Self-Driving Cars

Why did it need to be called FSD?

If you want to be charitable, it's because FSD is the eventual goal. But key word: eventual. It's not there yet.

And that really needs to be made abundantly clear to the public. Not in those little disclaimers and EULA's that nobody reads anyway.

They also just handed it out to everyone they could regardless of suitability or likelihood to listen to the rules.

So, now when something goes wrong for whatever reason, the complete lack of any proper media team really won't help with their attempts at playing catch-up. It's really not going to go well when done entirely by Elon Musk on an increasingly unusable site, and more or less you, Fighteer.

Jovian's point isn't meaningless. Comparative risk means nothing when this change is unilateral: Tesla gave a lot of people the option to use something that otherwise wouldn't have it or consider turning it on. The risk of someone improperly using the thing they wouldn't have if not for this beta availability, which removes all the screening for "use it correctly", is to be shouldered entirely by the public, while Tesla gets data from it.

Edited by RainehDaze on Mar 29th 2024 at 1:47:46 PM

Yeah I think Tesla could sell the public on this, the same way they sold the public on their EV charging infrastructure construction, by providing something back.

Obviously as a for-profit company Tesla aren’t going to make the training data instantly available for all self-driving devs, but either providing some data to academia or committing to a delayed release of the data down the line (pick a number of years after full release where major competitors would have caught up anyway and Tesla will be using newer data for its own latest designs) would be ways to sell this as something that produces mutual benefit.

I’d contend that for a lot of people it’s not a proper self-driving car until the technology is mature enough for the car to be sent out to collect a person who is unable to drive.

I will call it a self-driving car when it can in-fact drive itself without me in it.

Edited by Silasw on Mar 29th 2024 at 2:07:24 PM

Anything less than that is just "Driving Assistance". It's an autocorrect but for driving.

I mean, there's a semantic difference between a drunk person calling a robo-taxi and minors being picked up from school that I hope doesn't have to go through extensive explanation. There's no reason why a self-driving car couldn't handle both scenarios at a technical level; it's more a matter of custody. I find it hard to believe that any school would discharge a student to an empty car. Maybe waayyyyy down the line...

For what it's worth, Tesla has an open offer to license FSD. It doesn't seem as if anyone else has taken the company up on it, in large part because the AI/autonomy landscape is incredibly competitive and everyone who's in the market wants to be perceived as a leader. Given how many billions have been sunk into development, I can't see it being given away for free.

The idea of releasing earlier builds as a public service seems pointless, too. An earlier build would probably be less safe, and why would you put something like that into the world? "Here, have our leftovers. All you need is a multi-billion dollar data center and hundreds of top-tier AI engineers and you can catch right up!"

Currently, FSD Beta is classified as a driver-assistance system. It has the potential to be fully autonomous, hence the name, but regulators have not approved it for L4/5 yet. I am not sure if I mentioned it in my earlier posts, but from what I'm seeing on X, to take advantage of their free month of FSD Beta, customers have to visit a Tesla location and be given a short test ride. This should help with the education aspect.

Edited by Fighteer on Mar 29th 2024 at 10:17:11 AM

Honestly, if Tesla had called the current thing “Tesla Assisted Driving”, they could have spun a PR thing out of it.

Refer to it as TAD and do a whole thing about how the driving assistant is *named* Tad. It’d create a sort of personal connection (in the same way as Siri) and become a cultural thing.

Instead, we’ve got this super misleading name.

So you’re saying the process of developing this thing would be completely irrelevant and uninteresting to academics or the history of computing? Because releasing the information to academia was a specific thing suggested.

Edited by Zendervai on Mar 29th 2024 at 10:16:51 AM

Honestly, I'd find the fact they need anyone to collect the children beyond a certain age weirder, but I'm not sure what age ranges we're talking.

Yeah, depends on the age of the child (and the parents' understanding of them), I think. I walked home as an elementary school student, nobody was checking to see that the kids were getting discharged to their appropriate guardians. Granted, that was years ago, but I've seen plenty of kids walking home and taking public transportation without adult guardians after school in my neighborhood nowadays, too.

Edited by PointMaid on Mar 29th 2024 at 7:21:19 AM

This very much depends on local culture. In the U.K. you’d be fine to let a 12 year old walk/get the bus/train home from school alone (or so says the National Society for the Prevention of Cruelty to Children), so them getting the self-driving car home is just that but safer.

I’m not talking about releasing the actual full build, just the training data that gets gathered. Even outside of building a functional self-driving system I’m sure that academia and regulators could find lots of use for a vast pool of vehicle use data.

Edited by Silasw on Mar 29th 2024 at 2:25:13 PM

It *is* absolutely true that Tesla is gathering an enormous amount of traffic data that they’re apparently just kinda sitting on and using internally. I’d bet that most city planners would absolutely *love* to get that data because most methods of data collection tend to be way more specific (focused on specific intersections, for example) or are way more limited because they can’t track that many cars or the cars aren’t equipped with the sensors that a Tesla has.

Oh, the training data. Yes, I could see that getting released in some form down the line.

Speaking of data, here's an X thread from a long-time FSD Beta user showing that v12.3 resolved all of their persistent disengagement scenarios. These are locations and situations where previous versions would consistently take incorrect actions that required driver override.

That doesn't mean 12.3 has zero disengagements. Indeed, later in the thread there's a plot showing those cases. What's interesting is that they all appear to be unique situations rather than, "It fucks up every time it tries to navigate this intersection."

No matter who develops it, self-driving has several technical stages. One is getting the vehicle to follow the basic rules of the road and understand how to navigate its environment. Another is dealing with traffic conditions, lighting, weather, pedestrians, signage, all the stuff layered on top. The final stage is chasing the edge cases. FSD Beta seems like it's very close to that final stage, which is likely to never be finished.

It's what's known in engineering as "chasing the nines".

Edited by Fighteer on Mar 29th 2024 at 10:31:45 AM

The thing is we have data from. the limited testing so far that says, no tesla is not making you less safe, there making you safer ... hard data by the NTSB that says Teslas operating in autonomous mode cause accidents at a rate of roughly 2/3rds that of your avrage human driver. (17 accidents per million miles vs 26 per million for a human driven sedan, 34 per million if you want to count crossovers and suv)

But here is the problem your expecting them to prove a negative, all the times that a human absoltley would have crashed but the car did not cant exactly spread on social media because how do you prove that, only the accidents the car did cause can be proved... and those ingnore that in the time it has taken me to type this much of the post US drivers have caused 40 fatal acaccidents.

However by looking at aggregate data the results are very clear, but people just ignore that shit.

It strikes me as the technological equivilent of antivaxers "how do I opt out of this thing that the hard data says is substantialy safer because social media has convinced me the edge cases where it has gone wrong are normal"

Edited by Imca on Mar 29th 2024 at 11:37:12 PM

Wow, that's a bad faith argument. Calling us akin to anti-vaxxers.

I think this kind of folds again into Tesla not really having a PR department worth a damn. They’ve got occasional press releases and a bunch of tweets from a man with an overwhelmingly untrustworthy and slimy vibe. Like, I don’t know, maybe it’s really bad for Tesla’s main PR guy to be openly antisemitic, openly anti-queer, being sucked into extreme far right conspiracy theories at the drop of a hat, etc etc etc. They need to spin up actual advertising and get actual PR people in the mix and actually address this stuff and put it into the spotlight in a way that *doesn’t* involve a guy who obviously buys into the Great Replacement conspiracy theory and thought telling queer people to breed was a good look.

Arguing agianst hard data with the appeal to emotion of "I dont like new tech its scary?"

It's the same thing they do.

Agian, how does even a good PR team prove a negitive?

Hey our cars get into less accidents then a person, here is the national transportation saftey board saying so... doesnt exactly work.

Edited by Imca on Mar 29th 2024 at 11:39:21 PM

It's not and you know it.

We at least don't make up shit like FSD causes autism or something. We look at the times FSD fails where a human driver probably wouldn't and say "hey maybe this needs more testing".

Anti-vaxxer bullshit started with an outright lie. That's the big difference.

Edited by M84 on Mar 29th 2024 at 10:40:27 PM

Meh, I think the question of whether self driving cars are safer than non self driving ones is separate from the question "is calling a car that requires supervision 'self driving' dishonest and risky?". These three things can be true at once, and I think they are all correct.

It’s really very much not.

It’s one thing to be like “I am concerned about this technology being developed by a company where all the PR is filtered through someone with a proven track record of being wrong who has a lot of extremely fucked up views that aren’t backed up by anything” and “I don’t believe in extremely well established science.”

Edited by Zendervai on Mar 29th 2024 at 10:40:34 AM

Nah the anti self driving crowd just makes up that it is more dangerous then a human and regularly runs over preschoolers.

Totally diffrent.

As you just did right there, the data says a human would have failed too, one and a half times over infact.

Like even "its obvious a human would not have done this" is not obvious, I have personaly watched some one run a stop sign and broadside a fucking city bus, you know... the giant hybrid monstrosities the MTA owns?

There is no way they should have missed that, and yet they did.

Humans are more fallible then twitter thinks.

Edited by Imca on Mar 29th 2024 at 11:45:53 PM

The safety of self-driving cars is a statistical phenomenon that requires hard data, not emotional reactions. That's the bottom line.

My angle is “don’t call it full self driving if it isn’t full self driving” and that Tesla really needs an actual PR department so they can address this stuff properly.

Is this unreasonable?

Especially since an actual PR department could set up a lot of advertising things they could use to push back against negative PR.

Right now, it’s a guy who just told gay people that they need to breed which, really fucking goddamn disgusting, dude. Way to completely buy into the racist-ass Great Replacement theory.

Edited by Zendervai on Mar 29th 2024 at 10:43:28 AM

Your "angle" has nothing to do with the actual safety of the product, which is what matters in the end.

And I’m saying that Tesla is doing an absolutely garbage job at actually selling the safety of the product. It’s a string of bad stories and Tesla going “that’s not our fault” over and over and over and it looks really bad and it comes off like Tesla looking for technicalities and excuses.

Up to and including stories where they go “that’s not our fault, yes, the tracking was off and we can’t prove anything, but it’s not our fault”.

It doesn’t make the company sound even remotely confident. If every accident is an exception, what would make the company admit that it *did* make a mistake?

Edited by Zendervai on Mar 29th 2024 at 10:46:04 AM

Another big difference with anti-vaxxers is that they also go against the advice of medical professionals, as opposed to the pharma companies themselves.