4 ms·

My Tesla still detects about 90% of garbage bins on our street, but only about 60% of the school kids crossing the road (I live in Germany where kids walk to sc

My Tesla still detects about 90% of garbage bins on our street, but only about 60% of the school kids crossing the road (I live in Germany where kids walk to school). The rest it would kill. As I pass by that school daily on my way to work, my Tesla would probably kill about 10-20 kids per week. Yeah, good idea to hide the crash data.

mgoetzke 1y ago
You base that assumption on the visualizations in the center I guess ? They are not actually everything the car sees and reacts to. Especially not in our German FSD cars
- Fischgericht 1y ago
  I am aware that FSD has a different software stack. But it's the same hardware. So why would they make the detection of kids different on the standard firmware artificially worse? As Marketing for people who hate school kids? I find it laughable that there still are Musk fanboys who after a decade of lies about this still believe in "Robotaxis". 90% of them clearly have never tried to drive a Tesla in a scenario where the minimal protections for kids to use public street space is not "kids should get a SUV to not get killed". It is also amusing to watch videos of Tesla fanboys on YouTube who proudly show that their Tesla now can use FSD for up to 500 miles without a single crash (or "critical disengagement)". A human driver statistically causes a crash every 500,000 miles. But yes, we will have flying Robotaxis in 2 weeks from now, that will solve this problem. Musk said so. :)
  - robertlagrant 1y ago
    Other than the first paragraph, this all seems to be replying to something else?
  - yreg 1y ago
    > I am aware that FSD has a different software stack. But it's the same hardware. So why would they make the detection of kids different on the standard firmware artificially worse? As Marketing for people who hate school kids? Not sure what's your argument here. The visualization you get using "Enhanced Autopilot" is completely different to the one you get using "FSD Beta" because the software you are running is completely different as well.
    - Fischgericht 1y ago
      The point is not the visualization towards the driver. It's that the same data clearly is the base for the decisions this car makes. If it is not showing the kid crossing the street, you also will not get an emergency break warning, which I get in tons of other situations. What you see is what you get.
      - yreg 1y ago
        I'm sorry, I struggle to understand your point. Which one is it? - Do you think that a Tesla with an enhanced Autopilot would hit a kid because you don't see it in the visualization? - Or do you think that a Tesla with FSD Beta would hit a kid because it uses "the same data" as the one without it?
        lechatonnoir 1y ago
        I think the idea is, why would the visualization be so intentionally bad in the Autopilot version as to not detect the kids entirely? What benefit does that confer, or, from another perspective, what software constraint forces this to be the case?
        Fischgericht 1y ago
        Exactly that, yes. Thank you.
        yreg 1y ago
        It's not _intentionally bad_. Autopilot and FSD are _different products_ with _different tech_. It's not like they could simply copy `detect_children()` from FSD to Autopilot and call it a day.
        lechatonnoir 1y ago
        Hmm, that might be possible, but that's essentially not what I assumed. At the very least, they operate on the same hardware, so Autopilot is in some sense "intentionally bad" as a whole.
  - infecto 1y ago
    Not defending Musk, I don’t like him but I am not sure why you would think two separate software stacks should somehow be comparable. Maybe it’s my old age but I get tired of these style of rants where folks are fixated on a single thing.
fundatus 1y ago
I also find the opposite hilarious: The amount of things that Teslas detect as trash cans is absurd
- Sohcahtoa82 1y ago
  My favorite is that it identified my wife's Honda CRV as a trash can.
bell-cot 1y ago
> only about 60% of the school kids [...] The rest it would kill. [...] would probably kill about 10-20 kids per week. I'm no Tesla fan - but it would be real-world obvious if even 0.1% of Teslas actually were that "eager" to kill children. In most western countries, covering up child-killing accidents scales very poorly.
- avtolik 1y ago
  Well, we don't have FSD in Europe, and in US, I guess the children don't walk to school.
  - bell-cot 1y ago
    US kids walk to school far less than in the Good Old Days...but there's still a fair amount of walking. And on low-traffic residential streets, there can be quite a bit of de facto playing in the street. So it's still a "passably" target-rich environment for killer robocars.
  - usea 1y ago
    In the US, letting your children walk to school is taking a non-negligible risk that you'll be charged with a crime or have your children taken away. Their deaths from a motor vehicle are assumed by all to be a certain eventuality, and parents are more likely to be blamed for it than drivers.
thomastjeffery 1y ago
The worrying part is that if/when those percentages get better, you will be more likely to trust it enough to let it run over children.
- zelphirkalt 1y ago
  Soon promised to only have 1/10 of detection failures, better than ever before! Only 1 child per week! Rejoice! On a more serious note: Where do we as a society put the bar? What are the numbers, at which we accept the risk? Do we put the bar higher than for humans? Or same level? Or does the added convenience for car drivers tempt us to accept a lower bar?
  - thomastjeffery 1y ago
    Setting a bar is the mistake. We need to reframe the entire narrative. Safety implementation is never objective. You can only implement a system by subjecting it to context. Traffic safety is a world of edge cases, and each driving implementation will engage with those edge cases from a different subjective context. We are used to framing computation as a system of rules: explicit logic that is predictably followed. Tesla is using the other approach to "AI": statistical models. A statistical model replaces binary logic with a system of bias. A model that is built out of good example data will behave as if it is the thing creating that data. This works well when the context that model is situated in is similar to the example. It works poorly when there is a mismatch of context. The important thing to know here is that in both cases, it "works". A statistical model never fails: that's a feature of binary logic. Instead, it behaves in a way we don't like. The only way to accommodate this is to build a model out of examples that incorporate every edge case. Those examples can't conflict with each other, either. The model must be biased to make the objectively correct decision for every unique context it could possibly encounter in the future; or it will be biased to make the wrong decision. The only real solution to traffic safety is to replace it with a fail-safe system: a system whose participants can't collide with each other or their surrounding environment. Today, the best implementation of this goal is trains. Humans have the same problems that statistical models have. There are two key differences, though: 1. Humans are reliably capable of logical deduction. 2. Humans can be held directly accountable for their mistakes. Tesla would very much like us to be ignorant of #1, and to insulate their platform from #2.
gtani 1y ago
we have 2 very recent Tesla 3's here (in the US, tho i'm not sure which gen HW 3 or 4 they have and I don't drive them), i'm told (judging by center console) reliably identify anything they need to but FSD isn't happy in construction zones with orange cones and will go slow.
- Fischgericht 1y ago
  In Germany (and a lot of the world, really) town centers are very old and streets are narrow and are shared. Over here it is also totally legal to cross the road wherever you like. Also, due to the narrow roads it's standard practice to be in eye contact with other users of the shared space to make sure who drives/walks next. Car AIs can not hold eye contact, so this is where the problem starts. And, this one of course is very very specific just to Germany: On parts of the Autobahn you have to always expect another car approaching on the left lane with 250 km/h / 155 MPH, so you really have to use the rear view mirror very early to get an idea at what speed that car may be moving. The reach of the Tesla back camera is far too low for another driver at that speed being able to break so to not crash into your back. So, when it comes to Germany even if the system worked better, there simply is no place where you could really make use of it without either killing people or getting killed.

mgoetzke 1y ago

You base that assumption on the visualizations in the center I guess ? They are not actually everything the car sees and reacts to. Especially not in our German FSD cars

Fischgericht 1y ago
I am aware that FSD has a different software stack. But it's the same hardware. So why would they make the detection of kids different on the standard firmware artificially worse? As Marketing for people who hate school kids? I find it laughable that there still are Musk fanboys who after a decade of lies about this still believe in "Robotaxis". 90% of them clearly have never tried to drive a Tesla in a scenario where the minimal protections for kids to use public street space is not "kids should get a SUV to not get killed". It is also amusing to watch videos of Tesla fanboys on YouTube who proudly show that their Tesla now can use FSD for up to 500 miles without a single crash (or "critical disengagement)". A human driver statistically causes a crash every 500,000 miles. But yes, we will have flying Robotaxis in 2 weeks from now, that will solve this problem. Musk said so. :)
- robertlagrant 1y ago
  Other than the first paragraph, this all seems to be replying to something else?
- yreg 1y ago
  > I am aware that FSD has a different software stack. But it's the same hardware. So why would they make the detection of kids different on the standard firmware artificially worse? As Marketing for people who hate school kids? Not sure what's your argument here. The visualization you get using "Enhanced Autopilot" is completely different to the one you get using "FSD Beta" because the software you are running is completely different as well.
  - Fischgericht 1y ago
    The point is not the visualization towards the driver. It's that the same data clearly is the base for the decisions this car makes. If it is not showing the kid crossing the street, you also will not get an emergency break warning, which I get in tons of other situations. What you see is what you get.
    - yreg 1y ago
      I'm sorry, I struggle to understand your point. Which one is it? - Do you think that a Tesla with an enhanced Autopilot would hit a kid because you don't see it in the visualization? - Or do you think that a Tesla with FSD Beta would hit a kid because it uses "the same data" as the one without it?
      - lechatonnoir 1y ago
        I think the idea is, why would the visualization be so intentionally bad in the Autopilot version as to not detect the kids entirely? What benefit does that confer, or, from another perspective, what software constraint forces this to be the case?
        Fischgericht 1y ago
        Exactly that, yes. Thank you.
        yreg 1y ago
        It's not _intentionally bad_. Autopilot and FSD are _different products_ with _different tech_. It's not like they could simply copy `detect_children()` from FSD to Autopilot and call it a day.
        lechatonnoir 1y ago
        Hmm, that might be possible, but that's essentially not what I assumed. At the very least, they operate on the same hardware, so Autopilot is in some sense "intentionally bad" as a whole.
- infecto 1y ago
  Not defending Musk, I don’t like him but I am not sure why you would think two separate software stacks should somehow be comparable. Maybe it’s my old age but I get tired of these style of rants where folks are fixated on a single thing.

fundatus 1y ago

I also find the opposite hilarious: The amount of things that Teslas detect as trash cans is absurd

Sohcahtoa82 1y ago
My favorite is that it identified my wife's Honda CRV as a trash can.

bell-cot 1y ago

> only about 60% of the school kids [...] The rest it would kill. [...] would probably kill about 10-20 kids per week. I'm no Tesla fan - but it would be real-world obvious if even 0.1% of Teslas actually were that "eager" to kill children. In most western countries, covering up child-killing accidents scales very poorly.

avtolik 1y ago
Well, we don't have FSD in Europe, and in US, I guess the children don't walk to school.
- bell-cot 1y ago
  US kids walk to school far less than in the Good Old Days...but there's still a fair amount of walking. And on low-traffic residential streets, there can be quite a bit of de facto playing in the street. So it's still a "passably" target-rich environment for killer robocars.
- usea 1y ago
  In the US, letting your children walk to school is taking a non-negligible risk that you'll be charged with a crime or have your children taken away. Their deaths from a motor vehicle are assumed by all to be a certain eventuality, and parents are more likely to be blamed for it than drivers.

thomastjeffery 1y ago

The worrying part is that if/when those percentages get better, you will be more likely to trust it enough to let it run over children.

zelphirkalt 1y ago
Soon promised to only have 1/10 of detection failures, better than ever before! Only 1 child per week! Rejoice! On a more serious note: Where do we as a society put the bar? What are the numbers, at which we accept the risk? Do we put the bar higher than for humans? Or same level? Or does the added convenience for car drivers tempt us to accept a lower bar?
- thomastjeffery 1y ago
  Setting a bar is the mistake. We need to reframe the entire narrative. Safety implementation is never objective. You can only implement a system by subjecting it to context. Traffic safety is a world of edge cases, and each driving implementation will engage with those edge cases from a different subjective context. We are used to framing computation as a system of rules: explicit logic that is predictably followed. Tesla is using the other approach to "AI": statistical models. A statistical model replaces binary logic with a system of bias. A model that is built out of good example data will behave as if it is the thing creating that data. This works well when the context that model is situated in is similar to the example. It works poorly when there is a mismatch of context. The important thing to know here is that in both cases, it "works". A statistical model never fails: that's a feature of binary logic. Instead, it behaves in a way we don't like. The only way to accommodate this is to build a model out of examples that incorporate every edge case. Those examples can't conflict with each other, either. The model must be biased to make the objectively correct decision for every unique context it could possibly encounter in the future; or it will be biased to make the wrong decision. The only real solution to traffic safety is to replace it with a fail-safe system: a system whose participants can't collide with each other or their surrounding environment. Today, the best implementation of this goal is trains. Humans have the same problems that statistical models have. There are two key differences, though: 1. Humans are reliably capable of logical deduction. 2. Humans can be held directly accountable for their mistakes. Tesla would very much like us to be ignorant of #1, and to insulate their platform from #2.

gtani 1y ago

we have 2 very recent Tesla 3's here (in the US, tho i'm not sure which gen HW 3 or 4 they have and I don't drive them), i'm told (judging by center console) reliably identify anything they need to but FSD isn't happy in construction zones with orange cones and will go slow.

Fischgericht 1y ago
In Germany (and a lot of the world, really) town centers are very old and streets are narrow and are shared. Over here it is also totally legal to cross the road wherever you like. Also, due to the narrow roads it's standard practice to be in eye contact with other users of the shared space to make sure who drives/walks next. Car AIs can not hold eye contact, so this is where the problem starts. And, this one of course is very very specific just to Germany: On parts of the Autobahn you have to always expect another car approaching on the left lane with 250 km/h / 155 MPH, so you really have to use the rear view mirror very early to get an idea at what speed that car may be moving. The reach of the Tesla back camera is far too low for another driver at that speed being able to break so to not crash into your back. So, when it comes to Germany even if the system worked better, there simply is no place where you could really make use of it without either killing people or getting killed.