Elon: "Feature complete for full self driving this year"

[daniel]

Take what you pay for insurance. Divide it by 10. Multiply it by the number of FSD cars Tesla has sold.

Now multiply it by 10 for the inevitable negligence claims because they are a corporation and their engineering process will be under scrutiny.

Insurance is just part of the cost of doing business. It's passed on to the customers. Everything you buy has built into its price the insurance the company pays. Taking your numbers (which I don't actually agree with) the cost of insurance per car stays the same as now. But instead of each person paying for their own, the car company pays and passes the cost on to us when we buy the car. Bottom line: No difference.

But that does not kick in until the cars reach level 3, where the car actually takes responsibility. At level 2 the driver is fully responsible. We've had basic autopilot for five years and it's still Level 2 and a beta feature. "Feature Complete" is at least a year away, and I'm guessing three. After that there's at least another five years at level 2. That's at least 8 years before Tesla is willing to say its cars are responsible for decision-making, and we see any change in the way cars are insured.

 

[banned-66611]

Oh, and don't forget that commercial insurance for taxis is a lot more than personal insurance.

 

[willow_hiller]

Oh, and don't forget that commercial insurance for taxis is a lot more than personal insurance.

How much does commercial insurance for robotaxis cost?

 

[Daniel in SD]

Oh, and don't forget that commercial insurance for taxis is a lot more than personal insurance.

Is that because taxis drive way more miles than an average driver?

How much does commercial insurance for robotaxis cost?

That probably depends on robotaxis driving record.
In California the manufacturer needs to have $5 million is coverage.

 

[rnortman]

But that does not kick in until the cars reach level 3, where the car actually takes responsibility. At level 2 the driver is fully responsible. We've had basic autopilot for five years and it's still Level 2 and a beta feature. "Feature Complete" is at least a year away, and I'm guessing three. After that there's at least another five years at level 2. That's at least 8 years before Tesla is willing to say its cars are responsible for decision-making, and we see any change in the way cars are insured.

That seems like a realistic time frame, except I'd add that they may achieve "feature complete" on the current generation of hardware but the current generation will always and forever be level 2. The next generation might be level 3 or higher, years from now, and they will not retrofit anybody because outside of tweets they never promised that FSD meant you weren't responsible for paying attention at all times. And by then, there will already be level 4 cars from other companies that have been running in production for a while, so Tesla's level 3 will be a big yawn.

 

[Knightshade]

That seems like a realistic time frame, except I'd add that they may achieve "feature complete" on the current generation of hardware but the current generation will always and forever be level 2. The next generation might be level 3 or higher, years from now, and they will not retrofit anybody because outside of tweets they never promised that FSD meant you weren't responsible for paying attention at all times. .

I mean, they absolutely did promise that to buyers prior to March 2019.

"The system is designed to be able to conduct short and long distance trips with no action required by the person in the driver's seat."

Supervising, or intervening when needed, are actions from the person in the drivers seat. Which the FSD sold to such buyers explicitly says will NOT be required.

So for those guys they'll need to keep upgrading the HW till they meet that promise- or refund those folks I suppose and deal with the lawsuits from those who claim they bought the car specifically based on that promise and want buybacks.

For post March-2019 buyers they could probably do what you suggest though.

 

[daniel]

... The next generation might be level 3 or higher, years from now, and they will not retrofit anybody because outside of tweets they never promised that FSD meant you weren't responsible for paying attention at all times. ...

When I bought my car (and declined FSD because I didn't believe they could accomplish it on that hardware or while I owned my car) the explicit promise was that if I bought FSD I'd be able to use my car as a robotaxi, or send it empty to pick up kids and bring them home, or take them to school and drive itself home.

Elon absolutely promised what amounts to Level 5 autonomy on my car and with the hardware it had at the time of sale. He explicitly said my car had all the needed hardware. A year or so later he stopped making that promise when he realized that goal was too far away, but rather than admitting that he made a mistake and giving people their money back, he just quietly backed away from his promises and began describing what amounted to a Level 2 system that would extend Navigate-on-Autopilot to city driving, with a fully attentive driver ready to take over at his/her own initiative and without prior notice from the car.

Tesla has accomplished more than any other car company in bringing electric transportation to drivers and producing the most advanced driver's aids available for purchase to date. These are, by any measure other than towing ability, the best cars you can buy today. And the truck will remedy that one shortcoming. Elon just has to admit that he cannot fulfill the promise he made to early FSD buyers and deal with it by compensating them, and then move on with the job of developing the software and hardware needed for full autonomy.

 

[croman]

Analogies aren't great, but I think this fits:
Goal: Riding a two wheeled bike solo:
Approach 1: Training wheels (additional hardware to start that is not needed later)
Approach 2: A parent supporting the bike and walking/ running along side (more work needed during development, longer training period, less material cost)

Approach 3: Watching videos of other kids riding bikes and then hopping on and falling on your ass a bunch until you figure it out or are too broken (or bike) to continue.

 

[daniel]

Approach 3: Watching videos of other kids riding bikes and then hopping on and falling on your ass a bunch until you figure it out or are too broken (or bike) to continue.

Approach 4 (what happened to me): Put the kid on a bike that's way too big for him so he cannot touch the ground, hold onto the seat and promise not to let go, then break your promise and let go. Theory: Everybody instinctively knows how to ride a bike, and the kid just needs to realize that. Reality: Kid stays up for about five seconds after they let go of the seat, and then falls, because, damnitall, it just ain't true that it's instinctive. You have to learn how to stay upright.

 

[alcibiades]

they can! have people not seen this yet? it surprised me. its a little hard to see in the photo but the traffic light is a left red arrow and its visible in the console! not sure if this came with the 2020 sw update or was part of the old .7 release.

View attachment 508522

That's fascinating. I've looked really hard, and I haven't seen it. Nor have I had "what should be" a turn arrow light switch green when the arrow goes green. Shrug.

 

[Cirrus MS100D]

That's fascinating. I've looked really hard, and I haven't seen it. Nor have I had "what should be" a turn arrow light switch green when the arrow goes green. Shrug.

And you’ve got HW3, right?

 

[diplomat33]

Here is what Jim Keller, who was Tesla's VP for autopilot hardware engineering from 2016 to 2018, says to Lex Fridman in his recent interview:

"You don't have to be especially smart to drive a car so it's not like a super hard problem. I mean the big problem with safety is attention which computers are really good at, not skills."

"You can drive a car with 20/50 vision and you can train a neural network to extract a distance from any object and the shape of any surface from a video and data. It's really that simple... It's a simple data problem."

It is mind-blowing how naive and clueless he is. Even Lex Fridman pushes back several times and says that it is not that easy. I am not defending Tesla but I think it does explain a lot. Since guys like Elon and Jim apparently though that FSD would be easy and a simple data problem, it is not surprising that Tesla pre-sold FSD in 2016, claimed that AP2 was "FSD capable", promised "sleep in your car" FSD in a year etc...

I do think that Tesla will eventually get to "feature complete" and I think the new FSD rewrite is a good step in that direction. With the new FSD rewrite and the new push for more complicated neural networks, it seems Elon might be coming around now to how hard FSD really is. But it is obvious to me that Elon and co. grossly underestimated the problem of FSD when AP2 was launched. Jim Keller's quotes really support that.

 

[mikes_fsd]

It is mind-blowing how naive and clueless he is.

To be honest, the only "naive and clueless" person in this scenario are those who have NOT worked on the problem.
This man's experience and his expertise is not in question.

Yours on the other hand...

 

[diplomat33]

To be honest, the only "naive and clueless" person in this scenario are those who have NOT worked on the problem.
This man's experience and his expertise is not in question.

Yours on the other hand...

Oh, I am not questioning Jim Keller's expertise. He helped designed AMD chips and Apple's A4 and A5 chips. When it comes to chip design, he knows his stuff.

But when it comes to understanding FSD, his statements are very naive and clueless. Saying that driving is easy and does not take skill and that computer vision is a simple data problem are absolutely naive and clueless things to say.

 

[jimmym1981]

Oh, I am not questioning Jim Keller's expertise. He helped designed AMD chips and Apple's A4 and A5 chips. When it comes to chip design, he knows his stuff.

But when it comes to understanding FSD, his statements are very naive and clueless. Saying that driving is easy and does not take skill and that computer vision is a simple data problem are absolutely naive and clueless things to say.

I really think to get to L5, the traffic lights and stop signs will need a computer of their own to communicate with the cars. There are just too many unknowns when it comes to city driving.

 

[rnortman]

To be honest, the only "naive and clueless" person in this scenario are those who have NOT worked on the problem.
This man's experience and his expertise is not in question.

Yours on the other hand...

It is clear from the opinion he expressed that Jim Keller has not actually worked on problems of autonomous vehicles or robotics in general, though he clearly knows how to design silicon. Neither had Elon, in any serious way, when he made those pronouncements in 2016.

 

[Daniel in SD]

Many very smart people have been working on self driving for a decade. Tens of billions of dollars have been invested. If the problem were easy it would have been solved by now.

 

[AlanSubie4Life]

If the problem was easy it would have been solved by now.

I'm going to take issue with you not using the subjunctive conjugation in this sentence.

What’s the Difference Between Was and Were?

 

[Daniel in SD]

I'm going to take issue with you not using the subjunctive conjugation in this sentence.

What’s the Difference Between Was and Were?

fixed!

 

[alcibiades]

And you’ve got HW3, right?

Yup.