Hacker News Reader

Comment by dawnerd | Hacker News Reader

He was the guy that let the car drive into the pole. That completely goes against what everyone agrees to when joining FSD, and probably what employees agree to. Fact that he even posted and and tried to capitalize on it really shows...

https://electrek.co/2022/02/04/tesla-full-self-driving-beta-...

View on HN

This was considered the first confirmed crash on FSD beta. Unfortunately his firing is going to encourage other beta testers to only show positive clips.

We’ve already seen other testers try to hide near misses and failures. Here the testers suggest cutting out a clip where FSD beta lunges toward a bicyclist: https://mobile.twitter.com/omedyentral/status/14911468200972...

That same tester (Galileo Russel) nearly had beta access revoked after posting an earlier video where the beta crowded a pedestrian. He issued a DMCA takedown to try to stop the spread of the clip. https://news.ycombinator.com/item?id=28566376

Whatever you think of the beta (I think it’s blatantly irresponsible), Tesla should not be pressuring the testers to only share the good clips. That will lead to overestimation of the system’s capabilities and deaths in the long run, like what we’ve already seen from autopilot.

The guy laughing at the car almost running down the cyclist is a goddamn psychopath. They were moments away from injuring or killing someone.

That seemed like nervous/panic laughter to me.

Just nervousness/general emotion kind of "laugh", rather than "haha, that's funny" kind of laugh.

That's why they stopped using the mode after, right? Fuck em.

I think the entire concept of using this kind of beta-test on public roads is an extremely dubious one, so I agree with your "fuck em". I just don't think the guy is a "goddamn psychopath" based on a nervous emotional release laugh.

Indeed. The "goddamn psychopath" is the guy who shipped beta FSD software and thought it'd be fine to let randoms "try it" on public roads.

You're not wrong, but as someone who is forced to share the road with FSD Teslas, I'm happy that he was willing to show what one is capable of doing without proper interference.

Well we all share the air with aircraft. How would you feel about a pilot “demonstrating” controlled flight into terrain on autopilot?

I assure you my Lear’s autopilot would do it.

With nary a civilian or other airplane within impact distance, like in the video? Sure, let's do it!

Most of us don't have Lears to test that hypothesis out.

Aviation is severely regulated. After decades of horrendous accidents...

Self driving is new and it is painfully obvious the government hasn't figured out how to deal with it yet.

When the risk of an airplane crashing into my car/my bike/me as a pedestrian is 1/1,000,000 as likely as a car collision, we can talk. Meanwhile, if you want to splat into the edge of a mountain, I really have no feelings about it.

Or did you mean you would use your Lear's autopilot to try to do some extremely low level flying over a highway?

I think the point you're trying to make (maybe) is: aviation autopilots will also lead to disaster if used improperly, but that doesn't imply that they're bad or deficient technology (but rather that they were used improperly).

First, aviation autopilots are a mature, tested, reliable, certified technology and are deployed only under narrow, precisely delineated circumstances by trained professionals (not a beta product tested "in production" by amateurs).

Next, though, you imply that the FSD was used somewhat irresponsibly or improperly here. I fail to see that. He used it as it was supposed to, and when he saw that it was swerving into the pylon, he reacted immediately (within the limits of human reaction time), which was insufficient. How was that improper use?

That cars are capable of crashing was known before FSD was even a thing. Statistically, you should be much more afraid of the cars driven by human drivers, because they can do everything FSD can, and there's truly not many on the road using FSD.

The statistic I wonder about is -- self-driving cars will shrink the pool of available donor organs by about 20% ( https://futurism.com/neoscope/self-driving-cars-will-save-li...).

Is there an externality that should be priced in here? How?

Think of it this way... those organs are still being used to save someone's life, they're just doing it by staying in the original body.

In a strict utilitarian analysis, they are saving one person's life, but could probably save several peoples' lives and dramatically improve the quality of life of others.

Ah yes, the "let's kill people to get their organs" argument, which just seems so reasonable...

It's certainly not reasonable. But neither is responding to a question about the side effects of fewer donations as a result of car crashes with a dismissal about how the people in the cars will be alive.

One utilitarian perspective is that a world where you are at risk of your vital organs being reallocated at any time is such a bad world (deep anxiety for everyone all the time) that organ seizure would never be a reasonable policy choice.

On the other hand:

(1) Maybe this isn't true and we just think organ harvesting is an evil because of our innate status quo bias. Maybe like in Kazuo Ishiguro's 'Never Let Me Go' if there were a class of people who could be harvested at any time they would just ... Accept it and deal with it.

(2) maybe we do live in that world? People continued to travel to, work in, and do business with a place where there was credible evidence of an underclass who was being harvested ( https://en.wikipedia.org/wiki/Organ_harvesting_from_Falun_Go... )

(3) what if there were one person who just LOVED forcible organ harvesting, loved it ten billion times more than every other human on the planet hated it? Morally a utility maximizer would choose to bend to that monster's preference.

You're trying to make this polarized, but it's not about good vs evil. It's an observation that it's likely something in an equilibrium today is likely to shift somewhat in the future. No one is saying "now we need to harvest organs" but "there is going to be a new problem emerging in the future, and that problem will require new solutions".

Even as someone quite interested in autonomous vehicles, and optimistic that they will become extremely feasible, creating dramatic changes in our world in the process, I hadn't until now thought of the organ shortage they might create.

Does it mean we start creating people to harvest their organs? Of course not; please don't be ridiculous. Perhaps it means that biotech companies have yet another gap in medicine to think about tackling, though.

> That cars are capable of crashing was known before FSD was even a thing. Statistically, you should be much more afraid of the cars driven by human drivers, because they can do everything FSD can, and there's truly not many on the road using FSD.

It's not about statistics, it's about technical sensing limitations in Teslas leading to blind spots that have provably killed people. Human drivers with such blindness are not even allowed on the road. But Tesla refuses to use state of the art sensing technologies which have been shown to make autonomous cars safer for everybody. Tesla inexplicably and stubbornly refuses to use these sensors, it's really baffling to me. This leads to their cars having limitations in their sensing capabilities, such as trouble detecting large stationary objects. Therefore you see stories often reported of Teslas running into parked vehicles, especially large ones like fire trucks [0]. This problem dates back to 2016 when a man was decapitated because his Tesla couldn't recognize it was about to run into a tractor trailer [1]. In a sane world, that would have been the end of Tesla's experiment and they would have been forced to use safety features we all know they should be using (and Tesla knew or should have known at the time of the accident). But no, the same exact thing happened again in 2019, but this time with a Model 3 sliding under a tractor trailer [2]!

This is literally what driverless cars are supposed to excel at -- don't run into things -- and yet Teslas routinely do this because of limitations in the Tesla sensor stack, limitations we don't need statistics to call out as troubling for technical reasons. At this point, this choice has a confirmed and ever-expanding body count, and Tesla is not working to fix this problem, as it's been reoccurring for over 5 years now (wasn't level 5 autonomous capability predicted within that timespan?); nor are they willing to cease using public spaces to beta test their products, which in my opinion needs to end immediately.

[0]: https://www.msn.com/en-us/news/technology/why-teslas-keep-st...

[1]: https://www.theguardian.com/technology/2016/jun/30/tesla-aut...

[2]: https://electrek.co/2019/03/01/tesla-driver-crash-truck-trai...

It is entirely understandable that Tesla would wing it without lidar. Once one gets over the assumption that their priority is creating a level 5 system.

What Tesla cares about is selling expensive vaporware. If this vaporware depended on an expensive prop, this would cut into cash flow they want to spend on software dev. They've repeatedly said "we can do it without lidar". What they meant is "we cannot afford lidar in our situation." It doesn't look like it's going to work, but it also doesn't look like adding lidar solves the basic issues. So in the end at least they didn't con people into buying useless lidar. Just useless vaporware.

Statistically, yes, but I think people can be afraid of the lack of direct accountability for AVs. A distracted/drunk driver, for example, is something easy to attribute cause and blame to. And you're not afraid of the car, you're afraid of the potentially negligent, dangerous people driving it.

But an autonomous object that behaves wildly unpredictably simply because of a malfunctioning sensor or a software bug is something that defies known reasoning which you're normally trained to respond to. You can no longer make eye contact with drivers at a crosswalk, instead you can only assume the AV will behave as normal, and if it doesn't and runs you over there's nothing you could have done differently and you were just unlucky and lost the statistics lottery.

I am very uneasy of car manufactures and regulators writing me off as a factored-in statistic. You are basically then allowing vehicles on the road with a deterministic cause of incident.

Perhaps this is just a difference between two valid personal views.

For me, people are the thing that will be wildly unpredictable. I anticipate that autonomous vehicles will eventually make roads seem extremely machine-like & predictable.

I recognise that Tesla's FSD isn't there yet, but I anticipate that it eventually will get there.

"Let it hit" is pretty deceptive wording. You can see in the video it was not an intentional decision on his part. He trusted the car in a low-risk situation, where the only repercusion was small cosmetic damage. Perfect to see what the car can and cannot do in a responsible way. To me it does not reflect the least bit poorly on him.

I do find it weird that an employee ongoingly posts critical reviews though. I can understand why Tesla do not want that.

So much discussion here but no one has posted the referenced YouTube channel. Anyone have it? Not sure where to find it

https://www.youtube.com/c/AIAddict/videos

The video was embedded in the article he linked, but for maximum convinience the direct link is https://www.youtube.com/watch?v=sbSDsbDQjSU and the time of the collision is 3:30.

Posting unbiased full length videos of FSD was much more informative than self-censoring for Tesla, or "capitalizing" as you call it. The fact they removed his access to FSD makes it likely the few with access to FSD will feel the need to self-censor issues in the future.

He broke the terms of the FSD program. That should, for the safety of everyone on the road, lead to revocation. I'm part of FSD, I don't self-censor. It's terrible right now. I'm more worried about youtubers letting the car make mistakes just for the views. And honestly it's not safe for people to be trying to talk while paying attention to a car that constantly does dumb things.

I just don't see how you could follow his channel and watch that video and come to conclusion he let that happen for youtube views.

I've seen plenty of FSD Tesla abuse IRL with my own eyes. Infinitely more dangerous than what this guy did.

Is there a way to report these truly egregiously dangerous license terms offenders to Tesla via license plate along with my dashcam footage?

A lot of people in this thread are mixing up the FSD option (that you buy for your car) with the invite-only FSD beta. Currently the non-beta FSD will drive itself on limited access roads (basically taking you from onramp to offramp, changing lanes, taking interchanges, etc), but it's basically cruise control on surface streets. The FSD beta will self-drive pretty much anywhere (even parking lots) but it has only rolled out to a few percent of Teslas. Both versions use cameras inside the car to monitor the driver's attention. If you blind the internal camera, the car will refuse to engage autopilot or FSD beta.

If you're seeing "egregiously dangerous" behavior, you should call the police. The dash cam should be more than enough evidence to get the reckless driver charged.

In my experience the local police don't have time (or bandwidth) for it, unless the person is driving like a total criminal on an ongoing basis.

Is using FSD safely not part of the terms? What about the people I see cruising down the road glued to their phone?

You can't use autopilot or FSD while on your phone. I've tested this myself. Within a second or two you're alerted to grab the wheel and look ahead. If you don't, you'll get locked out of autopilot for the rest of your drive and your safety score will be dinged.[1] Tesla's ML algorithms even have a classification of "PHONE_USE" for the cabin camera.[2]

1. https://www.tesla.com/support/safety-score

2. https://twitter.com/greentheonly/status/1312607693249667073

No they’re not. Tesla rebranded enhanced Autopilot to full self driving. FSD Beta is what people paying for the option we’re promised.

This service is apparently "$12,000 up front or $199 per month" according to this article. You are saying that people should not be allowed to show that this very expensive service is not very good as it can't handle and empty street with a small pole in it? And that showing that this very expensive service is not very good is "against the terms" and should lead to "revocation" (presumably without refund)? That sounds like some border-line dystopian stuff here.

Anyway, this FSD beta is billed to "enable Tesla vehicles to virtually drive themselves both on highways and city streets by simply entering a location in the navigation system, but it is still considered a level 2 driver assist since it requires driver supervision at all times." Maybe, for the safety of everyone on the road, this entire thing shouldn't exist considering how badly it performs. But apparently showing this thing is kinda crap is the problem, rather than the company using the actual roads for the beta tests of their crappy software.

That video did not remotely seem like he intentionally crashed it like others are speculating. They even braked the car, which the TSLA didn't respect.

Not to mention, it ran a rolling right turn on a red light. FSD is definitely not ready for mass market, although TSLA seems to be willing to make the bets necessary to make that happen, the only way it can.

I’m not convinced the car is capable of not respecting your brake?

ABS is doing exactly that, just for very short periods of time.

ABS respects your brake, it just does it by apply counterintutive high-speed anti-brake pumping. It'd be like saying power steering doesn't respect your steering.

Power steering can override your input by just applying enough force.

What do you mean "let"? It looks like he took control within less than a second after it was clear the car was turning too much.

>He was the guy that let the car drive into the pole.

He "let" it, did he?

>That completely goes against what everyone agrees to when joining FSD, and probably what employees agree to.

This makes no sense.

>Fact that he even posted and and tried to capitalize on it really shows...

Someone posting "reality" has an agenda, whereas all of the people posting nothing but (highly edited) successes of FSD have zero agenda.

Reply on news.ycombinator.com