Hacker News new | ask | show | jobs
by MiscIdeaMaker99 485 days ago
I've owned a Model 3 for years now, and FSD is scary as hell. We haven't paid for it -- and we won't -- but every time we get a free trial of it (mostly recently this past Fall), I give it a whirl, and I end up turning it off. Why? Because it does weird shit like slow down at an intersection with a green light. I don't feel like I can trust it, at all, and it makes me more anxious than just using standard auto-steer and cruise control (which still ghost breaks sometimes). I don't get why anyone uses FSD.
11 comments

Don't even get me started. Here's a list of things my model Y regularly does:

- Try to accelerate to 45mph in a parking lot b/c it was within 10ft of the road

- Decelerate from highway speeds suddenly to 30mph, as though it saw something it might hit (I stopped it at 30-ish and hit the gas)

- Decelerate to 50mph because of "emergency vehicles" even though there were no vehicles around (sometimes it mistakes lights that strobe b/c they are seen through median dividers as "emergency lights")

- Take up two lanes because they gradually separated and the car thinks it should stay evenly between the left and right divider line

- choose absolutely bonkers limits, like 30mph on two lane country highways.

- Stop on the highway with a big red screen and a message that says "Take control now fatal error"

- Not so much a problem any more, but when I was first getting used to it, it would beep a message at me, then scold me for looking at the message (and not the road), then ask me to do some kind of hand grip on the wheel to prove I'm paying attention, but I have to look at the message to figure out what it wants.

My wife tells me "Just keep your foot on the gas to keep up the speed and your hands on the wheel to keep it in line" and I am just left wondering what FSD is for

Drive from Memphis to Nashville in my “long range” Telsa 3 that has almost an amazing 160 mile range at 70 Mph. FSD would periodically do something crazy and then ask me why I disengaged, adding further excitement to my drive.

I am now absolutely convinced that we will have full self-driving from Tesla when we have a beautiful wall all the way from the east to the west coast along both the Mexican and Canadian borders. Both will be beautiful.

My tesla model 3 on "autopilot" (just keep speed) will ghost break if it sees some cars merging into an adjacent lane. Really dangerous, nobody expects a car decelerating from 130kph down to 50 on the nearly empty Autobahn. My previous car (VW) got that right. Overall it is a nice car (with a massive and increasing brand toxicity problem), but how can one trust the "full self driving" if it can't/doesn't even keep speed in supposedly trivial cases where the driver has control?
> what FSD is for

Hype & Marketing

Also boosting the stock price.

Musk has been lying about FSD for a decade and the lies boosted the stock price as intended:

https://motherfrunker.ca/fsd/

https://electrek.co/2024/08/24/tesla-deletes-its-blog-post-s...

It turns out lying works.

And it’s no secret he has been lying either. A few years ago Tesla’s senior most liaison to California’s road safety folks was on the record as saying hey we’re years away from true FSD, while Musk was at the very same time telling investors that they were going to crack it and deliver it in a year.
Eeesh. Yeah, I was afraid it might still be that bad after 10 years of between "this year" and "within three years".

At the time, 2016, I trusted their promotional video showing it driving hands-free; I'm not going to make the mistake of taking them at their word again after it was revealed to have not been as it appeared: https://www.businessinsider.com/tesla-faked-video-in-2016-pr...

> I am just left wondering what FSD is for

The vision and promise, or the actually demonstrated use case?

The demonstrated use case is to charge people more money for the same product.

The vision? That is exactly what Musk keeps saying: in principle, a self-driving car never gets tired or drunk, so it can be safer than the mean human even if it only operates at the level of the median human. And it wouldn't need to be limited to median human level, as the whole fleet could learn from every member, so gain experience a million times faster than any human.

But at this point, I'm sufficiently skeptical of all of this, that I think they (and everyone else) should be banned from direct observation of the entire fleet's cameras — it's a huge surveillance network operating on every public road and several private ones.

Why on earth are you using FSD in parking lots?
What else would the "Full" in "Full Self Driving" mean?
Why are you using full self driving as part of your driving?

Weird ask imo.

I hope you guys got rid of your teslas. Preferably under press. They are a danger to people around you.
They'd probably just replace it with another car, defeating the purpose.
Other cars don’t claim to offer FSD. The limitations are more explicit, therefore it’s safer for people around it.
If you were going to get rid of your Tesla for that reason, then you'd know about the limitations. Getting rid of it creates a chance for it to go to someone who did not.
Not based on their safety record.
Got any source for that, or anything to refute the reference below that shows that Teslas are actually the most dangerous in terms of fatal accidents/miles driven at twice the average?
Those are about the safety of people in the car. The comment saying they were dangerous was about people not in the car.
“The study's authors make clear that the results do not indicate Tesla vehicles are inherently unsafe or have design flaws. [...] “The models on this list likely reflect a combination of driver behavior and driving conditions, leading to increased crashes and fatalities,” iSeeCars executive analyst Karl Brauer said in the report.”

See also: https://www.iseecars.com/most-dangerous-cars-study#v=2024:~:...

It's also interesting in the context of this argument that they exclude models older than 2018.

I wonder if driver behavior is influenced by Tesla’s design of certain systems, in which case Tesla vehicles would be inherently unsafe and have design flaws to the extent that unsafe driver behaviors are enabled or encouraged.

“I wonder” is facetious here. Of course driver behaviors are influenced by Tesla marketing and design choices, and of course Teslas being the Most Dangerous Car Brand is an indication that Teslas are inherently unsafe and have design flaws.

Just like every other vehicle on the road.
It's for the share price. Just like it was for uber.
What’s amazing to me is that FSD to this day cannot recognize active school zones. Even my 6 year old Audi onboard cameras can do that. If you put FSD on and you go through a school zones, the Tesla will happily zip at full speed completely ignoring the school zone.
It's terrifying that companies are allowed to beta test buggy software out in the real world by shooting huge machines with sharp pointy corners through schools.
I hope this isn't too off-topic, but I'm always amazed at how many otherwise smart people hold the naive belief that FSD is remotely close because 99.9% of the time it works fine.

Self-driving in my opinion will require an AI that is, if not very close to, an AI capable of general intelligence.

Why?

Because in the real world to be able to drive a car as well as a human across all of the edge cases a human can you probably need something approaching general intelligence.

Humans understand that a person isn't just something with 4 limbs, but also can be that thing that looks like a white sheet with eyes by the side of the road on Oct 31st. And its these types of weird edge cases that humans instinctively understand because they have a deep world model to reason about which cannot be reasoned about by the narrow FSD AI systems we currently have.

When you think about what humans need to do when driving it's so much beyond just watching the road and turning a wheel that it seems almost absurd to imagine our current AI is anywhere near capable of handling all of the edge cases humans currently are.

And I also don't buy this argument that the goal should be to simply to reduce the total number of accidents per mile... I'd grant that it's very possible that FSD could reduce the total number of accidents per mile driven because most miles are driven in the much more narrow environment of highway driving. And here AI probably could do better job than a human on average when you factor into the equation human tiredness and distractibility. But no one is going to be comfortable with FSD occasionally plowing into a group of kids outside a school because statistically the total number of people who die in road traffic accidents is reduced on a per mile basis.

I'd be interested if anyone strongly disagrees.

I think you can implement self driving without general AI, but it has to be really defensive, err on the cautious side, and that means it can't travel faster that 10 to 25 kph, like a bicycle basically. That car will have a "safe zone" around it, monitored with radar and/or lidar, and if anything enters just outside that "safe zone", the car stops before hitting anything.

The market for such cars would be very limited IMO.

Totally agreed with the fist paragraph but after that you're ignoring the existence of Waymo today, which people generally feel comfortable in and around where they are. Elons marketing is so strong even the doubters feel like Tesla is the leader of the pack.
> But no one is going to be comfortable with FSD occasionally plowing into a group of kids outside a school because statistically the total number of people who die in road traffic accidents is reduced on a per mile basis.

I really don't see why not. Since those deaths must be counted too, if it still is safer with than in mind then it can't be something that happens even rarely.

But when are these edge cases, are they like the edge cases when I need 4 wheel drive? Basically never and if I did need it I just wouldn't go there or would rent a car for the day or something.
Most safety problems, including the child in a Halloween costume, can be solved by the advanced AI technique called "Don't hit anything. If it looks like you're about to hit something, slow down or stop."

Trouble is, when the company deliberately ties one hand behind its back by insisting on camera-only vision, it is never going to be perfect at not hitting stuff. Either multispectral imaging, radar, or lidar would help avoid edge cases like the Halloween costume. The camera might not even realize there's a three-dimensional object in front of it if there's snow on the ground. Stupid, stupid, stupid.

That "don't hit anything under any circumstances" is incompatible with USA-typical "stand our ground" and "right of way" philosophy of doing things.
> Trouble is, when the company deliberately ties one hand behind its back by insisting on camera-only vision, it is never going to be perfect at not hitting stuff.

I have a Y with FSD. I think that underplays it a little. Yes, they don't have LIDAR, but LIDAR is very expensive and fragile. I can understand that.

But they also removed the windscreen rain sensors in favour of using the cameras. Consequently you can be driving down the highway on a clear day, and the windscreen wipers will start. They also don't have ultrasonic distance sensors. Consequently the car won't warn you about short bollards at the corners of the car. It seems to be completely unaware of them.

Unlike LIDAR these sensors cost almost nothing, and these are high end cars. Keeping them until they have the camera version reliable would seem prudent. In fact they could have used them to improve the camera version, by say comparing what the rain sensor said to the camera output. It's seems like they've allowed an AI religion to cloud their engineering decisions.

It's so surprising to hear these issues with FSD and it makes me nervous even though I haven't encountered any problems in v13. I regularly use it back and forth between work and home and mostly rush hour with a lot of difficult merges and weird situations.
I suspect part of the difference is what kind of roads you are on. Whenever I'm in the Bay area (or Southern CA in general), I'm amazed by the quality of the roads. The pavement is even and smooth and the lines are crisp, fresh paint that is easy to see.

Meanwhile in the Midwest, we have potholes, uneven roads, sometimes roads with different surfaces mixed together (gray concrete with black asphalt patches). Lines are often badly worn by the weather and road salt and can be quite difficult to see.

I strongly suspect with no evidence that FSD likely has more problems on roads that are in poor condition.

I had to reread your comment multiple times because i couldn’t believe I was reading someone complimenting Bay Area and Socal roads.

I say that in the most innocent, sorta joking way possible considering I curse those very roads at every turn. Maybe I have to break out of the bubble!

I agree. I've used FSD v13 on a Model Y with hardware version 4 for a couple of months now. Checking my mileage, that's over 2,000 miles, most of which was with FSD enabled (road trips on interstates, backroads, two-lane country roads without lane markings, interstates, highways, etc.). It's been absolutely fantastic.

Even my parents and sister use FSD v13 regularly now in their Teslas.

It's come a long way from the early days when I first started testing it.

It makes me wonder how many people are using Autopilot (included as standard) instead of FSD on a newer Tesla with the new AI hardware?

It's pretty wild to be able to start from park. Tap a button, and go.

Just the other day, it managed merging onto the interstate and then immediately changing 7 lanes to the left to merge onto the next interstate exit heading north. It performed flawlessly.

I test drove a model 3 ~3 years ago and FSD was terrifying. I had no idea what it could or couldn’t do. IMO Hyundai (and others with similar features) have it perfect with adaptive cruise + active lane assist. I know exactly what it can do, it does 90% of the driving on long trips, and it doesn’t do so much that I’m tempted to put too much trust in it.
Prepare for Mayhem:

"Tesla Plans Robotaxis in Texas by June, in a State with No Regulations" - https://www.auto123.com/en/news/tesla-robotaxi-texas-regulat...

inb4 autonomous only zones in cities
It's crazy, because any negative criticism of FSD will have a ton of fanboys pouring out of the walls to tell you how great it is, how great the latest update is, how your anecdotal "evidence" is not typical, etc.

Except all you have to do is go try it and it becomes clear to any layperson that it's probably getting there but, and this is really crucial, it's not there yet.

I feel like FSD has been "getting there" since before even Tesla started marketing it. I remember Google's early self driving cars and everyone thought they were only a few years away from being practical.

I think FSD definitely has utility, but not in the hands of laypeople. There are still far too many edge cases that it just doesn't handle well, and your average person can't be trusted to stay alert and attentive while using a feature so heavily marketed as not needing either of these things.

> I remember Google's early self driving cars and everyone thought they were only a few years away from being practical.

...and that turned out to be a little optimistic, but they really are "there" now IMHO in San Francisco. I rode in one for the first time last year. Subjectively, I felt safer moving through San Francisco traffic in the self-driving car than I do when I'm driving there myself or when I'm being driven by a human in a Lyft. It was attentive, cautious, and smooth, and I got there in a reasonable time with no fuss. And crucially, I see a notable lack of stories about it making dangerous decisions, despite the total passenger miles.

Why is Waymo there now and not Tesla? I think a combination of factors, including: (a) the head start, (b) the willingness to use LIDAR and RADAR to overcome limitations, (c) the focus on self-driving (they design and operate self-driving systems; they don't manufacture electric cars), (d) the service model (easier problem to focus on a mapped region with good weather and monitor everything vs. sell a car expected to work anywhere/anytime without (as much?) telemetry), (e) frankly, caring more about safety and less about hype. Of those differences, the "head start" one is shrinking relatively speaking, but the others will likely remain significant enough that I don't expect to trust Tesla's systems any time soon.

Which is why the fan boys always tell you that it's the next version that will fix all the bugs.
It's the next version will fix all the bugs all the way down.
It's like C++ :-)
I was going to say Java…
Interestingly common trait in fan-boyism, the <thing> is always just a few steps away from being right.
Also, a fairly mundane reality of technological progression.
exactly... this is HN of course so I expect nothing less. my favorite is when I frequently (as early as couple of days ago) get comments like "go see some videos on youtube before commenting like that" :D soooo funny. the thing is absolute garbage but elon can sell garbage better than anyone that ever lived
I like that you've built a strawman for anyone who might disagree with you. "Oh there aren't any positive reviews they're just fanboys." You may as well write "Anyone who disagrees with me is wrong."
I don't get it, what do you expect them to do? just reinforce your view? the data is pretty clear how many people use FSD without issues. What's equally weird is you guys preemptively smearing folks who defend FSD. I don't get it, nothing will make you guys change your mind I guess.
For me, evidence would work. That's why California's regulations to report miles driven and # of accidents and disengagements is so nice. It's a standard to compare, measure and regulate. And when Tesla hides away from such and instead moves to unregulated Texas markets, it makes me predisposed to think they are shying away from gathering just the evidence that would convince me it's safe. If it works for you, great. But this difference makes me happy to live in California.
> the data is pretty clear how many people use FSD without issues.

What data? The data Tesla chooses to share with you?

It wouldn't surprise me to find out this incident had FSD disengage moments before colliding with the pole, thus continuing 100% FSD safe driving.

Who else can share the data Tesla is collecting??
1) Tesla has not actually shared real datasets, only cherry picked snippets and highlights.

2) This is only the data Tesla chooses to share, at times when Tesla chooses to share it.

They could have other auditors validate the data. They could release full datasets on regular schedules. They could do a lot of things, but they don't. They give little highlights of results without explaining methodologies. They pick seemingly random timeframes. They're not open with this information in the slightest.

> the data is pretty clear how many people use FSD without issues.

Serious question - what data? And who is supplying that? Tesla? And what is the emotional human equivalent for the level of confidence that we should assume is "safe"? 1% error rate? 0.5%? 0.00001%?

who should share the data Tesla is collecting? Tesla? or should some one steal their data and share it?
It’s a little hit or miss (pun intended). I’ve used it off and on in my model 3 for the past 2 years and the rate of improvement has been significant.

Still though it has quirks.

On long trips, I LOVE it. LOVE it. Being able to just tap in and relax, make phone calls, listen to an audiobook, etc is so nice. The first time I ever used it I had to leave early from the All Things Open conference in Raleigh because I was getting sick. Having it essentially drive me home for 5 hours when I wasn’t well, including stopping to charge, was a huge relief.

It’s also great in traffic jams where you’d otherwise be dealing with stop and go traffic until you get through it. Just tap in and relax til you’re on the other side.

Day to day driving, it’s a little more iffy. I’ve dealt with seemingly random slowdowns on otherwise empty roads. It feels odd especially because it’s sudden.

Early on it would have difficulty on roads without well marked lines too.

I’ve never felt like it was going to run into an object though. Usually it errs on the “too cautious” side and I just take over to get where I’m going quicker.

> Being able to just tap in and relax, make phone calls, listen to an audiobook, etc is so nice.

My 12 year old Ford Focus does that

Not while it drives you hours to your destination without your intervention…
The transmission still works?
Its a manual, and works fine
I always slow down a little bit when speeding through intersections, just in case I need to react to someone illegally putting themselves in my path.

It was this guys fault for not monitoring the car, but also Tesla's for using a double-speak name like Full Self Driving.

If FSD is a statistically significant enough risk factor for injury above Teslas that don't use it, it should be banned.

Well, FSD it is officially in beta. Supposedly it would actually be safe once it reaches release; Which decade that might be, your guess is as good as mine.
I'm on AI4 v13 and havent had a safety intervention in several thousand miles. It's incredible and extremely smooth
While we're sharing anecdotes, I have FSD (13 now) on my model Y and love it. I was anxious at first and remain very guarded while using it (which you're obligated to do anyway) but it's taken away a lot of the tedium and fatigue of commuting and long highway driving. I occasionally use it door-to-door but often turn it off on certain roads, where I'll do a better job avoiding potholes, for example. It's not done anything unsafe, but it did change lanes once without signaling and I had to intervene. No one was around and the lines were hard to see, so perhaps that's why. Overall it feels like a far safer driver than a lot of people I've been in the car with.
I feel this way too. It was scary the first couple weeks, but I'm glad I gave it more of a chance. Over time you learn when to trust it and what situations it you need to take over, and then it just becomes a normal part of driving with less stress on me while my brain energy is spent looking for problems instead of keeping up with traffic or in the lane.

Really a human + AI hybrid experience.

I too have FSD 13 on my CT and use it for 99% of driving with no issues. I have done a number of long city to city drives (30-100+ miles) with zero interventions. The roads would be 10,000x safer if every car was using FSD, even in extreme edge cases like the original post.
How do you like the CT? Is it your first Tesla?
It’s fantastic and Tesla service has been really good for the one issue I had over the last 12 months. My wife bought a Model Y in 2021 so it’s the second Tesla in the house.
There are upsides, definitely. For slow moving traffic FSD can remove so much of the tedium of matching speed, spacing, or stop-and-go.

I like the level 1 to level 3 features: Lane keeping, emergency braking (when there's something there), adaptive speed control, etc. But a new minivan has all those too.

For long highway driving it does remove 99% of the things I hate, but there's 1% of the time it just annoys the hell out of me, and it tarnishes the whole experience.

> While we're sharing anecdotes, I have FSD (13 now) on my model Y and love it.

At what ratio of good anecdotes to bad anecdotes should we trust it? For me, the ratio has to be astonishingly high, such that if there are a few people in the discussion saying it did something suspect (much less dangerous), they're always going to be the ones I listen to. Not that I'm doubting your experience; it's just not enough to outweigh the other.

Obligatory xkcd: https://xkcd.com/937/

I have no horse in this race at all, but people seem less likely to tell good anecdotes. There's little interesting in "I used it, it worked adequately."
I guess proper statistics are the thing.