Hacker News Reader

Comment by NickM | Hacker News Reader

Article is misleading. Tesla sent a letter to the CA DMV requesting approval for a feature which they referred to as "City Streets" which they say is level 2, and will continue to be level 2 as it is developed. They specifically say in this very document that they will develop higher levels of autonomy in the future, so it's pretty clear that the "final release of City Streets will continue to be level 2" quote is just referring to the particular subset of functionality that they're seeking approval for.

From the very same letter, there is also this quote:

Please note that Tesla’s development of true autonomous features (SAE Levels 3+) will follow our iterative process (development, validation, early release, etc.) and any such features will not be released to the general public until we have fully validated them and received any required regulatory permits or approvals.

View on HN

That is where there is a disconnect between the Tesla marketing language and technical language.

They publicly advertise and collect cash for pre-sales of "Full Self Driving". Most Hacker News technologists (a generalization for sure) would reserve a phrase that bold for Levels 4 and 5.

But here they are in writing stating that they are currently only at Level 2.

So a couple core questions around Tesla are:

- Is it acceptable to take pre-orders from the average consumer for a feature that might be 10 years away?

- Is it ethical to use exaggerated marketing speak for specific technical functions (to give a non-Tesla example, this is like when I see companies implementing rule-based decision trees or linear regression and calling their apps "Artificial Intelligence")?

It's not just technical and marketing, it's legal too. And this article references emails from tesla legal team. So legal is telling dmv that it's L2 so it doesn't have to get permits, but that doesn't mean the tech isn't destined to be L3 when it's fully developed and they are ready to submit themselves to regulatory scrutiny.

Your theory here is they're purposefully downplaying their level of self-driving technology to avoid regulatory processes?

Are they going to keep winking at customers and letting them think it's level 3 so they continue to take their hands off the steering wheel and do other things like countless Youtube videos already capture?

It's a little disturbing how people will jump through so mental hoops to give Tesla the benefit of the doubt.

I don't really see that as giving Tesla the benefit of the doubt. I read it as describing what may be a ploy to avoid the regulatory processes and that may be a valid point.

> this is like when I see companies implementing rule-based decision trees or linear regression and calling their apps "Artificial Intelligence"

AI means so many things that it has become as meaningful as "magic". Some don't consider anything less than human-level an AI (this is also called AGI or strong AI). On the other hand enemy behavior in video games is commonly called "AI" even when it is as simple as pacman ghosts. In the same way that magic can literally mean "impossible" or be applied to card tricks accessible to 5 year olds.

"AI" is ambiguous but "Full Self Driving" has a pretty clear plain meaning. It means that the car is self-driving. Not partially self-driving, not mostly self-driving, fully self-driving.

To the layperson, "full self driving" means "I can ride it drunk without going to jail"

And I think that's the meaning car companies should use in public communications. I get why it's helpful for automotive engineers to talk in more detailed shades of grey, but for the rest of us, a car where I can't prop up my feet and start playing Minecraft is "assisted driving".

This disconnect seems to be growing globally, be it marketing, technical or legal.

It seems strange considering the wide availability of information and communication, but apparently there either is no feedback loop back to the people who write like that, or there is but it's mostly positive (be it more sales, less lawsuits or a greater following). At some point the meaning of those words will have hollowed out so much that we can't use them in normal conversation anymore.

But they haven’t financially recognized the full amount paid for FSD as revenue...

They’re basically long-term 0% loans like the Cybertruck deposits. They keep it in escrow and slowly recognize it feature by feature.

Cynical take: Tesla GC says "We completely understand and agree that we won’t deploy any autonomous vehicle feature without a deployment permit" [ https://www.plainsite.org/documents/242a2g/california-dmv-te... ]. Absolutely certain that Tesla won't take liability for city driving unless they absolutely must. Especially when so many Tesla drivers are so open to accepting the liability. If Tesla's competitors (e.g. Waymo etc) launch taxis, then Tesla might change course and do the same. But if Tesla doesn't have to do that, they're going to play a game with California where they get to argue over if their cars are level 2 or actually self-driving or whatever .... all the while Tesla users continue to assume liability and help train Tesla's software.... which is what Tesla really wants to continue as long as possible.

These docs don't really shed light on Tesla's FSD plans so much as they illustrate the lengths to which Tesla will go to skirt regulation and liability. All the chatter and rumors about Tesla FSD just drive more attention and more potential users (trainers) into their pool.

Having more "trainers" probably doesn't matter one bit to Tesla, but their $10k is quite useful

> Tesla sent a letter to the CA DMV requesting approval

Can you cite that? It directly contradicts the linked article which claims that the CA DMV contacted Tesla to clarify if Tesla were testing a vehicle without proper permit.

If Tesla were seeking some kind of approval it makes their email very confusing as Tesla is insisting that they don't need additional permits/approval for the "City Streets" beta.

Here's the letter: https://www.plainsite.org/documents/242a2g/california-dmv-te...

That seems to confirm what the article said. The first email appears to be dated December 20, 2019 from the CA DMV to Tesla.

Here's the original email sent to Tesla:

> Subject: Announcement of Full Self Driving Feature.

> Al we’ve seen press reports that Mr. Musk announced on Twitter that a “Tesla holiday software update has FSD sneak preview...” Many people generally translate “FSD” to be “full self-driving.” Is Tesla releasing a full self-driving software update to any California Tesla owners? As you are aware, the deployment of autonomously driven vehicles on public roads in California requires a permit to deploy. At this time Tesla does not have a permit to deploy. Please provide an update on what this announcement for the deployment of the feature means in terms of autonomous operation on public roads in California?

I don't see anywhere where Tesla are seeking approval, quite the opposite in fact. Tesla are saying they shouldn't be required to seek approval for "City Streets."

Oh I see what you're saying. I believe they are "seeking approval" for City Streets ... as a Level 2 product. Likely some discrepancy between what Musk says and what the rest of the company does.

They were already approved for Level 2 driving, the DMV is asking if City Streets is level 2 or per Tesla's own statements FSD which they were not permitted for testing.

If "Tesla sent a letter to the CA DMV requesting approval" per the above comment, that would have meant Tesla were trying to get an FSD permit in CA (the opposite of what happened).

"City Streets" is part of "Full Self Driving" $10,000 package though. They are calling it "Full Self Driving", but a subset of functionality will always be level 2? And the FSD Beta is also level 2 because it only works on city streets? That seems clearly misleading.

What's next? "Really, truly full self driving" that's actually level 3+?

These are incremental releases of gradually improving functionality. Their plan is to get to level 3+.

People who buy it know what they are getting, and if they don't, it's entirely willful ignorance. I don't even own a Tesla yet I've read plenty about the actual capabilities and the fact that these cars can't truly drive themselves.

I don’t think we’ll see “full self driving” within the lifespan of several cars from now.

Yet, if I bought a Tesla I’d still need to sweat the decision of buying the FSD option because Musk insists the resale value of cars without it will be significantly smaller without it.

I always assumed by "full" they meant that it doesn't just do autopilot on highways, but is capable of handling local roads, traffic lights, stop signs, as well. Basically it can drive from local address A to local address B assuming good conditions.

I'm a robotics person though so it was always pretty clear to me that I couldn't possibly think of being able to deploy an L4 system with their set of sensors.

When you call it "full" and "self driving", I don't know how many qualifiers you can attach to it. I know it's not anywhere close to driving by itself, but between the naming and Elon Musk constantly talking about how level 5 (I mean, level 5 is the ultimate form of autonomous driving) is just around the corner, it's pretty easy to see that as misleading customers. All while they don't even bother to get a driverless testing permit from CA DMV.

Yeah, of course. But it was always clear to me.

The problem is (a) the stock market is full of idiots that won't listen to real engineers like us, and listen to some bullshitty CEOs and analysts instead (b) the SEC and stock market is perfectly optimized and incentivized to maximize the amount of marketing bullshit coming out of company PR.

If people wanted more honestly worded things out of companies, we need to stop nailing companies every time they don't "beat" earnings estimates every quarter, and begin to understand how real science and real technological advancement happens.

The way NASA is funded, for example, is the exact opposite, and minimizes bullshit, but NASA and Tesla share a lot in common in terms of what they're trying to accomplish for humanity.

EDIT: Okay maybe NASA wasn't a good example. But in any case, the current structure of public companies and letting a bunch of gullible <strike>sheep</strike> average citizens collectively determine the value of a company based on what a bunch of <strike>clowns</strike> non-engineers in suits at Wall Street try to make of a highly technical venture isn't a system optimized for minimizing bullshit, it's quite the opposite.

> The problem is (a) the stock market is full of idiots that won't listen to real engineers like us

So you are a real engineer and Musk is not?

> begin to understand how real science and real technological advancement happens

By pushing for it hard, investing your credibility and a lot of man power and resources into it? Like literally what Elon Musk does with all his projects.

> The way NASA is funded, for example, is the exact opposite, and minimizes bullshit

NASA claims nonsense literally all the time. The #JournyToMars under Obama for example.

> So you are a real engineer and Musk is not?

Real engineers don't bullshit. Real engineers understand that there is a certain responsibility that comes with people trusting their lives to their engineering skills. Real engineers can give you confidences on everything in hard numbers with data. Real engineers can provide realistic estimates of timelines. Real engineers make mistakes sometimes, admit them candidly, and know perfectly well how to adjust those timelines based on past estimate errors and extrapolate adjusted project timeline estimates with basic math. Real engineers know the limits of systems and how to write spec sheets that describe them accurately. Real engineers don't always know the answer to every question, and know when to say "I don't know".

It's managers, investors, analysts, and sometimes customers who don't want to listen to those extrapolations and estimates and uncertainty. And so many real engineers gradually become increasingly fake, because the other (corrupt) parts of the system spread their toxic thinking into the engineering.

I have a lot of respect for Musk, who was a real engineer, probably REALLY wants to stay a real engineer, and is still mostly an engineer. But he is indeed forced to be a part-time <strike>bullshitter</strike> non-engineer by the system, which demands bullshit by its very incentive structure.

For Musk his bullshitting is still nowhere close to the competitors, who said for a long time that they can manufacture millions of electric cars if they really want to.

When I told my friend about self driving Tesla who's a BMW fan, he said that BMW will just sell self driving cars when it's ready.

Really? Without gathering any data? I'd say BMW is the bigger bullshitter. Elon needs to update the computers maybe, but everything else in the cars is ready for self driving im the near future.

> his bullshitting is still nowhere close to the competitors

I mean, as I said, the whole stock market is full of bullshit. My point is the system is designed to incentivize bullshitting, so it's expected given the incentive system created. Companies that don't bullshit don't win, and if that's the loss function you define, bullshit is what you're going to get.

> everything else in the cars is ready for self driving

I disagree with this. I think significant hardware upgrades to Tesla cars will be necessary to achieve L4. Any attempts at L4 without those upgrades will result in a lot of accidents. I won't be surprised if they come out 5 years from now asking former Tesla owners to get their cars retrofitted with some new sensors if they want L4.

L2 is achievable with their existing hardware and that's what they are doing.

They already evaluated retrofiring sensors. Doing it would cost about as much as a new car. They've made a bet that this sensor suite will be enough and I'm sure Musk got green light on it from his team. Time will tell if they were right.

I personally agree with them that technologies like PseudoLIDAR will be sufficient.

I believe that for a good human driver Tesla sensor suite would be sufficient to drive safely if you covered all windows. You could project the camera views on covered windows and keep the current bird eye visualization.

The issue really is with control of the car. Deep Reinforcement learning sucks. Training Deep RL is unstable, hyper-parameter sensitive, even random-seed sensitive, and requires ridiculous amount of data for even the simplest applications. And after training it does not generalize.

>everything else in the cars is ready for self driving im the near future

Completely false.

They can get a license to use Waymo hardware/software in their cars, right?

> But it was always clear to me.

So when the CEO went on stage and said cars would be self-delivering via fully autonomous driving it was obvious to you that his statement was a lie?

Yes. I know not to listen to CEOs when they say things like that. They're effectively puppets.

I've also probably coded way more lines of autonomous driving code than Elon Musk himself. That isn't to say I don't respect him and his vision, but I'm pretty certain I have a better idea of timelines than the BS he's forced to say to the public.

In my opinion Musk actually believes in his timelines. He does exactly the same mistakes without financial incentives e.g. Falcon Heavy. His "COVID panic is dumb" and "it's gone by April" are memes by now. Then he will also openly say that he believes shares are too high, so of course shares dropped. Not to mention that SEC is going after him and Tesla for his stupid tweets.

Either he is an alien playing 5 dimensional chess with time-travel, or he is pathologically overoptimistic. He is especially overoptimistic about AI, so it leads him to make stupid predictions about when we will have cars without steering wheels.

> They are calling it "Full Self Driving", but a subset of functionality will always be level 2?

Well, yes? That's how subsets work.

Thought experiment, pretend "City Streets" was called "Level 2 Mode". Would you be upset that "Level 2 Mode" will always be level 2? Or that "Level 2 Mode" is a subset of "Full Self Driving"?

I’d be upset that level 2 mode was released 3 years after they promised level 5 would be ready.

> Thought experiment, pretend "City Streets" was called "Level 2 Mode".

That would be an accurate name for "city streets". Except currently, it's part of an umbrella feature called "full self driving" that's supposedly going to be "level 5" soon. This is it with most things with FSD. Qualifiers after qualifiers for a clearly misleading name.

It's funny people think Level 2 is a subset of Level 4 and 5.

Kind of like calling NNs a "subset" of General AI

The functionality is a subset, yes.

No, that's not how subsets work.

If a subset of features that is part of "full self driving" cannot ever reach level 4/5, then "full self driving" is not possible.

"full" is a marketing terminology but the English meaning of it would be that it works comprehensively in the majority of cases. If it doesn't work on city streets, which is a major subset of use cases, how can it be "full"?

Did you skip the rest of my post?

There is an enormous difference between "city streets" the location, and "City Streets" the name.

They're not saying the car will be stuck at level 2 on streets in cites.

They're saying that this particular feature which is named "City Streets" will always be level 2.

Eventually (they promise) the car will have level 4 driving on city streets. Level 4 driving inside cities will either have a different name or no name at all. At that point they will probably delete "City Streets", because it's a subset of level 4 and there's no reason to keep it after level 4 is finished.

Edit: especially because the full name is "Autosteer on City Streets". When they have level 4 support, they definitely won't be calling it "autosteer"

if people watched the fsd videos on youtube, they would understand that fsd beta is full self driving, but it requires driver attention and gives the user a quick way to take over control.

I think the only one that doesn't get this are journalists, regulators, and class action lawyers.

> "Really, truly full self driving" that's actually level 3+?

Prediction:

* Ridiculous Self Driving

* Ludicrous(+) Self Driving

* Plaid Self Driving

Yeah, they just keep moving the carrot. By the time Tesla comes with FSD the current model lines will have reached their end of live and the $6k+ that people paid will be lost with the car.

An attorney general from a state will start to go after them if they don't come up FSD any time soon. My guess is the settlement will be that they can carry it forward to another Tesla. No of the features really work with the current FSD like smart summon. And there's probably tens of millions of money poured into a feature that doesn't limit up to expectations.

If they were performing as their CEO represented, the cars would be have driving themselves cross-country to delivery themselves to customers years ago. The time to have them up about lying what would be delivered and when was a looooong time ago.

If you think it trough, Tesla is misleading in making anyone velieve their curren tech will ever have full self-driving capabilities.

What's really required for full self-drivung capabilities is infrastructure support. This will come as large eatablished car manufacturers and governments start defining standards for that.

Here's a very simple example. Imagine a stop sign. Tesla's and Google's aproach is to detect the stoo sign via cameras. This can be problematic when the canera can't see the stoo sign (weather, tree branches blocking it partly, a truck in front of you). It would be much simpler if the stoo signbwas able to communicate to self driving cars that they have to stop. Now imagine an accident because your Tesla didn't recognize the stoo sign. Who is going to pay for the damages? Tesla for false advertising? The city for not cutting down the three? Your insurance? You?

This is precisely why fully self driving on existing infrastructure won't happen. I don't understand how anybody can believe that it will.

Yeah, the word “current” in the headline is doing some heavy lifting

Reply on news.ycombinator.com