Tesla and Uber Unlikely to Survive (Vol. 2)

[Cheeses of Nazareth]

lothianJim said:

Human drivers don't perform consistently in all conditions. Yet insurance companies still have to estimate risk for different occupations, ages, engine size....

I got a quote for my performance, that was nearly twice the price that I was quoted a few days earlier. Turns out buying a 500bhp car with one days notice, was interpreted to mean I was more likely to make compulsive and unpredictable errors than someone who planned weeks ahead.

If anything, I would guess actuaries have an easier time working out premiums for cars that are mostly driven on autopilot, as even with all it's flaws, it's probably more predictable kind of unpredictability, than the average human driver.

But you are a single risk individual, and assessed as such.

And when they find ( not that they have to look hard) the flaw, it wont be the driver that pays , it will be whoever is on the hook for the software.

Lets assume 1000 people drive a car into a wall... 1000 different reasons.. nobody cares 1000 different claimants , 1000 different people

Telsa drives 1000 cars into a wall... 1 set of software 1 reason.. Try mitigating that .

[off_again]

Brother D said:

They are already 10 times safer than normal cars running auto pilot on average:

https://cleantechnica.com/2020/08/01/tesla-autopil...

That is very misleading. Since Autopilot cant be used on certain roads, situations or conditions and the vast majority of users, just turn it on for longer journeys on comparatively safe freeways and interstates - it completely ignores the higher risk factor miles covered and paints it to be perfect, which it is not.

This is positioning from Tesla to try and sell this from a regulatory point of view to get it approved on the basis of security. Its misleading, frankly incorrect and hopefully the legislative and regulatory bodies see through the spin of statistics to support their marketing efforts.

[Tuna]

lothianJim said:

Human drivers don't perform consistently in all conditions. Yet insurance companies still have to estimate risk for different occupations, ages, engine size....

I got a quote for my performance, that was nearly twice the price that I was quoted a few days earlier. Turns out buying a 500bhp car with one days notice, was interpreted to mean I was more likely to make compulsive and unpredictable errors than someone who planned weeks ahead.

If anything, I would guess actuaries have an easier time working out premiums for cars that are mostly driven on autopilot, as even with all it's flaws, it's probably more predictable kind of unpredictability, than the average human driver.

Insurance companies don't come into it - this is about the law and who is legally in charge of a vehicle. By law, we have to have a test that takes into account how human beings react, and tests the 'edge cases' for our behaviour - can we see, can we control the vehicle, can we react and can we behave responsibly. Note that the test has baked into it the idea that a human driver with reasonable eyesight can tell the difference between a stop sign and an advertising hoarding.

Now how do we 'test' the software? We don't know if it suffers from 'digital' optical illusions (actually, we have a strong idea it does - phantom braking is one example, but researchers have shown how even tiny changes in a scene, equivalent to a cigarette packet lying in the road can fool neural nets into completely misinterpreting the scene). We don't know how it prioritises options (if a child steps out into the street, but a bus is in the opposite lane, which way will it swerve?), and we have no mechanism for checking whether a software update changes those priorities.

Software can absolutely speed up response times, and that will certainly reduce a particular type of accident. However, there are just as many situations where the software has to make a choice - and that is where we have a huge can of worms that doesn't just go away because cars with "fast brakes" are better at avoiding rear-end shunts. Basic things like reliable positioning in the road are key not only to avoiding your own accidents, but also not causing other people to have accidents avoiding you.

[Tuna]

off_again said:

That is very misleading. Since Autopilot cant be used on certain roads, situations or conditions and the vast majority of users, just turn it on for longer journeys on comparatively safe freeways and interstates - it completely ignores the higher risk factor miles covered and paints it to be perfect, which it is not.

This is positioning from Tesla to try and sell this from a regulatory point of view to get it approved on the basis of security. Its misleading, frankly incorrect and hopefully the legislative and regulatory bodies see through the spin of statistics to support their marketing efforts.

"People who've just bought brand new expensive cars have fewer accidents than average shocker"

[gangzoom]

How anyone can say this latest FSD beta build isn't a massive setup from anything else avaliable to the public is beyond me.

Again it would fascinating to see how Waymo would deal with on coming traffic and parked cars across half the lane.

https://twitter.com/kimpaquette/status/13207562610...

https://twitter.com/WholeMarsBlog/status/132059302...

Edited by gangzoom on Monday 26th October 19:04

[Gandahar]

gangzoom said:

How anyone can say this latest FSD beta build isn't a massive setup from anything else avaliable to the public is beyond me.

Again it would fascinating to see how Waymo would deal with on coming traffic and parked cars across half the lane.

https://twitter.com/kimpaquette/status/13207562610...

https://twitter.com/WholeMarsBlog/status/132059302...

Edited by gangzoom on Monday 26th October 19:04

It certainly is a setup, to keep the share price up to get Elon more money ...

Let's see how it goes. The interesting thing will be if someone makes a mistake whilst driving the car, then blames it on the FSD and sues Tesla, even when not Teslas fault. I can see Tesla getting snowed under by lawsuits ( you know how much people love those nowadays ) by third parties and Tesla owners.

Meanwhile this is interesting,

https://finance.yahoo.com/news/toyota-panasonic-ba...

Japanese aiming to get more efficient and catch up.

[gangzoom]

Now its slowing down for speed bumps at night, managing roundabouts (slowly), and tried to crash into a train!

Apparently this FSD beta build isn't actually running on Tesla's latest NN (HW4 maybe needed??). So the software is clearly got lots of scope for improvement.

Cannot wait to see wider release of this at some point .

https://youtu.be/RN5Qoei7v1k

[Smiljan]

It still has such a long way to go, even with the relatively simple roads in that video the driver has to take control so many times. It’s an impressive bit of kit but at the moment seems to increase the driver workload keeping any eye on what it’s doing rather than being a useful tool.

There’s a lot of talk on YouTube videos of data being fed back to the neural net to help it learn but no one has actually shown that this is happening or how it is possible. The data transfer from the cars would be enormous, does every Model 3 really update Tesla with its driving data in this way?

They still have the issue that it’ll cut out completely in heavy rain, snow, dirt on the cameras etc... but I guess that’s why they release the beta to owners in typically dry desert like states.

This still seems so far away from self driving robotaxis that were promised this year, perhaps it’s getting better slowly but progress is just that, slow.

[Order66]

gangzoom said:

How anyone can say this latest FSD beta build isn't a massive setup from anything else avaliable to the public is beyond me.

I agree with what you're saying, but that still doesn't mean its good. In many ways it drives like a bad learner - hesitant, unpredictable, inappropriate actions, and there has to be a fully licensed driver alongside poised to take control at any second - a long way from "Full Self Driving"

[NDNDNDND]

gangzoom said:

Now its [...] tried to crash into a train!

Cannot wait to see wider release of this at some point .

Ummm...

[Gandahar]

gangzoom said:

Now its slowing down for speed bumps at night, managing roundabouts (slowly), and tried to crash into a train!

Apparently this FSD beta build isn't actually running on Tesla's latest NN (HW4 maybe needed??). So the software is clearly got lots of scope for improvement.

Cannot wait to see wider release of this at some point .

https://youtu.be/RN5Qoei7v1k

As a Tesla critic I think it would be unfair to get too wound up with this beta software having a few issues. As you say, lots of room for improvement and also Tesla have taken that step, unlike others so far. Pushing the envelope.

Having said that ( critic hat back on) what version of the hardware and software are these beta testers running? The hardware needs to be 4, as you rightly said and also the new beta "ground up" software that Elon said.

Anyone else with a Tesla and who doesn't have those two things, even though they paid a lot for it, are going to have to wait for upgrades...

You know there is going to be a HW5 as well .. for the promised land.

I get the feeling Musk thought he had to do this because of all his past promises for 2020. He's better sticking to trying to get Chinese people to buy Chinese made Teslas rather than excess shipping to Europe .....

[jjwilde]

I wonder how long it will be before it can drive from John o' Groats to Lands End without interaction (only stopping to charge)?

I'd say that would be a pretty significant achievement.

[Digga]

jjwilde said:

I wonder how long it will be before it can drive from John o' Groats to Lands End without interaction (only stopping to charge)?

I'd say that would be a pretty significant achievement.

Yet what chance of driving from one end of Birmingham to the other, on a busy rush hour when M6 and M5 are snarled up to buggery, without a band? I'd give it zero. No computer can compensate for the sort of random driving the real world throws up.

[Gandahar]

jjwilde said:

I wonder how long it will be before it can drive from John o' Groats to Lands End without interaction (only stopping to charge)?

I'd say that would be a pretty significant achievement.

The thing is that if you wanted to drive that route then you would want to actually drive it. For the experience. Otherwise get a taxi.

This is Pistonheads, so not sure why people saying "not driving" is a good thing.

Forgetting 700+ miles for the moment, apparently warranty wise Tesla will not cover you for a quarter mile



"YOU WILL HAVE TO DEAL WITH IT FOR NOW"





That is the writing on the wall for Tesla if they keep that attitude into 2021 when more and more EVs come on stream.

They have always been a company more concerned with selling cars and getting quotas rather than support afterwards. That is their achilles heel.

I like the expression " From previous experience you will most likely eventually get a fault but that is what needs to happen before parts can be replaced"

So they know it's going to blow up, but rather than be preventative, they say just wait until it fecks up, then suffer ..

He has to just deal with it for now .

[lothianJim]

I haven’t seen many ‘stop start traffic’ or motorway experience track days.

Personally I’m much more likely to go for a leisure drive if I can automate the boring bits in between.

[Gandahar]

lothianJim said:

Personally I’m much more likely to go for a leisure drive if I can automate the boring bits in between.

Maybe the Model X has roof racks that can fit an MX5 on the top, would suit you sir.....


I never give much head to sentences starting off Personally ... obviously you are typing out this marketing blurb.

[lothianJim]

Gandahar said:

Maybe the Model X has roof racks that can fit an MX5 on the top, would suit you sir.....


I never give much head to sentences starting off Personally ... obviously you are typing out this marketing blurb.

I’ve served my 2 seater time already. these days I’d rather hoon up to the highlands in comfort with a bike in the boot.

Autopilot on A roads. 480bhp for the B roads... no fun at all

Edited by lothianJim on Thursday 29th October 16:16

[Order66]

Gandahar said:

They have always been a company more concerned with selling cars and getting quotas rather than support afterwards.

Yeah, I picked up my Telsa a month ago - by the time I had got home from picking it up I had figured that the right indicator was running at double-speed, something that isn't entirely obvious in the car as it "ticks" at normal speed. I then figured out the whole rear light cluster wasn't working.

Raised as an issue with Tesla - the best they could do was a service appointment some 3 weeks later. So deliver a car in an illegal state and not give a flying fk when pointed out. In frustration I stripped off the interior panels in the boot to get to the light and discover the cluster isn't even plugged in. Plugged it in, all working. How the fk did it get to delivery without this being caught?

[anonymous-user]

Do it yourself PDI. It's the future.

[off_again]

Order66 said:

Yeah, I picked up my Telsa a month ago - by the time I had got home from picking it up I had figured that the right indicator was running at double-speed, something that isn't entirely obvious in the car as it "ticks" at normal speed. I then figured out the whole rear light cluster wasn't working.

Raised as an issue with Tesla - the best they could do was a service appointment some 3 weeks later. So deliver a car in an illegal state and not give a flying fk when pointed out. In frustration I stripped off the interior panels in the boot to get to the light and discover the cluster isn't even plugged in. Plugged it in, all working. How the fk did it get to delivery without this being caught?

And thats the problem.

Any production line, including Tesla, will have multiple checkpoints for quality control. Some of it automatic, some of it manual. Either way, there will be a number of these that will occur during the build process. Then there will be one as it leaves the factory and another when it arrives - just to make sure there is no damage in transit - usually required for insurance purposes. Then the local 'store' will carry out the legal requirements pre-delivery. The fact that no one either cares or looks is the problem.

I believe that the problem is that most of the Tesla employees will be on share schemes, which is heavily tied to sales and delivery numbers as well as some extra terms in there too. So hit the delivery numbers, get your shares and given its riding high at the moment, there is a lot of money to be made. The incentive is not to deliver a quality product, but to get them delivered to customers regardless. They can fix it as part of the after sales program, right? Of course, this is the most expensive way to do it, leads to annoyed customers and is frankly stupid, but hey, its all about the cash right?

And that leads me to the final point - I wonder how long these cars will actually last? Given the build quality issues, will they stand the test of time for paint, interior quality and overall build? Will they stand the abuse of the second / third owner? Yes, they are more reliable than the IC versions due to simplicity, but what about all of the other stuff? I wonder how they will stand 100,000 miles of ownership in somewhere other than California (which doesnt use salt on the roads in winter and has famously mild weather and 240 days of sun).

Cant help thinking that your EV South Korean model is going to look better, last longer and ultimately be the better longer term purchase for many. Not all, but for owners who are looking for a longer-term ownership, not sure it makes that much sense at the moment.