Firmware 9 in August will start rolling out full self-driving features!!!

[1375mlm]

if he really believed his own hype he would have put some tangible guarantees behind it - notice how Elon is making sure in the last 2-3 years to make sure there is nothing definitive in his promises including any definite timelines.

I agree with your entire post, except for this.

The way I see it, Elon actually believes his estimates. I think the reason he now says “about” or “maybe” is because of previous experiences
He said it himself that he had never built a mass market car, and “How am I supposed to know with precision when it’s gonna get done?”
I think he’s gotten so much hate for his wildly inaccurate estimations that now he’s cautious when making estimations for the future.
I don’t think Elon has much to gain from lying to his customers.
It seems to me that he looks at progress, makes his own genuine estimation, and somewhere down the line encounters an unforeseen problem, which delays the project. I presume that it is because of his inexperience in the area. But the way I look at it is that despite being years late, the company is still years ahead of competition.

 

[aronth5]

I would strongly suspect stop signs would require double confirmation.... confirm to slow to a stop. Confirm to start back up.

There’s no way people would want their cars randomly slowing down for false positive stop signs no matter how slim the chances it would happen.

When approaching a stop sign I just want some notice on the display that the car has recognized the sign without the need to confirm the stop.
And I want the notice to appear consistently at the same distance within reason from the sign.

 

[whitex]

I agree with your entire post, except for this.

The way I see it, Elon actually believes his estimates. I think the reason he now says “about” or “maybe” is because of previous experiences
He said it himself that he had never built a mass market car, and “How am I supposed to know with precision when it’s gonna get done?”
I think he’s gotten so much hate for his wildly inaccurate estimations that now he’s cautious when making estimations for the future.
I don’t think Elon has much to gain from lying to his customers.
It seems to me that he looks at progress, makes his own genuine estimation, and somewhere down the line encounters an unforeseen problem, which delays the project. I presume that it is because of his inexperience in the area. But the way I look at it is that despite being years late, the company is still years ahead of competition.

Elon did not get hate for wildly inaccurate estimations, rather he got hate for for taking money for features he never delivered and never will, due to his wildly inaccurate estimations. That's a big difference. If he said he'll have FSD in 2018 and didn't get there till 2028, he would still get love. However when he takes people's money in 2016 for FSD and then delivers FSD for any cars newer than 2027, then he gets hate.

Elon's strategy seems to be - promise a diamond, deliver a piece of glass, then dismiss the argument that he way under-delivered as "boring" because "the current product will deliver 2 diamonds", rinse-lather-repeat. AP3 promises (TBD) will be greater than AP2 promises (AP2.0 has all the hardware for full self driving, where you can summon the car from across the country, you can have your Tesla drive you family and friends and even work for Tesla car sharing network), which were greater than AP1 promises (your car will find you anywhere on private property), even though AP1 didn't come close, AP2 never will come close, and in all likelyhood neither will AP3.

 

[BigD0g]

Elon did not get hate for wildly inaccurate estimations, rather he got hate for for taking money for features he never delivered and never will, due to his wildly inaccurate estimations. That's a big difference. If he said he'll have FSD in 2018 and didn't get there till 2028, he would still get love. However when he takes people's money in 2016 for FSD and then delivers FSD for any cars newer than 2027, then he gets hate.

Elon's strategy seems to be - promise a diamond, deliver a piece of glass, then dismiss the argument that he way under-delivered as "boring" because "the current product will deliver 2 diamonds", rinse-lather-repeat. AP3 promises (TBD) will be greater than AP2 promises (AP2.0 has all the hardware for full self driving, where you can summon the car from across the country, you can have your Tesla drive you family and friends and even work for Tesla car sharing network), which were greater than AP1 promises (your car will find you anywhere on private property), even though AP1 didn't come close, AP2 never will come close, and in all likelyhood neither will AP3.

Yep.

 

[1375mlm]

Elon did not get hate for wildly inaccurate estimations, rather he got hate for for taking money for features he never delivered and never will, due to his wildly inaccurate estimations. That's a big difference. If he said he'll have FSD in 2018 and didn't get there till 2028, he would still get love. However when he takes people's money in 2016 for FSD and then delivers FSD for any cars newer than 2027, then he gets hate.

Good point.
I still would like to believe that he accepted money because he thought it would be done soon.
The Tesla website even said EAP would be done before end of 2017 in the details of the AP option

I think he has the best intention,
Elon said in the 60 min interview
“People should not ascribe to malice that which can easily be explained by stupidity."

It’s probably why the stopped accepting payments for FSD- Tesla’s official way of saying “oh crap didn’t realize it would take this long”

 

[whitex]

The Tesla website even said EAP would be done before end of 2017 in the details of the AP option

2016, not 2017 - so 2 years ago, not 1. Most things in this EAP description are still not delivered despite the claimed 40 times more power than AP1.

Good point.
I think he has the best intention,

He's too smart of a guy to be so out of touch with reality. So maybe his best intentions tell him the best thing to do is to sell something even though he knows it cannot be delivered - something along the lines of "better to lie and survive than tell the truth and have the company die"?

 

[MarkS22]

2016, not 2017 - so 2 years ago, not 1. Most things in this EAP description are still not delivered despite the claimed 40 times more power than AP1.
View attachment 365323


He's too smart of a guy to be so out of touch with reality. So maybe his best intentions tell him the best thing to do is to sell something even though he knows it cannot be delivered - something along the lines of "better to lie and survive than tell the truth and have the company die"?

I agree with you, however, I believe the break-up with MobileEye was much more of a setback than most people realize. In that specific case, I think he did think it could be delivered roughly in that timeframe. I'm still of the mindset that the DrivePX2 (HW2.0) was supposed to include an EyeQ3 to process the visual data and help validate Tesla's code more rapidly. When they couldn't do this, I think it set them back a good two years. We're only just at feature parity (give or take) now. And we're just seeing those promised EAP features come online. Once we get more advanced Summon (as promised/teased recently) and lane changed without confirmation (also teased recently), it's just about there.

 

[electronblue]

I agree with you, however, I believe the break-up with MobileEye was much more of a setback than most people realize. In that specific case, I think he did think it could be delivered roughly in that timeframe. I'm still of the mindset that the DrivePX2 (HW2.0) was supposed to include an EyeQ3 to process the visual data and help validate Tesla's code more rapidly. When they couldn't do this, I think it set them back a good two years. We're only just at feature parity (give or take) now. And we're just seeing those promised EAP features come online. Once we get more advanced Summon (as promised/teased recently) and lane changed without confirmation (also teased recently), it's just about there.

Much of this is probably true except one crucial bit: The timeline of the announcements does not match with your thinking there. MobilEye break-up happened in July 2016. AP2 announcement — including promised timelines — happened later, in October 2016. Even if and likely when what you mention was the plan, it had already changed by the time the announcements were prepared and made.

Elon knew MobilEye would not be on board when he announced what he did. AP2 as Level 5 capable hardware. There were several reports and resignations on the disagreements from his team too, so it could not have come as a surprise really.

Elon knew.

It is not like this is the first time this happened. The same thing ”deliver previously sold promise only in a future product” happened with P85D, P85D Ludicrous, P90D V1 to V3 promised performance that was only eventually delivered in the P100D, with each incremental step before it falling short of what was sold.

AP1 is another example. People forget that AP2 features like traffic lights and even Navigation on Autopilot like automated navigation-based exit taking were actually announced for AP1 first but never came (in addition to the summon will meet you at the curb and find you on private property for AP1 already mentioned). Navigation-based Autopilot for AP1 was announced for firmware 8.1 in late 2016.

Fool us once, eh?

 

[emmz0r]

I agree with you, however, I believe the break-up with MobileEye was much more of a setback than most people realize. In that specific case, I think he did think it could be delivered roughly in that timeframe. I'm still of the mindset that the DrivePX2 (HW2.0) was supposed to include an EyeQ3 to process the visual data and help validate Tesla's code more rapidly. When they couldn't do this, I think it set them back a good two years. We're only just at feature parity (give or take) now. And we're just seeing those promised EAP features come online. Once we get more advanced Summon (as promised/teased recently) and lane changed without confirmation (also teased recently), it's just about there.

Sign reading... ?

 

[whitex]

I agree with you, however, I believe the break-up with MobileEye was much more of a setback than most people realize.

Tesla+Mobileye partnership ended before they Elon marketed AP2 (July 2016). Are you saying that Elon one of those "most people" who were clueless about how much of a setback it was? If yes, then he really doesn't have a good grasp on his own company technology and/or doesn't have proper advisers (or possibly he doesn't listen to anyone who tells him not what he wants to hear). If Elon did realize the impact of the breakup, then he deceived everyone by marketing AP2 the way he did.

PS> I think Elon was an evil genius in this case - planning and actually attempting the the first major AI IP heist by using someone else's AI to train your own. I wonder what Elon would do if someone did the same thing - bought a bunch of Tesla AP2 or AP3 hardware and used it to train a new competing AI. It also interesting to watch how long before the legal system catches up with this type of intellectual property theft. Given that Elon has thought of doing this, I bet he also thought of someone else doing it to him. I have a very strong suspicion that AP3 will be heavily focused on security to try to protect Tesla IP (for example, inputs come in, steering commands come out, but no way to get back labeled videos as in this article - being able to feed raw videos and get back labeled ones is a goldmine to AI training.

 

[electronblue]

I think Elon was an evil genius in this case - planning and actually attempting the the first major AI IP theft by using someone else's AI to train your own. I wonder what Elon would do if someone did the same thing - bought a bunch of Tesla AP2 or AP3 hardware and used it to train a new competing AI. It also interesting to watch how long before the legal system catches up with this type of intellectual property theft. Given that Elon has thought of doing this, I bet he also thought of someone else doing it to him. I have a very strong suspicion that AP3 will be heavily focused on security to try to protect Tesla IP (for example, inputs come in, steering commands come out, but no way to get back labeled videos as in this article - being able to feed raw videos and get back labeled ones is a goldmine to AI training.

I wonder if anything has been stopping Tesla from this anyway — I mean they could still have development EyeQ3 chips on their NN training cars/systems even if they lack them in the customer cars?

I think the customer reality is a bit simpler independent of what they may or may not have planned internally for EyeQ3. In customer cars I think Tesla wanted to use EyeQ3 to boost AP2 initially just like they are using radar and ultrasonics to boost the system initially even though the eventual plan is Tesla Vision only. (Boost in the sense of compensating for limitations of Tesla’s own vision system.) With EyeQ3 Tesla could have basically shipped AP1 on AP2 cars initially and then built from there. Not surprisingly we hear the first AP2 software indeed was mimicking EyeQ3 even on the data structure level to fit into the existing framework...

 

[trm2]

Elon's strategy seems to be - promise a diamond, deliver a Fart app.

Fixed it for you.

 

[whitex]

I wonder if anything has been stopping Tesla from this anyway — I mean they could still have development EyeQ3 chips on their NN training cars/systems even if they lack them in the customer cars?

There are likely NDAs and terms for engineering development for the few engineering samples Tesla did get their hands on. This is purely an educated guess on my part, but between the legal risk and the limited number of chips Tesla got their hands on, it probably doesn't make any sense to even try.

 

[electronblue]

There are likely NDAs and terms for engineering development for the few engineering samples Tesla did get their hands on. This is purely an educated guess on my part, but between the legal risk and the limited number of chips Tesla got their hands on, it probably doesn't make any sense to even try.

That is possible.

I doubt the chip amount necessarily is limited after all Tesla still maintains AP1 cars with the same chip but legal or contractual reasons certainly could play into it.

 

[diplomat33]

Elon's strategy seems to be - promise a diamond, deliver a piece of glass, then dismiss the argument that he way under-delivered as "boring" because "the current product will deliver 2 diamonds", rinse-lather-repeat. AP3 promises (TBD) will be greater than AP2 promises (AP2.0 has all the hardware for full self driving, where you can summon the car from across the country, you can have your Tesla drive you family and friends and even work for Tesla car sharing network), which were greater than AP1 promises (your car will find you anywhere on private property), even though AP1 didn't come close, AP2 never will come close, and in all likelyhood neither will AP3.

Elon's ways of doing things is to push the envelope. He's a visionary and ridiculously over the top ambitious. It is how he gets things done. We see this with all his endeavors, from Tesla to SpaceX to the Boring Co. With SpaceX, he's pushing for a "starship" that sits atop a Big F'ing Rocket that he claims will be capable of sending humans to Mars and even to the edges of the solar system. He's given talks about Mankind becoming an interplanetary species and colonizing Mars and beyond. Talk about bold!! With Boring Co, he wants nothing less than a revolutionary hyperloop transportation network that will allow people to travel underneath cities and between cities at ludicrous speeds. With Tesla, he wants a network of fully self-driving electric cars that can travel coast to coast with no driver, recharge at automated superchargers and can charge completely off the grid. It sounds crazy. He's not like the other automakers, playing it safe, building their average hybrids year after year. And I think most of us on this forum love that about Musk. At the end of the day, maybe we don't have FSD cars yet but we are driving really innovative and cool cars that we never would have gotten from those generic automakers.

And as the tech gets better, the probability that the hardware will be able to deliver FSD gets higher. Which is why I think over time, the skepticism of the naysayers who say Tesla's hardware won't be able to do FSD becomes less and less believable. AP1 was never going to do FSD. AP2 was iffy. The cameras and radar are good enough but clearly the processing power is not quite there yet. AP3 can process 2000 frames per second, 10x times better than AP2. AP3 has the best chance yet of achieving FSD.

 

[rnortman]

Elon's ways of doing things is to push the envelope. He's a visionary and ridiculously over the top ambitious. It is how he gets things done.

Nobody takes issue with being ambitious and visionary. I, and many others, take issue with accepting payment for something which does not exist and will never be delivered. (AP1 promises will clearly never be delivered as they have ceased new feature development for AP1; it looks likely that AP2 EAP promises of On-Ramp to Off-Ramp and Smart Summon will never be delivered on HW2 at least and they've never said EAP buyers will get HW3; you are welcome to remain optimistic about FSD on HW3 for FSD purchasers, but I remain skeptical.)

By the way, I didn't resurrect this thread to shame you. I resurrected it to shame Elon and Tesla. And indirectly shame anybody who continues to give any credibility to Elon's promises.

 

[diplomat33]

Nobody takes issue with being ambitious and visionary. I, and many others, take issue with accepting payment for something which does not exist and will never be delivered. (AP1 promises will clearly never be delivered as they have ceased new feature development for AP1; it looks likely that AP2 EAP promises of On-Ramp to Off-Ramp and Smart Summon will never be delivered on HW2 at least and they've never said EAP buyers will get HW3; you are welcome to remain optimistic about FSD on HW3 for FSD purchasers, but I remain skeptical.)

On accepting payment for non existent features, I would agree with you. I am on record as saying that it was a mistake for Tesla to accept payment for "FSD" back when HW2 was first announced since "FSD" was not ready for the public.

I am optimistic because I know "FSD", as I have defined it, will definitely happen. It is only a matter of time. And I love my Model 3. I think EAP works incredibly well. My previous car just had dumb cruise control so for me, AP2 is a radical improvement. I look forward to future improvements to EAP and eventually "FSD".

By the way, I didn't resurrect this thread to shame you. I resurrected it to shame Elon and Tesla. And indirectly shame anybody who continues to give any credibility to Elon's promises.

No offense taken. I actually thought it was Bladder who resurrected this thread when he asked me about my botched predictions. Although, I have to admit I am not sure how shaming people is constructive. Are you hoping to shame Elon into changing his ways? What do you hope to accomplish by shaming people who still give credibility to Elon's promises?

 

[MarkS22]

Sign reading... ?

Apparently, that's in current engineering code. However, I think it's another case where truth/reality can be blurred. For example, MobileEye has worldwide signage. I suspect they have a version of it working on California (USA?) roads only.

 

[MarkS22]

Tesla+Mobileye partnership ended before they Elon marketed AP2 (July 2016). Are you saying that Elon one of those "most people" who were clueless about how much of a setback it was? If yes, then he really doesn't have a good grasp on his own company technology and/or doesn't have proper advisers (or possibly he doesn't listen to anyone who tells him not what he wants to hear). If Elon did realize the impact of the breakup, then he deceived everyone by marketing AP2 the way he did.

PS> I think Elon was an evil genius in this case - planning and actually attempting the the first major AI IP heist by using someone else's AI to train your own. I wonder what Elon would do if someone did the same thing - bought a bunch of Tesla AP2 or AP3 hardware and used it to train a new competing AI. It also interesting to watch how long before the legal system catches up with this type of intellectual property theft. Given that Elon has thought of doing this, I bet he also thought of someone else doing it to him. I have a very strong suspicion that AP3 will be heavily focused on security to try to protect Tesla IP (for example, inputs come in, steering commands come out, but no way to get back labeled videos as in this article - being able to feed raw videos and get back labeled ones is a goldmine to AI training.

I don't think he was clueless. I think, for example, early progress on duplicating features went really fast. Low hanging fruit. I think there was some hubris mixed in with the marketing bravado. There is a reoccurring theme of hubris throughout all this. And I think you need it to believe you can accomplish much of it. Be it Tesla or SpaceX, you can't achieve all these things if you don't believe it's possible. The truth lies somewhere in the middle of "this is easier than people think" (Narrator: "It's not.") and "It'll eventually be available, but if not, we'll deliver other distracting features. Better to ask for forgiveness than permission."

I don't think it's "Let's completely lie" so much as, "The engineering team says there's a 20% chance it'll be ready in 2016" and he runs with it, thinking if he sleeps in the room and hires even better talent it'll actually be possible.

Agreed on the "evil genius" part. Maybe "evil" was strong, but Mobileye surely put themselves in a dangerous position. As AI gets more sophisticated, how do we stop a competitor from simply training itself? Sure, you can make it tougher by encrypting internal communications, but ultimately IP law is going to have to catch up.

 

[rnortman]

I am optimistic because I know "FSD", as I have defined it, will definitely happen. It is only a matter of time.

Yes, FSD will definitely happen. But I doubt it will happen (as you have defined it -- truly driverless and unsupervised) on currently-produced Tesla vehicles, at least not without substantial retrofits of new hardware (not just HW3).

Your certainty is rather odd, given that nobody has ever released a true L4/L5 driverless car. Given that it has never been done, any certainty about what it will require is unjustified -- nobody knows for sure if it can be done with the AP2 sensor suite. Nobody knows for sure it can be done in most environments even with Waymo's much more sophisticated sensor suite and vastly superior computing power (yes, superior even to HW3). The only reasonable thing to do is to acknowledge that uncertainty.