Hacker News Reader

Comment by GuB-42 | Hacker News Reader

Tesla charges $10k for not really self driving, with the promise that it will improve, and you should buy it now, because the price will increase in the future. I.e. they make you pay for something you don't have, with no guarantee that you will have it later. If it isn't a money grab, I don't know what is. And of course you can take a loan on that, so it isn't an upfront cost either.

At least, with VW system, you pay when you use it. Doesn't work in your area, you don't pay. And if you realize that you can't pay, you can simply drive, your car won't turn into a brick.

And of course it is to extract more money from consumers, that's the goal of every company. But personally I find VW approach more honest than Tesla's. The best would be to offer working self driving out of the box, with no promises and no recurring fees, but the reality is that we don't have the tech yet.

Note that I don't have anything against the $10k package from Tesla except for the way it is advertised. If they just sold you an advanced supercruise, there would be no problem. But they also sell you the idea that your car will fully drive itself in the future and that you are making an investment. Kind of like a Kickstarter, except it is a trillion dollar company, not a bunch of people in a basement.

View on HN

As someone with a non Tesla car with some smart cruise control features, I think you’re being too harsh on Tesla.

I have a 2020 Hyundai, and there has been no major refresh in the 2021 or 2022 models of the same car. But the software features have each gotten slightly better. The cruise control will only slow down and stop you on my model, but on the newer model, it will stop and then start going again. My model will sometimes correct the steering at the last minute if I start veering outside the lane, the newer model will more gradually keep you centered in the lane so you don’t even start to veer out. Pretty minor stuff like that.

There’s no way the hardware on my car is the limiting factor, but I will never get any of these improvements. Tesla is still the only car maker to decide that car software can be treated like almost any other software people use, and actually updated independently of the hardware.

Even if the 10k self driving charge is a little too ambitious, in my opinion Tesla are still miles ahead of all the other major brands in user-friendly practices.

It’s always interesting that people get so used to the status quo even if it’s actually inefficient or wasteful or arbitrarily worse than it easily could be. But if a company does something ambitious and new, but it has some downsides like a higher cost in the beginning, those downsides get judged so much more harshly than the downsides of the status quo.

> There’s no way the hardware on my car is the limiting factor

I'm not sure why you'd make that assumption. The newer models that can do more may very well have more/better sensors, faster processor(s), more RAM, etc. Just because the interior and exterior look the same doesn't mean that everything else is the same too.

I think the point stands that there is probably a ton most manufacturers could do in terms of over the air updates with existing hardware that they just don’t.

I see that as a separate and larger point, but it is true. The reasons for it are also more complex than "they just don't" though they aren't technological reasons.

They will get sued for this one day.

For what? Putting better stuff in newer models? That's how basically every product works.

For not being transparent with consumers. What will happen when one day you turn on your Tesla and it relays a message to you that FSD is disabled because you are missing hardware required by the new version?

Well a lawsuit would be 100% justified in that case. I think the far more likely scenario is that Tesla would release an update that simply wouldn't apply to vehicles with the missing hardware, which would stay back on the older version.

You are exactly right. And then they will eventually depreciate old cars because if google can’t maintain android development for old phones why do you think Tesla will be able to do it for old cars?

I guess I can't know for sure about any one particular feature. But as far as I can tell, the company doesn't ever backport any smart/autonomous features to older models. Whereas Tesla manages to do this continuously, and years earlier than other brands. I doubt that every single incremental improvement to one of these features relies on brand new hardware or sensors that was not available the year before. Some of them are clearly just software improvements.

Tesla builds their cars with this in mind, and pretty much nobody else does. Even for updates that are purely software, that software may not even be updatable without physical access to the module(s) that run it.

The idea of a car you bought being updated new features from the manufacturer down the road didn't even exist until Tesla. The legacy car manufacturers can't just turn on a dime and start doing the same thing for many reasons, ranging from accounting and contractual issues to actual technological ones.

None of it is as simple as it may seem.

I'm sure these are the same kinds of excuses the execs within the company give as to why they can't do what Tesla is doing. And that's part of why Hyundai stock is up ~30% over the past 5 years and Tesla is up over 1000%.

> There’s no way the hardware on my car is the limiting factor, but I will never get any of these improvements. Tesla is still the only car maker to decide that car software can be treated like almost any other software people use, and actually updated independently of the hardware.

You also won't have any features removed, like Tesla has also pioneered.

I have no idea what you are talking about here. Cars sometimes lose features between model years, or even during mid year refreshes. Tesla is no pioneer there.

Tesla has gimped range on cars that were already sold with an OTA update. This is far different from a mid-year refresh.

That's fair, and the limited charging speeds have been even more significant at times. This is one of the big downsides of the warranty being so weak, IMO.

I'm talking about OTAs being pushed to cars that have already been purchased. Not all of them add features. Some have removed or modified features in ways that owners did not want.

OTAs are great when they add features. They suck when they remove/alter ones you previously liked.

Yeah, that's fair. Not being able to opt out of them can be a problem. Though its tough as there are some safety related ones mixed in with the others.

> My model will sometimes correct the steering at the last minute if I start veering outside the lane, the newer model will more gradually keep you centered in the lane so you don’t even start to veer out.

On my 2017 VW, this enhancement was merely a feature flag waiting for the customer to enable using VCDS/OBDeleven. It's an amazing difference!

Another good one I found and enabled is lane change assist, which makes it so that when you're on cruise control but slower than your set point (due to tailing a slow lead car) putting your turn signal on will immediate begin the acceleration that otherwise would only occur once you have completed the lane change and are no longer impeded by a slow lead car, so you're better matched with the flow of the new lane as you start entering it.

VCDS and OBDeleven are essentially emulators of the factory scan tools that only VW dealerships normally have. Way, way beyond OBDII code reading.

> At least, with VW system, you pay when you use it. Doesn't work in your area, you don't pay. And if you realize that you can't pay, you can simply drive, your car won't turn into a brick.

I think this is a very optimistic take. We are already seeing the industry push out subscriptions for features like Ford's Blues Cruise. The costs are low now, but obviously they can go up.

I'd be really surprised if VW doesn't start pushing something similar well before they have anything approaching L5 autonomy.

It isn't too crazy either, as these advanced features are already useful.

This is a great argument. Plus VW has an incredible incentive to make it great.

Tesla auto pilot is like vaporware: selling software that may never be released, as requires a break through in Aai capabilities.

> And of course it is to extract more money from consumers, that's the goal of every company.

No it isn't. We don't have companies so that they can extract money from consumers. We have companies to help efficiently distribute wealth through the community to maximise the probability of social needs being filled.

If the goal was to extract money from consumers and give it to those with spare capital, there are vastly more efficient ways to do that. Why would they even bother producing valuable goods?

It's the only legal way.

Have you had a look at the healthcare industry in the US vs healthcare in the rest of the world?

"Why would they even bother producing valuable goods?"

Like opium by the East India company or drugs by mexican cartels?

Getting people to buy something is a lot easier than chasing them with a gun, just look at how bad we are at colecting taxes.

"We have companies to help efficiently distribute wealth"

There is no economic theory, from marxist to neoliberal, that claims conpanies are there to distribute wealth. They organise capital and labour, and maybe they fullfill needs, unless they are doing HFT or something similarly murky.

I guess they "distribute wealth" in the sense that they first extract it, then concentrate it and finally distribute it to a very limited audience. It's not usually the primary or emphasized act though...

Perhaps I should have said "organise capital" instead of "distribute wealth". I mean they cause it to be available to certain people.

Not to be curt, but this is one of the most naive things I've read on this website.

90% of the time, people start companies to make money, not to "efficiently distribute wealth through the community."

He or she said:

> We have companies to help efficiently distribute wealth through the community to maximise the probability of social needs being filled.

We here presumably means society in general. We set rules and give privileges to people to run companies in order to better society. In a capitalist system this is done through the medium of 'enlightened self interest' so yes, one starts a company to make money but that is not the reason we have companies.

You are arguing against a point he didn't make.

"so yes, one starts a company to make money but that is not the reason we have companies."

I don't understand this at all. The reason people create companies is the reason that companies exist. They're by definition the same thing - companies don't exist without people creating them. So if people start companies to make money, the reason companies exist is to make the people who start them money.

You can't just create a company - you need a legal framework for a company to exists.

> You can't just create a company - you need a legal framework for a company to exists.

Not really, because “company” is a generic term for any organization of people with a common business purpose independent of whether it has a distinct recognized legal recognition, existence, or treatment separate from the constituent individuals.

A corporation, LLC, or other legally recognized business entity requires a legal framework.

> Not really, because “company” is a generic term for any organization of people with a common business purpose independent of whether it has a distinct recognized legal recognition, existence, or treatment separate from the constituent individuals.

This isn't really true. A company doesn't include the employees of the company, not the way I used it and not the way the parent used it. So it doesn't refer to a generic term for any organisation of people with a common business purpose. It really only refers to those which have some category of legal framework.

No, this logic is circular. You're defining a company as a thing that we've legally defined as a company, ergo you must have legally defined something as a company for a company to exist.

That's not true, though - the non-legal definition of a company is "a commercial business."

You can absolutely have a commercial business without a legal framework for it. People engaged in work and trade well before anyone wrote up a legal framework for companies.

I think you overestimate the extent to which our economy is planned. Companies sprouted up because people wanted to create them, not because society decided they were efficient.

I appreciate your curtness, but I don't think I was naive. Can you answer my question: So why do they bother producing valuable goods? Why don't we adopt more efficient means of transferring wealth from those without spare capital to those who have it?

> At least, with VW system, you pay when you use it. Doesn't work in your area, you don't pay.

"AWS has billed us tens of thousands of dollars because we forgot to turn something off/confusing billing" hasn't taught anyone anything?

There's nothing honest about VW's approach. They won't have self-driving either, it will be at best the same as Tesla's. And yet, they want to charge you, and charge you significantly more than 10k upfront (because they expect to extract this payment every time, all the time, for the entire lifetime of the car).

That's the same argument against software-as-a-service

Maybe I'm not bothered about self driving for 90% of my journeys, but on occasion I'm going somewhere unusual or want to travel overnight or after drinking or whatever I'll just pay for a few hours.

Imagine a world where there was just a set fee of $x per hour and $x per mile to travel. Would allow a far easier choice between different modes of transport, as the marginal cost becomes the actual cost.

(Of course self driving is years away so this is all just spin)

It won't be long after self driving arrives and you won't get that option. Humans are terrible drivers, even in the best cases. Once self driving is better survival than humans it won't be legal to drive on public roads.

A future where humans are no longer able to drive will have to be a decade or more after self driving arrives. It’s impossible for everyone to replace their non self driving cars immediately.

The used car market is a $150bn industry. A quick spot check found that a 2010 Honda Civic still costs between $5k and $10k depending on mileage, etc.

By the time that self driving is required by the NHSA or whatever body requires backup cameras and the like in new cars self driving will be standard.

A $20k Subaru will come with it out of the box, no add ons, just like a backup camera today.

It might be possible to add self-driving features as an aftermarket upgrade. My 2013 model car already has drive-by-wire features (for automatic parallel parking), so it just needs more cameras and a replacement CPU. Heck, as a last resort you could put a robot in the driver's seat, physically operating the steering wheel and pedals.

<A future where humans are no longer able to drive will have to be a decade or more after self driving arrives.

The National Driver's Association will arise with bumper stickers: "You can pry the steering wheel from my cold, dead hands"

...38,000 times a year

As self driving becomes common, people won't bother with getting a driving license. States that do can actually put in decent levels of restrictions and competency to get one rather than letting an 16 year old with 5 minutes driving round a car park out on the road.

As fewer people get new licesnse, the demand for self-driving goes down as people who can drive the legacy way die off.

Maybe where you live. There's a 0% chance human driving will be banned in America until we are all well passed driving age.

I wouldn’t be so confident. I think you’re drastically underrating how much convenience will drive changes in behavior and social expectation. If self driving works (big if, I’m dubious about this), then the vast majority of us will switch over to using it for regular trips very quickly. As that happens, the perception about what is safe and socially acceptable will change quickly, as will the types of vehicles consumers demand.

This type of change isn’t instant, but it can happen faster than a human lifespan. Think about someone born in the 70s and what they’ve experienced with cigarettes. They were born into a world where close to half the US population smoked, and they’re now retiring in a world where smoking is de facto banned in most non-residential buildings. That is a drastic change, and remember that tobacco is addictive, which hampers change much more effectively than habits like driving.

I wouldn't be so confident 100% ("steering-wheel-less") L5 self-driving is possible. Long tail is a thing. There will for a long time be a need for manual override, for things like tricky parking or slow-speed maneuvering, bad weather, weird lightning conditions, etc.

That rules out de facto driving ban. That leaves driving ban by fiat. Also never gonna happen in this century, unless it carves out an exception for "classic cars."

As far as hazard ratios go, cigarette smoking is orders of magnitude more dangerous than driving. Not a good comparison.

> I wouldn't be so confident 100% ("steering-wheel-less") L5 self-driving is possible.

I’m not, you didn’t read what I wrote. I specifically predicated my theory on if it was possible, and indicated my reservations there.

Even ignoring smokers - second hand smoking kills more than cars

https://www.cdc.gov/tobacco/data_statistics/fact_sheets/heal...

https://en.wikipedia.org/wiki/Motor_vehicle_fatality_rate_in...

It wont be banned, it will just cost much more to insure (since it will be directly tied to risk, computers already drive better than humans). If you drive manually, you will need to pay for the privilege.

> since it will be directly tied to risk, computers already drive better than humans

They do not! Are you comparing computer driving in "easy" situations with human driving in all situations?

I think the comparison is made over the entire range of situations.

If we can get rid of, or vastly reduce, the problems of inattention, exhaustion, rage, chemical impairments and the like, we have a huge improvement. I think that improvement is vastly more, cost measured in human life, than the problems we see with computer control in odd situations.

So yes. They do drive better than humans. The fact that their failings are wack-doodle to us (you drove _right_ into that truck??) is swallowed in the fact that their failings are profoundly fewer than ours.

> their failings are profoundly fewer than ours

Better start showing off some statistics along with that claim.

Insurance is already directly tied to risk - car ensurers are profitable companies with current rates

Yeah, I don't think it will ever be completely banned. It might be restricted a little bit, though.

But the big benefit that I see of self-driving is that a lot of manual driving becomes safer. Autonomous vehicles won't overlook motorcycles and Miatas.

Autonomous cars will go a long way to making small, fun vehicles safe again.

Maybe on restricted access highways, but self driving isn't coming to dirt and gravel roads any time soon.

From a technology standpoint, this seems much much easier for AVs vs humans than many other AV challenges. When it comes to measuring physics precisely, complex computation, and rapid control, computers beat humans by a million miles. This is in stark contrast to, say, driving through a crowded urban environment with all manner of rule-breaking peds and bikes popping out of everywhere.

The ability to calculate the physics of driving dynamics has been solved for decades. But, if you feed garbage data to your physics models, you will get a garbage result. Computers are still very very awful at image processing as compared to humans.

Yes, I'm saying that environments with diet and gravel roads are a much easier perception problem than dense urban environments, and perception is one of autonomous driving's weak spots. Conversely, quick physics computation and response is a relative weakness of human drivers, and is exacerbated by these environments.

So this model works well - rather than paying $10k for a feature that only works for 100 hours a year you spend on the freeway, and not the 400 hours you spend on your dirt road in smallville, you get the benefit when you go and visit Aunt Mabel without having to pay a large fee upfront.

> Humans are terrible drivers, even in the best cases.

Humans drive correctly 99.99999% of the time. We can't can't even keep a computer up and running that well, never mind driving perfectly.

See this thread with the math: https://news.ycombinator.com/item?id=12012166

The future you are expecting is never going to happen.

> Humans drive correctly 99.99999% of the time. We can't can't even keep a computer up and running that well, never mind driving perfectly.

This comparison doesn't make any sense. You're comparing _continuous_ operation with the aggregate of the average human's bouts of driving. Even robotaxis aren't going to run constantly. A sensible analogue would be something like aircraft software, which I'd guess does have near the reliability of human drivers (and can last much longer continuously than humans do).

> You're comparing _continuous_ operation with the aggregate of the average human's bouts of driving.

I don't understand what you mean. I'm comparing actual time behind the wheel - hours of correct driving vs hours (well minutes) of incorrect driving.

Do the math yourself, take total hours behind the wheel and divide by the accident rate, like I did. You'll get the same results. Depending on your assumptions anywhere from 99.999 to 99.99999% correct driving by humans.

> Even robotaxis aren't going to run constantly.

Again, I'm not comparing constantly, I'm comparing actual hours behind the wheel.

84 billion hours of driving[0] for 35,485 deaths[1]. That's one death per 2.4 million hours, so that's an indicator of the magnitude that's needed before even talking about it.

That number also means no situations where the computer gives up and hands control back to the driver.

If autonomous cars can never hand back (at most stopping legally, safely and out of traffic) and cause 1 death per 10 million hours, that would seem about right.

[0] https://www.volpe.dot.gov/news/how-much-time-do-americans-sp...

[1] https://en.wikipedia.org/wiki/Motor_vehicle_fatality_rate_in...

Is this not a reference to uptime (as in servers)?

> We can't can't even keep a computer up and running that well

That's the comparison that I quoted in my comment and that I was responding to. You're comparing continuous uptime to humans' disconnected (and brief) bouts of driving.

>A sensible analogue would be something like aircraft software

Does automobile software development have the level of oversight that aircraft development has in terms of regulation and approval for use? I know the NHTSA approves safety with crash tests (and are currently working on AV standards) but I'm not sure it's an equivalent to what the FAA currently does.

Not to say they won't be equivalent in the future, but I don't know how well the analogue holds up if they don't have equivalent levels of quality oversight currently.

> Does automobile software development have the level of oversight that aircraft development has in terms of regulation and approval for use?

Nothing in my comment implied this, and I don't disagree that it doesn't. The question is immaterial to the specific topic of discussion: the comparison of server uptime to human-driver reliability as an indicator of the feasibility of functional AV software.

The tone of your response seems to imply I’ve was attacking your position, but that wasn’t my intent.

I am legitimately asking how much parity there is regarding oversight regulation between flight and autonomous software, which would help make your point clearer since it has a direct relationship to reliability.

However, I do think it’s relevant to the point. If you are making an analogy about reliability but one is legally forced to demonstrate reliability while the other is not, does that comparison really make sense with that additional context? While I've worked in the aerospace software domain, I don't know to what standard AV is held which is why I was asking.

> seems to imply I’ve was attacking your position, but that wasn’t my intent.

Not quite attacking, but disagreeing with an implied assumption that I didn't make (ie, that I was arguing that AVs were ready for primetime instead of narrowly disputing the flawed reliability comparison in the parent comment).

> I am legitimately asking how much parity there is regarding oversight regulation between flight and autonomous software, which would help make your point clearer since it has a direct relationship to reliability

The reason I figured it was rhetorical is that I thought it was fairly universally known that there's no comparable regulatory apparatus for AVs (yet?). But you're mistaken that this is relevant to my point: the parent comment was discussing the feasibility of AVs by saying "we can't build computers with the uptime of human drivers", and I was pointing out that he had this backwards.

> If you are making an analogy about reliability but one is legally forced to demonstrate reliability while the other is not, does that comparison really make sense with that additional context?

I didn't say AVs were as reliable as aircraft software, nor that they would definitely get to that pt (perhaps they don't need to, as safe stops are much easier on the ground). I was disputing the claim that computer system uptime can't be made as reliable as humans in the current day, using aircraft systems as an in-production example.

Ah, ok. Thanks for taking the time to clarify. It sounds like we’re making the same point but maybe talking past one another.

My point was that AV-aircraft comparison is only valid iff they are held to the same quality standard. It sounds like we agree they are not.

I think it’s important to include that caveat, that’s all. If you assume everybody already knows that, I can understand why you didn’t include it in your original point.

Averages are extremely misleading here, on the human side you're lumping drunk drivers (that could be precluded from driving by the simple mandatory adoption of interlock system), commercial drivers, people driving in really bad weather, across all types of roads and with cars in variable state of maintanance. Remove all those factors (specially DUI) on the human side, then add all those conditions were self-driving doesn't even attempt to try, and the sheer superiority of the human visual cortex alone is indisputable.

And then you forget to click a "don't use autopilot" setting buried in three levels of settings, and will pay for a month-worth of driving.

Seems to me it would be difficult to not notice your car driving itself.

It doesn't matter if the car drives itself or not. What matters is how money is withdrawn.

It's enough to have "on/off" somewhere in settings, and have it "enabled" even if you don't use it.

Good grief, there are countless articles even here on HN about dark patterns where people are tricked into subscribing to something, or being unable to unsusbcribe except when going through the hoops. But sure, a car manufacturer charging maney for a subscription will never ever do that as long as you "don't use a feature".

You're inventing hypotheticals to say why they're going to charge you more, without evidence.

Pay-per-minute autonomous driving is far more like old pay-per-minute phone plans than it is mystery-AWS-data-charges. You know when it's driving and when it's not. There's no unforeseeable per-lambda charges about it.

I think it's reasonable to worry that WV or another company might start charging a variable rate for different "types" of autonomous driving. Pay-per-minute phone plans could be just as mysterious when you factored in long distance and peak usage hours.

It's not a huge logical leap to think that a subscription based service like this could lead to things such as:

You want to travel between point A and B several miles apart in rural Kansas on a straight road with a road density of 2 vehicles per mile? "Sure, that's 8.50 an hour."

Oh, you want to travel between Newark and Manhattan in the middle of a snowstorm? "That will be 15 an hour because we have to cover the costs for the real-time feed of data from our weather and control center"

Oh, you have a camper, too? "That's an additional 2 bucks an hour for our towing package"

Oh, you're traveling across state lines and the state over imposes an additional tax on autonomous driving. "we need to charge you 8.75 to cover the tax and our costs but only in that state."

> You're inventing hypotheticals to say why they're going to charge you more, without evidence.

Ah yes. There's no evidence. Except the countless instances of dark patterns everywhere on the web tricking into you paying money. Countless instances where you can't cancel a subscription, and have to go through multiple hoops to do it. Countless instances when there's a runaway billing in a SaaS that leaves people with huge bills.

But sure. These are all just "hypotheticals", and not the reality how the absolute vast majority of subscription business works.

> You know when it's driving and when it's not. There's no unforeseeable per-lambda charges about it.

Ah yes. Because of course it will be as simple as "when it's on, it will be charged", and not "when an obscure setting is on, you will be charged, regardless of whether it will be self-driving or not".

There is very little reason to believe Tesla will be best let alone first when it comes to self driving technology. I personally feel they are backing the wrong technology and will ultimately be unsuccessful without LIDAR sensors like Waymo and other startups and the beta testers with current model Teslas are going to ultimately end up disappointed.

I don't own a car currently (nor any Tesla stock) so I'm happy to wait and see if I'm wrong or right.

Tesla can already generate a pretty good "lidar point cloud" out of just camera input.

But we'll see.

One issue with Tesla's approach is that they do not have enough cameras to do this at every angle. Its going to be especially tough in parking lots, as they have very poor coverage of the road directly beneath them.

Mobileye's camera based point cloud approach appears to be much more robust.

Why would the car need to see the road directly beneath while parking? Because of positional error (drift) or what? It should know the painted markings from earlier frames.

Good point though. I'd like Tesla to add more cameras.

Reply on news.ycombinator.com