposed to supervise the operation and stay alert.
Musk has proven time and again that things his critics say are impossible/unrealistic ends up being achieved late. See anything from reusing rocket stages to the goals from Tesla's 2018 Compensation Plan[1] ("If Mr. Musk were somehow to increase the value of Tesla to $650 billion — a figure many experts would contend is laughably impossible and would make Tesla one of the five largest companies in the United States ...")
Or the Arianespace guy saying SpaceX is "selling a dream". To quote: "I think a $5 million launch or a $15 million launch is a bit of a dream. Personally, I think reusability is a dream."
[1] https://www.nytimes.com/2018/01/23/business/dealbook/tesla-e...
[2] https://arstechnica.com/space/2024/06/some-european-launch-o...
For how many seconds after FSD is disengaged do accidents still get attributed to FSD?
True, we need to see the details of how they compute all this data. I do remember reading that they include crash data up to like a minute after FSD has been disengaged so these types of crashes should be included.
Nevertheless, the latest version of FSD I can certainly believe is seven times less likely to get into an accident than the average driver. I experienced it daily.
Are they saving lives? The latest NHTSA FARS analysis I saw (iSeeCars, report is through 2022) shows Tesla at 5.6 fatalities per billion miles vs the 2.8 average.
[flagged]
[flagged]
Or they have eyes and ears and have seen the footage.
You realize that almost every politician has done a similar type of pose right? So if there’s a picture or a video of your favorite politician doing that pose you’re automatically going to assume they are Nazi?
Just a minimal knowledge of Musk‘s ties to Israel, and friendships with Netanyahu, Jared Kushner, and others you should realize that there’s no way he’s a Nazi or antisemitic. If anything, he is highly racist against Muslims.
Yes, if I see a video (not a still) of someone doing a nazi salute two times in a row, I'm gonna assume they're a nazi. Absolutey no disagreements with you there.
Doubly so if they also happen to tune into an AfD and a Tommy Robinson rally.
Maybe we don’t have the same definition of what a nazi is, because Tommy Robinson is a huge Israeli schill, Elon musk is friends with Netanyahu, Jared kushner, and many other Zionists.
There is a difference between waving and doing a Nazi salute. There is a difference between seeing a single frame of a wave out of context and a full video where he does it twice.
Don't forget he promised robotaxis in 2019:
“Next year for sure, we will have over a million robotaxis on the road,” said Musk on October 21, 2019. “The fleet wakes up with an over-the-air update. That’s all it takes.”
https://www.thedrive.com/news/38129/elon-musk-promised-1-mil...
Yes, we all know that when Elon makes promises about the future, he's always late delivering on them. But this is about the capabilities of something that's already here, and those kind of claims are way more reliable.