Tesla and Uber Unlikely to Survive...
Discussion
DonkeyApple said:
as the streets are flooded with minicabs that can completely undercut all competition including private ownership costs.
And that is ultimately what's been 'sold' here, a total and complete transformation in personal transportation, from ownership to cost to manufacturing, and pretty much all business models in existence today.What Musk is proposing seems like utter madness and pure SciFi.
I guess we'll see how it plays out, either his right and Tesla will keep on going, or his wrong and Tesla will be done.
Forget short sellers Musks personal ambition is by far the biggest threat to Tesla.
Tesla could have done well been purely a niche premium brand, using resources to put in soft touch platics in the S/X, instead gambled it all on producing the 3. Now the gamble is even bigger, and the risks far higher, but I suppose thats how this guy functions.
gangzoom said:
And that is ultimately what's been 'sold' here, a total and complete transformation in personal transportation, from ownership to cost to manufacturing, and pretty much all business models in existence today.
What Musk is proposing seems like utter madness and pure SciFi.
I guess we'll see how it plays out, either his right and Tesla will keep on going, or his wrong and Tesla will be done.
Forget short sellers Musks personal ambition is by far the biggest threat to Tesla.
Tesla could have done well been purely a niche premium brand, using resources to put in soft touch platics in the S/X, instead gambled it all on producing the 3. Now the gamble is even bigger, and the risks far higher, but I suppose thats how this guy functions.
I'd happily go for that - an on demand car to take me wherever I wanted, when I wanted, with close to zero lead time, for a fee based on usage / access to the service.What Musk is proposing seems like utter madness and pure SciFi.
I guess we'll see how it plays out, either his right and Tesla will keep on going, or his wrong and Tesla will be done.
Forget short sellers Musks personal ambition is by far the biggest threat to Tesla.
Tesla could have done well been purely a niche premium brand, using resources to put in soft touch platics in the S/X, instead gambled it all on producing the 3. Now the gamble is even bigger, and the risks far higher, but I suppose thats how this guy functions.
I admire Musk for quite boldly heading there. He can't, of course, do it in the timeframe he mentions to a scale that will enable me to sign up to the service anytime soon. But that doesn't mean he shouldn't aim there.
That said, he shouldn't forget the day job: CEO of Tesla
Let's see what tomorrow's quarterly update brings.
RobDickinson said:
Tuna said:
And, as has been made clear, getting people to drive your cars for you and then send (by definition) limited data back to you is not 'training' in any meaningful sense.
Its exactly how you train neural nets.Here's how you train neural nets. You shove bucket loads of data into them, along with the 'correct' outcome you're looking for - and you compare that correct outcome with what actually happened. You do this for normal situations, and for the edge cases (of which, in a real world driving situation, there are billions). OK, so nothing controversial there.
But here's the thing - we know Tesla are not collecting all of the data. With 21 cameras plus radar etc, each car should be collecting gigabytes (if not terrabytes) of data on each and every journey Does your Tesla upload gigabytes of data every day? No, it really doesn't.
So Tesla are sampling. "Tell me when something interesting happened". But how do they know when something interesting happened? They're dependent on the same system that they're meant to be training to identify the right data to train on. Can you see the issue there?
Sure, they can look for extreme events - "My car had to brake hard unexpectedly... why was that?", but what about the journey where, due to a slightly shiny road surface, the car spent half the time driving way too close to the next lane? It's completely unaware that it's going wrong, and will not send those gigabytes of information back because everything was just fine - wasn't it? When you just sample real world data you also have to have a strong signal for what is the correct response from the neural network, and the cars themselves cannot generate that automagically.
Human drivers often drive distracted - looking at the stereo, shouting at the kids in the back, trying to fish the phone out from the side of the seat... and the vast majority of the time they get away with it because roads are relatively open and traffic is relatively predictable. Autopilot can do the same, for exactly the same reason - it's only once in a few thousand times that the distraction translates to a serious course correction, or an accident. But that means that there is no training data being generated to stop the autopilot equivalent of being 'slightly wrong'. We get away with 'slightly wrong' most of the time - right up to the point where a crash barrier, or a person is in the way and something really bad happens.
Tesla's training can certainly make AP safer - it can detect 'emergency' situations sooner and react better - because it can train from actual emergencies. But there is a huge question over whether it is getting the right data for 'normal safe driving' - because all that's being collected is random snapshots of journeys that didn't end badly. Not ending badly is not the same as driving safely.
The two points to make here are:
(1) How is Tesla going to be able to 'prove' its cars can drive themselves safely, if the only information they have is a small sample of driving situations? At the moment Tesla drivers are self selecting, and drive in a tiny subset of the nation's roads - that's not representative of the places the average taxi driver has to go. For Tesla's billions of miles of driving data (which is seriously filtered), the other companies are simulating trillions of miles. In the billions of miles that Teslas have driven, there should only be a few thousand incidents to train on - and only a few hundred accident situations. That's not training data, it's random samples.
(2) Given the burden of proof and regulatory compliance, how is Musk promising fleets of driverless taxis next year? You'd need cars in public testing four years ago to be able to hit that sort of promise. It's a promise so out there that it has to be questioned. As it seems to be a pitch to explain how they can get private individuals to finance a nationwide fleet of cars that Tesla themselves couldn't possibly afford, it looks perilously close to a pyramid scheme. That sort of naked financial manipulation requires a better proof of capability than "Of course we can do it.. because.".
AstonZagato said:
My wife doesn't even want to take the Tesla to the airport to be valet parked. The thought that she'd let it go off and do mini-cabbing...
Mini cabbing no way, but for family use it would fantastic. My parents don't like Mway driving, so even though Leicester to London is only 100 miles they take the train, which as anyone will tell you isn't cheap, nor convenient.If they get a Model 3 that can do 'full self driving' problem solved. Even more exciting would a Model 3 to 'share' between the extended family, need to send the car to school for pick up whilst we are both at work no issues - even if both grandmothers don't like to drive they can still sit in the car (my mum cannot).
Even for online shopping or returning stuff in the post. Yes its not hard to drive to the collection depot/post office etc, but wouldn't be much easier to send the car to a depot with the stuff needing be collected/sent whilst we are at work??
IF Tesla can pull this off its a game changer, and offer plenty new bussiness models to appear. But most important IF it gives us more family time due to no longer the need to drive to A and B for daily chores it would be fantastic.
Bear in mind the cost per mile of X including tyres/fuel is about 6p per mile, that is very cheap cost to pay the convince and freed time.
Note that none of that means we can't develop FSD cars using purely optical tracking at some point - just that the reality gap that Tesla is demonstrating makes it look less and less likely that it'll be them that delivers it.
Musk has a habit of making a virtue of his shortcomings. A normal dash in a traditional car is an extremely expensive and complex thing to develop - so the Model 3 makes do with a plank of wood and an iPad - "Look! Minimalism!". LIDAR would blow the budget for the Model 3 out of the water - so instead it gets cameras that cost a few dollars in components. HD mapping is incredibly expensive and resource intensive - so they are doing without. And so on..
FSD might miraculously fall out of these experiments and be genuinely cheap and reliable with no further investment, but that's phase two in the famous Gnome Underpants profit scheme - it's a happy accident rather than a planned outcome that any large business could depend on. Musk is completely reliant on keeping investment coming for long enough that a happy accident can happen. "Stay with me till my number comes up".
Musk has a habit of making a virtue of his shortcomings. A normal dash in a traditional car is an extremely expensive and complex thing to develop - so the Model 3 makes do with a plank of wood and an iPad - "Look! Minimalism!". LIDAR would blow the budget for the Model 3 out of the water - so instead it gets cameras that cost a few dollars in components. HD mapping is incredibly expensive and resource intensive - so they are doing without. And so on..
FSD might miraculously fall out of these experiments and be genuinely cheap and reliable with no further investment, but that's phase two in the famous Gnome Underpants profit scheme - it's a happy accident rather than a planned outcome that any large business could depend on. Musk is completely reliant on keeping investment coming for long enough that a happy accident can happen. "Stay with me till my number comes up".
Edited by Tuna on Tuesday 23 April 12:14
Tuna said:
FSD might miraculously fall out of these experiments and be genuinely cheap and reliable with no further investment, but that's phase two in the famous Gnome Underpants profit scheme - it's a happy accident rather than a planned outcome that any large business could depend on. Musk is completely reliant on keeping investment coming for long enough that a happy accident can happen. "Stay with me till my number comes up".
Edited by anonymous-user on Tuesday 23 April 12:14
Also, cost engineering is valid innovation. This kind of concession to practicality can be spun both ways.
Edited by anonymous-user on Tuesday 23 April 12:59
Humans make do with just vision (albeit, stereoscopic). We have no LIDAR or HD mapping. Theoretically, a solution should be achievable based on purely video input.
However, theory and practice are very different things. We have brains that have evolved over millennia (and are evolved from creatures with hundreds of millions of years) to interpret and react to that data. I struggle to believe that FSD is happening anytime soon - particularly when I use AP2 regularly. I realise that FSD requires a computational hardware upgrade but, even so, the AP2 is really not very impressive.I don't think I will be putting money down for the FSD option on my current car.
However, theory and practice are very different things. We have brains that have evolved over millennia (and are evolved from creatures with hundreds of millions of years) to interpret and react to that data. I struggle to believe that FSD is happening anytime soon - particularly when I use AP2 regularly. I realise that FSD requires a computational hardware upgrade but, even so, the AP2 is really not very impressive.I don't think I will be putting money down for the FSD option on my current car.
sambucket said:
Sounds to me like it's more than an experiment though. Seems like musk has been betting the farm on FSD since as early as 2016. Puts some of the internal angst into perspective.
Hard to tell without knowing more about the chips themselves. Having a mildly custom neural net accelerator might be justifiable just to avoid paying over the odds for whatever Nvidia comes up with - remembering that if they sell you chips that are good enough for FSD, they'll sell those same chips to the competition. Building your own silicon doesn't have to be that expensive, and protects your IP nicely.I neither know, nor (if im honest) care much either way about autonomous cars from Tesla or anyone else.
What I do wonder is why Musk doesn't put his efforts into building better cars for an affordable price. Surely that would generate sales quicker than making nebulous promises of something most don’t care about?
What I do wonder is why Musk doesn't put his efforts into building better cars for an affordable price. Surely that would generate sales quicker than making nebulous promises of something most don’t care about?
REALIST123 said:
I neither know, nor (if im honest) care much either way about autonomous cars from Tesla or anyone else.
What I do wonder is why Musk doesn't put his efforts into building better cars for an affordable price. Surely that would generate sales quicker than making nebulous promises of something most don’t care about?
From what I understand, the announcements weren't for the benefit of his customers - they were purely there to keep investment coming in. Musk needs to keep investors interested far more than he needs sales at the moment, particularly when he seems to be resource constrained on manufacture.What I do wonder is why Musk doesn't put his efforts into building better cars for an affordable price. Surely that would generate sales quicker than making nebulous promises of something most don’t care about?
AstonZagato said:
My wife doesn't even want to take the Tesla to the airport to be valet parked. The thought that she'd let it go off and do mini-cabbing...
It’s for the fiscally desperate not those who can actually afford the car. Want a yacht but can’t afford it, put it out to charter. Want a super car but can’t actually afford it, put it out wo wedding hire. Want a Merc but can’t actually afford it, put it out for minicabbing. If Musk can convince a whole load of people who can’t afford to lob $50k at a chattel by getting them to think they’ll be able to get 50% yields, or frankly any yield, then good luck to him. He is a genius salesman but this move really requires finding hundreds of thousands of punters who genuinely can’t think straight but can still get $40 odd k funding. A big ask even for America.
As someone else mooted, yesterday has probably more to do with trying to sow a positive seed ahead of the other news this week.
Gassing Station | EV and Alternative Fuels | Top of Page | What's New | My Stuff