Hacker News Reader

Comment by computerex | Hacker News Reader

This is why when I spot a Tesla on the road, I make every effort to try and get as far away from it as possible. Placing a machine vision model at the helm of a multi-ton vehicle has got to be one of the dumbest things the regulators have let Elon get away with.

View on HN

Tesla isn't the only brand that is doing that, and most of us didn't pay for it and are driving ourselves. So maybe pretend to be terrified by something else.

Other brands tend not to rely so much on a vision model for their driving assistance features, instead relying more on a variety of sensors, traditional computation, and curated databases.

ChatGPT is bad...but ChatGPT with a radar is something I'll trust.

If it has a radar, then it should have something to do with the sky... hmm, what would be a good name?

Agreed, they tend to drive extremely badly and I don't trust them. Avoid them, anything nissan and lifted trucks (especially dodge RAMs). Stats show that a solid 5% of all ram 1500 drivers have a DUI on record. It's like 0.1% for prius drivers.

re:machine vision - What's particularly bad is that they don't actually put enough sensors in the cars to be safe.

Pre-collision systems, blind spot monitors, radar, parking sensors, etc are all so helpful and objectively good for drivers. Doing a vision only implementation and then claiming "full self driving" is where it gets awful.

Has it been proven to be worse than normal drivers that pass their test?

Where would that data live and how would we disentangle it from income effects?

I don't think you would need to disentangle it from income effects to address the concern of someone who actively avoids Teslas on the road, unless they also actively avoid cars that look like they belong to people with low income.

> I don't think you would need to disentangle it from income effects to address the concern

Good point, I was mainly thinking of the 'AI saves lives' argument.

> of someone who actively avoids Teslas on the road, unless they also actively avoid cars

Probably they should avoid hitting _any_ cars =)

What does worse mean? Tesla's AI drives worse than an 80 year old with cataracts, but on the other hand it can react faster to obstacles than the fastest race-car drivers.

I don't own a Tesla so have no direct experience, but my guess would be that it might crash less than a human, but has far more near-misses.

Worse probably means something like "accidents per mile".

Edit: was curious, so I did the math. The standard unit seems to be 'crashes per 100 million miles' so according to the Tesla safety report (only source on Autopilot safety I could find easily) that works out to [0]: one accident for every 4.41 million miles driven = 22.7 accidents per 100 million miles.

Without autopilot, the AAA foundation (again, easily available source, feel free to find a better one, this is from 2014) [1] says that our best drivers are in the 60-69 age range, and have 241 accidents per 100 million miles. Our worst drivers are our younger teens (duh) with 1432 crashes per 100 million miles.

So unless you can find better data that contradicts this, Autopilot seems like a net benefit.

[0] https://insideevs.com/news/537818/tesla-autopilot-safety-rep...

[1] https://aaafoundation.org/rates-motor-vehicle-crashes-injuri...

I think your analysis is a good one given the data we have, and can be used to draw some conclusions or guide general discussion. However the analysis is indeed limited by the data available. The AAA data does not consider variability by gender, race, socioeconomics, location, etc. Further it does not normalize variability in the types of driving being done (teslas have limited range, are not towing trailers, etc), nor other technological advances (modern vs older vehicles).

Autopilot only works on highways and decent weather and visibility — i.e. the least accident-prone scenarios possible. AFAIK, AAA’s stats count accidents in all situations

That is an apples to oranges comparison for several reasons, including:

- People don't use autopilot in all situations, its use is biased towards the easiest driving conditions.

- Your non-autopilot statistics include old cars that have fewer safety systems.

A real comparison would normalise for these factors, which isn't trivial to do.

Id wager for a modern car it doesn't crash any less. My modern Toyota also stops faster than I can thanks to a big radar unit hooked up to the brakes.

In fact, almost all new cars, Teslas included, since 2022, have an Automatic Emergency Breaking system (AEB), which will hit the breaks if you're about to hit something. If I were walking in a parking lot and had to step in front of an older SUV or a Tesla, I'd step in front of the Telsa.

https://www.jdpower.com/cars/shopping-guides/what-is-automat...

Do you need it to be proven scientifically to take caution on something? I take caution around every car.

Whether it's "proven" to do so probably depends on what kind of proof you're looking for.

But there are plenty of news articles and youtube videos that show it doing ridiculous unsafe things (including running into a truck that was driving perpendicular to the road). So I highly doubt it's as good as a normal driver, in fact I'd be shocked if it is.

>But there are plenty of news articles and youtube videos that show it doing ridiculous unsafe things

There are also plenty of videos of human drivers doing absurd things, so this is hardly an argument.

All that matters is whether a random Tesla that you see on the street is more likely to crash into you than a different car. I know that Tesla has published there own statistics which say that it isn't, but I would be very interested in seeing an independent study about this.

Consider this:

A tesla’s fsd code had gone through the sdlc they choose, involving project management.

I distrust project management.

Therefore: I only trust a tesla’s fsd mode as much as I trust a project manager’s driving.

It has been shown to fail in less predictable ways. We have a good intuition of the stupid things people are going to do on the roads.

Saw a Tesla get pulled over on the Autobahn for holding the left lane and being too slow. That was one of my favorite moments on the road. Second only to watching a team of truckers box in an asshole driver while going up a mountain.

Reply on news.ycombinator.com