Ultimately, anecdotes and testimonials of a product like this are irrelevant. But the public discourse hasn't caught up with it. People talk about it like it's a new game console or app, giving their positive or negative testimonials, as if this is the correct way to validate the product.
Only rigorous, continual, third party validation that the system is effective and safe would be relevant. It should be evaluated more like a medical treatment.
This gets especially relevant when it gets into an intermediate regime where it can go 10,000 miles without a catastrophic incident. At that level of reliability you can find lots of people who claim "it's driven me around for 2 years without any problem, what are you complaining about?"
10,000 mile per incident fault rate is actually catastrophic. That means the average driver has a serious, life threatening incident every year at an average driving rate. That would be a public safety crisis.
We run into the problem again in the 100,000 mile per incident range. This is still not safe. Yet, that's reliable enough where you can find many people who can potentially get lucky and live their whole life and not see the system cause a catastrophic incident. Yet, it's still 2-5x worse than the average driver.
> Only rigorous, continual, third party validation that the system is effective and safe would be relevant. It should be evaluated more like a medical treatment.
100% agreed, and I'll take it one step further - level 3 should be outright banned/illegal.
The reason is it allows blame shifting exactly as what is happening right now. Drivers mentally expected level 4 and legally the company will position the fault, in as much as it can get away with, to be on the driver, effectively level 2.
They're building on a false premise that human equivalent performance using cameras is acceptable.
That's the whole point of AI - when you can think really fast, the world is really slow. You simulate things. Even with lifetimes of data, the cars still will fail in visual scenarios where error bars on ground truth shoot through the roof.
Elon seems to believe his cars will fail in similar ways to humans because they use cameras. False premise. As Waymo scales, human just isn't good enough, except for humans.
It can be misleading to directly compare disengagements to actual catastrophic incidents.
The human collision numbers only count actual incidents, and even then only ones which have been reported to insurance/authorities. It doesn't include many minor incidents such as hitting a bollard, or curb rash, or bump-and-run incidents in car parks, and even vehicle-on-vehicle incidents when both parties agree to settle privately. And the number certainly excludes ALL unacceptably close near-misses. There's no good numbers for any of these, but I'd be shocked if minor incidents weren't an of magnitude more common, and near misses another order of magnitude again.
Whereas an FSD disengagement could merely represent the driver's (very reasonable) unwillingness to see if the software will avoid the incident itself. Some disengagements don't represent a safety risk at all, such as when the software is being overly cautious, e.g. at a busy crosswalk. Some disengagements for sure were to avoid a bad situation, though many of these would have been non-catastrophic (such as curbing a wheel) and not a collision which would be included in any human driver collision statistics.
As a robotaxi, yes. That's why Teslas rollout is relatively small/slow, has safety monitors, etc...
FSD, what most people use, is ADAS, even if it performs a lot of the driving tasks in many situations, and the driver needs to always be monitoring it, no exceptions.
The same applies to any ADAS. If it doesn't work for in a situation, the driver has to take over.
If there was actually a rate of one life threatening accident per 10,000 miles with FSD that would be so obvious it would be impossible to hide. So I have to conclude the cars are actually much safer than that.
FSD never drives alone. It's always supervised by another driver legally responsible to correct. More importantly we have no independently verified data about the self driving incidents. Quite the opposite, Tesla repeatedly obscured data or impeded investigations.
I've made this comparison before but student drivers under instructor supervision (with secondary controls) also rarely crash. Are they the best drivers?
I am not a plane pilot but I flew a plane many times while supervised by the pilot. Never took off, never landed, but also never crashed. Am I better than a real pilot or even in any way a competent one?
Above I was talking more generally about full autonomy. I agree the combined human + fsd system can be at least as safe as a human driver, perhaps more, if you have a good driver. As a frequent user of FSD, it's unreliability can be a feature, it constantly reminds me it can't be fully trusted, so I shadow drive and pay full attention. So it's like having a second pair of eyes on the road.
I worry that when it gets to 10,000 mile per incident reliability that it's going to be hard to remind myself I need to pay attention. At which point it becomes a de facto unsupervised system and its reliability falls to that of the autonomous system, rather than the reliability of human + autonomy, an enormous gap.
Of course, I could be wrong. Which is why we need some trusted third party validation of these ideas.
As time goes on, Tesla's fiasco becomes more and more embarrassing. Waymos are all over the place in the cities they serve, doing pretty much what they're supposed to do. Nuro has some fully autonomous vehicles running around. Baidu's Apollo Go is deployed in 16 cities in China, although they use remote driving as a backup.
Tesla, though, is still hyping a technology that seems to have maxed out years ago.
Not sure if it’s evident in broader statistics yet, but I think that because Tesla got the early adopter market (tech savvy people), they are now losing that same market first.
I had a party at my house a couple months ago, mostly SF tech people. I found the Tesla owners chatting together, and the topic was how much FSD sucks and they don’t trust it.
I asked and no-one said they would buy a Tesla again. Distrust because they felt suckered by FSD was a reason, but several also just said Elon’s behavior was a big negative.
I own six EVs (three cars, one of which is a Tesla, and three motorcycles). My first EV was my Tesla.
We're on the cusp of trading the Tesla in for a Rivian most likely. I should be Tesla's target customer, but instead I'm exactly who you described:
- I don't like the brand. I don't like Elon. I don't like the reputation that the car attaches to me.
- I don't trust the technology. I've gotten two FSD trials, both scared the shit out of me, and I'll never try it again.
- I don't see any compelling developments with Tesla that make me want to buy another. Almost nothing has changed or gotten better in any way that affects me in the last four years.
They should be panicking. The Cybertruck could have been cool, but they managed to turn it into an embarrassment. There are so many alternatives now that are really quite good, and Tesla has spent the last half a decade diddling around with nonsense like the robot and the semi and the Cybertruck and the vaporware roadster instead of making cars for real people that love cars.
I partially agree. FSD seems fine to me but I wouldn’t buy a second Tesla. Tesla seems to have stopped caring about being a car company that caters to nerds/tech enthusiasts.
Mine has been an extremely well done vehicle and I was (and kind of am) bullish on FSD as a driver assistance technology, but a car is a 6-7 year investment for me and I have big doubts about their direction. They seem to have abandoned the idea of being a car company, instead chasing this robotaxi idea.
Up until 2023/2024 was fine for my 6-7 year car lifecycle. Tesla was really cool when they let you do all sorts of backwards-compatible upgrades, but they seemed to have abandoned that.
I’ve found it incredibly disappointing seeing their flailing direction now.
Rivian seems to still have a lot of the magic that Tesla had. They’re definitely a strong contender for my next vehicle in a year or two.
I just had a thought - a Waymo car costs $200k (maybe more) from a quick google search. YoY returns of $200k on S&P are about 10%, while an Uber driver takes home about $40-$50k - so in terms of cost, they are about 2x-2.5x of each other, with the Waymore likely needing expensive maintenance/support infrastructure, bringing the total much closer.
Which means if Tesla can really build that Cybercab - with an underpowered motor, small battery, plastic body panels, just cameras (which I think they promised to sell under $20k) - they'll be able to hit a business expense level and profitability that Waymo will only be able, in say, 10 years.
Even if you don't want to talk about non-existing hardware, a Model 3's manufacturing cost is surely much lower than a Waymo.
Once (if) they make self driving work at any point in time before Waymo gets to the same level of cost - they'll be the more profitable business.
Not only that, they'll be able to enter markets where the cost of Waymo and what you can charge for taxi rides is so far apart that it doesn't make sense for them - in this sense, they'll have a first mover advantage.
Waymo cars are basically priceless at this point. As in: the car cost doesn't matter. They've so far spent multiple times their fleet's costs on R&D. The fact that they're getting some pocket cash from paid fares is inconsequential for their bottom line.
Any realistic mass deployment will use cheaper cars, more suitable for taxi service.
Not to mention, Waymo is moving from Jaguar to Zeekr for its next-gen fleet, meaning 100% import tariffs on those Chinese-built base vehicles before it even begins the expensive retrofit process.
The core problem with Waymo’s model is its lack of (economically rational) scalability. Shipping finished vehicles to a second facility to strip and rebuild them as robotaxis is inherently inefficient, and cannot be made efficient by scaling up. To achieve meaningful volume, Waymo would need to contract an automaker to build finished robotaxis, ideally domestically or where tariffs are sufficiently low.
Obviously Tesla's solution only works if their vision-only strategy bears fruit. Assuming it does (a wildly controversial assumption in this space, but let's go with it for now) the economics are utterly wild. It's difficult to imagine how any competitor could come close to competing on cost or scale. And that's assuming the Model Y, ignoring the as-yet hypothetical Cybercab.
I suppose Alphabet could buy the corpse of Canoo. I suspect that if it had a plausible manufacturing ramp, they would have been snapped up quickly. Automotive-scale manufacturing is a crucible, and it destroys most who attempt it. In fact most die long before they begin frfr.
Tesla's FSD has indeed made significant improvements in the past year (still way behind where it was promised to be even half a decade ago), but they are FAR from being able to operate an actual robotaxi service. Austin is an embarrassment. It seems that tesla believes they can make more money on fooling investors than they can on any core business model
Sorry but while I am a happy Waymo user, this is overblown and just incorrect. Waymo has datacenter oversight. (which, who cares, the product is great).
Tesla FSD 12 -> 13 was a massive jump that happened earlier this year. 14 is still rolling out.
Testing out 13 this weekend, it drove on country roads, noticed a road closure, rerouted, did 7 miles on non divided highways, navigated a town and chose a parking space and parked in it with zero interruptions. It even backed out of the driveway to start the trip. I didn't like the parking job and reparked; other than that, no hands, no feet, fully autonomous. Unlike 12, I had no 'comments' on the driving choices - rotaries were navigated very nicely, road hazards were approached and dealt with safely and properly. It was genuinely good.
Dislike Elon all you want, but Tesla FSD is improving rapidly, and to my experienced eyes adding 9s. Probably two or three more 9s to go, but it's not a maxed out technology.
You lack data to draw this conclusion. The most important factor is deaths per mile, which is sparse, so it requires aggregating data from many drivers before you have enough statistical power.
I can imagine it doing fine on highways for a thousand miles. FSD has literally never managed to complete a trip involving city driving for me without disengaging or me having to stop it from doing something illegal. I'm not sure how many attempts I'm supposed to give it. Hell, autopilot doesn't even manage to consistently stay safely in a lane on I-40 between Durham and Asheville.
This scam is true for all AI technologies. It only "works" as far as we interpret it as working. LLMs generate text. If it answers our question, we say that the LLM works. If it doesn't, we say that it is "hallucinating".
Im sorta beginning to think some LLM/AI stuff is the Wizard of Oz(a fake it before you make it facade).
Like why can an LLM create a nicely designed website for me but asking it to do edits and changes to the design is a complete joke. Lots of the time it creates another brand new design (not what i asked all) and it's attempts at editing it LOL. It makes me think it does no design at all rather it just went and grab one from the ethers of the Internet acting like it created it.
It's not a scam because it does make you code faster even if you must review everything and possibly correct (either manually or via instruction) some things.
As far as hallucinations go, it is useful as long as its reliability is above a certain (high) percentage.
I hear you, but GenAI also gets the opposite fork from people who hate it: It's good result that used GenAI at any point => your prompting and curation and editing is worthless and deserves no credit; it's not good result => that proves AI isn't real intelligence.
As with Marmite, I find it very strange to be surrounded by a very big loud cultural divide where I am firmly in the middle.
Unlike Marmite, I wonder if I'm only in "the middle" because of the extremities on both ends…
> Would at least like to use it for stop and go traffic, which is about the only thing I trust FSD for.
Depends on the type of stop and go driving. Crawling along at 15mph, sure. But the most dangerous driving scenario - whether human or machine is the driver - is a scenario with large variations in speed between vehicles and also limited visibility.
For example suddenly encountering a traffic jam that starts around a blind corner.
That's also the most tiresome part of driving and has the least risk due to low speeds. Easy win for FSD. But for all other cases it becomes a complicated ethical question.
> So, for example, when a Florida driver on Autopilot drops his phone and blows through a stop sign, hitting a car which then hits two pedestrians, killing one, Tesla will claim “this driver was solely at fault.” In that case, a judge agreed that the driver was mostly at fault, but still assigned 33% of blame to Tesla, resulting in a $243 million judgment against the company.
His foot was on the gas though
Looking at this author's other articles, he seems more than a bit unhinged when it comes to Tesla: https://electrek.co/author/jamesondow/ Has Hacker News fallen for clickbait? (Don't answer)
A couple of facts on the Florida case: it was a jury verdict, not a judge. The jury found Tesla 33% at fault for a 2019 Key Largo crash. Damages were $129M compensatory (Tesla responsible for 33% of that) plus $200M punitive, for $243M total.
The driver admitted he looked down after dropping his phone and blew a stop sign; Tesla argues his foot was on the accelerator, but the jury still assigned partial fault because Autopilot was allowed to operate off limited-access highways and the company didn’t do enough to prevent foreseeable misuse. The driver had already settled separately.
is there any blame to be associated to Tesla for its feature? what's the right percent for you? 20%? 10%? 5%? 0%?
If the wheels of the car fell off, whould Tesla have any blame for that? If we had laid wires all along the road to allow for automatic driving, and Tesla's software misread that and caused a crash, would it be to blame?
When is Autopilot safe to use? Is it ever safe to use? Is the fact that people seem to be able to entirely trick the Autopilot to ignore safety attention mechanisms relevant at all?
If we have percentage-based blame then it feels perfectly fine to share the blame here. People buy cars assuming that the features of the car are safe to use to some extent or another.
Maybe it is just 0%. Like cruise control is a thing that exists, right? But I'm not activating cruise control anywhere near any intersection. Tesla calls their thing autopilot, and their other thing FSD, right? Is there nothing there? Maybe there is no blame, but it feels like there's something there.
0%. This is entirely on the driver. He's someone who should spend a few years in prison, and then never be allowed to have a license again.
A foot on the gas overrides braking on autopilot and causes it to flash up a large message up on the screen that "Autopilot will not break / Accelerator pedal is pressed"
They don’t even work well in California. Driving into sun? Car loses all lane visibility and yells at you. Driving late at night? Can’t detect lanes and complains about blinded cameras.
If only there was some kinda technology that didn’t rely on optics that could see in pitch dark or when the sun is shining.
Next up: they put LCD layer with selective dimming over cameras and claim world's first instead of investing in lidar/radar as they should from the start rather than dropping it "coz it is not needed"
According to users, that issue appears to be solved as of FSD v13. The solution may be reliant on the higher quality camera modules shipped with hardware 4.
Technically you kind of get this in Nevada when using Tesla insurance and if you drive 100 % FSD. If you drive manually you are pretty much doxed for random Front collision Warning which is super sensitive
Having driven Tesla FSD and coded with Claude/Codex, it suffers from the exact same issues- Stellar performance in most common contexts, but bizarrely nonsensical behavior sometimes when not.
Which is why I call it "thunking" (clunky thinking) instead of "thinking". And also why it STILL needs constant monitoring by an expert.
Readit News
Only rigorous, continual, third party validation that the system is effective and safe would be relevant. It should be evaluated more like a medical treatment.
This gets especially relevant when it gets into an intermediate regime where it can go 10,000 miles without a catastrophic incident. At that level of reliability you can find lots of people who claim "it's driven me around for 2 years without any problem, what are you complaining about?"
10,000 mile per incident fault rate is actually catastrophic. That means the average driver has a serious, life threatening incident every year at an average driving rate. That would be a public safety crisis.
We run into the problem again in the 100,000 mile per incident range. This is still not safe. Yet, that's reliable enough where you can find many people who can potentially get lucky and live their whole life and not see the system cause a catastrophic incident. Yet, it's still 2-5x worse than the average driver.
100% agreed, and I'll take it one step further - level 3 should be outright banned/illegal.
The reason is it allows blame shifting exactly as what is happening right now. Drivers mentally expected level 4 and legally the company will position the fault, in as much as it can get away with, to be on the driver, effectively level 2.
The human collision numbers only count actual incidents, and even then only ones which have been reported to insurance/authorities. It doesn't include many minor incidents such as hitting a bollard, or curb rash, or bump-and-run incidents in car parks, and even vehicle-on-vehicle incidents when both parties agree to settle privately. And the number certainly excludes ALL unacceptably close near-misses. There's no good numbers for any of these, but I'd be shocked if minor incidents weren't an of magnitude more common, and near misses another order of magnitude again.
Whereas an FSD disengagement could merely represent the driver's (very reasonable) unwillingness to see if the software will avoid the incident itself. Some disengagements don't represent a safety risk at all, such as when the software is being overly cautious, e.g. at a busy crosswalk. Some disengagements for sure were to avoid a bad situation, though many of these would have been non-catastrophic (such as curbing a wheel) and not a collision which would be included in any human driver collision statistics.
FSD, what most people use, is ADAS, even if it performs a lot of the driving tasks in many situations, and the driver needs to always be monitoring it, no exceptions.
The same applies to any ADAS. If it doesn't work for in a situation, the driver has to take over.
I've made this comparison before but student drivers under instructor supervision (with secondary controls) also rarely crash. Are they the best drivers?
I am not a plane pilot but I flew a plane many times while supervised by the pilot. Never took off, never landed, but also never crashed. Am I better than a real pilot or even in any way a competent one?
I worry that when it gets to 10,000 mile per incident reliability that it's going to be hard to remind myself I need to pay attention. At which point it becomes a de facto unsupervised system and its reliability falls to that of the autonomous system, rather than the reliability of human + autonomy, an enormous gap.
Of course, I could be wrong. Which is why we need some trusted third party validation of these ideas.
Dead Comment
Tesla, though, is still hyping a technology that seems to have maxed out years ago.
I had a party at my house a couple months ago, mostly SF tech people. I found the Tesla owners chatting together, and the topic was how much FSD sucks and they don’t trust it.
I asked and no-one said they would buy a Tesla again. Distrust because they felt suckered by FSD was a reason, but several also just said Elon’s behavior was a big negative.
We're on the cusp of trading the Tesla in for a Rivian most likely. I should be Tesla's target customer, but instead I'm exactly who you described:
- I don't like the brand. I don't like Elon. I don't like the reputation that the car attaches to me.
- I don't trust the technology. I've gotten two FSD trials, both scared the shit out of me, and I'll never try it again.
- I don't see any compelling developments with Tesla that make me want to buy another. Almost nothing has changed or gotten better in any way that affects me in the last four years.
They should be panicking. The Cybertruck could have been cool, but they managed to turn it into an embarrassment. There are so many alternatives now that are really quite good, and Tesla has spent the last half a decade diddling around with nonsense like the robot and the semi and the Cybertruck and the vaporware roadster instead of making cars for real people that love cars.
Mine has been an extremely well done vehicle and I was (and kind of am) bullish on FSD as a driver assistance technology, but a car is a 6-7 year investment for me and I have big doubts about their direction. They seem to have abandoned the idea of being a car company, instead chasing this robotaxi idea.
Up until 2023/2024 was fine for my 6-7 year car lifecycle. Tesla was really cool when they let you do all sorts of backwards-compatible upgrades, but they seemed to have abandoned that.
I’ve found it incredibly disappointing seeing their flailing direction now.
Rivian seems to still have a lot of the magic that Tesla had. They’re definitely a strong contender for my next vehicle in a year or two.
Which means if Tesla can really build that Cybercab - with an underpowered motor, small battery, plastic body panels, just cameras (which I think they promised to sell under $20k) - they'll be able to hit a business expense level and profitability that Waymo will only be able, in say, 10 years.
Even if you don't want to talk about non-existing hardware, a Model 3's manufacturing cost is surely much lower than a Waymo.
Once (if) they make self driving work at any point in time before Waymo gets to the same level of cost - they'll be the more profitable business.
Not only that, they'll be able to enter markets where the cost of Waymo and what you can charge for taxi rides is so far apart that it doesn't make sense for them - in this sense, they'll have a first mover advantage.
Any realistic mass deployment will use cheaper cars, more suitable for taxi service.
The core problem with Waymo’s model is its lack of (economically rational) scalability. Shipping finished vehicles to a second facility to strip and rebuild them as robotaxis is inherently inefficient, and cannot be made efficient by scaling up. To achieve meaningful volume, Waymo would need to contract an automaker to build finished robotaxis, ideally domestically or where tariffs are sufficiently low.
Obviously Tesla's solution only works if their vision-only strategy bears fruit. Assuming it does (a wildly controversial assumption in this space, but let's go with it for now) the economics are utterly wild. It's difficult to imagine how any competitor could come close to competing on cost or scale. And that's assuming the Model Y, ignoring the as-yet hypothetical Cybercab.
I suppose Alphabet could buy the corpse of Canoo. I suspect that if it had a plausible manufacturing ramp, they would have been snapped up quickly. Automotive-scale manufacturing is a crucible, and it destroys most who attempt it. In fact most die long before they begin frfr.
Tesla FSD 12 -> 13 was a massive jump that happened earlier this year. 14 is still rolling out.
Testing out 13 this weekend, it drove on country roads, noticed a road closure, rerouted, did 7 miles on non divided highways, navigated a town and chose a parking space and parked in it with zero interruptions. It even backed out of the driveway to start the trip. I didn't like the parking job and reparked; other than that, no hands, no feet, fully autonomous. Unlike 12, I had no 'comments' on the driving choices - rotaries were navigated very nicely, road hazards were approached and dealt with safely and properly. It was genuinely good.
Dislike Elon all you want, but Tesla FSD is improving rapidly, and to my experienced eyes adding 9s. Probably two or three more 9s to go, but it's not a maxed out technology.
Dead Comment
Like why can an LLM create a nicely designed website for me but asking it to do edits and changes to the design is a complete joke. Lots of the time it creates another brand new design (not what i asked all) and it's attempts at editing it LOL. It makes me think it does no design at all rather it just went and grab one from the ethers of the Internet acting like it created it.
As far as hallucinations go, it is useful as long as its reliability is above a certain (high) percentage.
I actually tried to come up with a "perceived utility" function as a function of reliability: U(r)=Umax ⋅e^(−k(100−r)^n) with k=0.025 and n=1.5 is the best I came up with, plotted here: https://imgur.com/gallery/reliability-utility-function-u-r-u...
Wasted 1 hour each of your 5 co-workers who ended up reviewing unusable slop? Silence.
As with Marmite, I find it very strange to be surrounded by a very big loud cultural divide where I am firmly in the middle.
Unlike Marmite, I wonder if I'm only in "the middle" because of the extremities on both ends…
The family of the first person killed by that will know who to sue for a punitive trillion dollars.
Depends on the type of stop and go driving. Crawling along at 15mph, sure. But the most dangerous driving scenario - whether human or machine is the driver - is a scenario with large variations in speed between vehicles and also limited visibility.
For example suddenly encountering a traffic jam that starts around a blind corner.
His foot was on the gas though
Looking at this author's other articles, he seems more than a bit unhinged when it comes to Tesla: https://electrek.co/author/jamesondow/ Has Hacker News fallen for clickbait? (Don't answer)
The driver admitted he looked down after dropping his phone and blew a stop sign; Tesla argues his foot was on the accelerator, but the jury still assigned partial fault because Autopilot was allowed to operate off limited-access highways and the company didn’t do enough to prevent foreseeable misuse. The driver had already settled separately.
If the wheels of the car fell off, whould Tesla have any blame for that? If we had laid wires all along the road to allow for automatic driving, and Tesla's software misread that and caused a crash, would it be to blame?
When is Autopilot safe to use? Is it ever safe to use? Is the fact that people seem to be able to entirely trick the Autopilot to ignore safety attention mechanisms relevant at all?
If we have percentage-based blame then it feels perfectly fine to share the blame here. People buy cars assuming that the features of the car are safe to use to some extent or another.
Maybe it is just 0%. Like cruise control is a thing that exists, right? But I'm not activating cruise control anywhere near any intersection. Tesla calls their thing autopilot, and their other thing FSD, right? Is there nothing there? Maybe there is no blame, but it feels like there's something there.
A foot on the gas overrides braking on autopilot and causes it to flash up a large message up on the screen that "Autopilot will not break / Accelerator pedal is pressed"
It can't be healthy to be so obsessed with something/someone you dislike.
Drivers have tackled this problem by wearing polaroid sunglasses.
I really hope someone asks Tesla how they plan to solve the Sun glare issue.
If only there was some kinda technology that didn’t rely on optics that could see in pitch dark or when the sun is shining.
Why was that OK? Why was it safe to let people use like that without informing them?
Having driven Tesla FSD and coded with Claude/Codex, it suffers from the exact same issues- Stellar performance in most common contexts, but bizarrely nonsensical behavior sometimes when not.
Which is why I call it "thunking" (clunky thinking) instead of "thinking". And also why it STILL needs constant monitoring by an expert.