2017 Investor Roundtable:General Discussion

[mmd]

Not a chance. Did you even watch the video?

Posting random stuff IS an attempt at "special effects"

Yes, I watched the video. It is too good compared to what we read in the AP 2.0 threads on TMC.
You are underestimating the power of modern computers and video processing softwares

Here is another possibility/question. How do we know this video wasn't with AP 1.0 car? As of now, AP 1.0 appears to be more capable than AP 2.0. The title of the video says "Autopilot Full Self Driving Demonstration Nov 18 2016 Realtime Speed". Technically, it could be AP 1.0 or AP 2.0 or any other version of AP.

I also noticed, the car avoided going under any overpass. People are reporting AP slamming brakes in those situations.

 

[Bangor Bob]

EPA may go away completely by 2018

Florida Congressman Drafts Bill To ‘Completely Abolish’ The EPA | The Huffington Post

<shrug> They can do that, but the Clean Water Act and Clean Air Act will still need to be enforced.

If congress doesn't want so many regulations, they should quit passing legislation that says "The <agency> will create and enforce regulations to control <x>" and instead have the balls to set some standards themselves. Of course, that would leave the congresscritters liable for consequences come reelection time. Can't have that.

 

[AudubonB]

Lord Vetinari:

If you're missing your oh-so-clever posts, you can find them in Snippylandia.

1: Stop with the attacking of other members.
2: Stop with the pretending that you're a moderator.

***Any more from either of you and you're banned***

 

[JRP3]

Any other reports of this 10 day retooling?

On Dec 28, 2016 my order was confirmed with a delivery date of Late Jan - Early Feb. I didn't believe this would happen. Yesterday my Delivery Specialist called me (after a few emails) and said I'll be getting the car in early March. She said the Fremont factory is undergoing a 10 day retooling which impacts production.

Owners delivery stories, photos and first impressions

 

[vgrinshpun]

Yes, I watched the video. It is too good compared to what we read in the AP 2.0 threads on TMC.
You are underestimating the power of modern computers and video processing softwares

Here is another possibility/question. How do we know this video wasn't with AP 1.0 car? As of now, AP 1.0 appears to be more capable than AP 2.0. The title of the video says "Autopilot Full Self Driving Demonstration Nov 18 2016 Realtime Speed". Technically, it could be AP 1.0 or AP 2.0 or any other version of AP.

I also noticed, the car avoided going under any overpass. People are reporting AP slamming brakes in those situations.

If you did watch the video, you were not paying attention, probably because preoccupation with your outlandish ideas. The problem is not in me underestimating power of modern computers, rather you discounting any common sense. The video has numerous FSD software errors/bugs. If video's is not live recording of a FSD drive, but faked as you claim, these would not be there.

Here is the list, followed by the snapshots:

• Time stamp 0:20. Overhead structure - no slumming the brakes.
• Time stamp 3:53. Car stopped in the middle of the road, confused that one of the ladies is "in-path"
• Time stamp 4:25. Car stopped in the middle of executing right turn, presumably deciding that another car parked on the opposite side of intersection is "in-path"
• Time stamp 6:09. Car identifies an owner of the dog as an object, but not the dog
• Time stamp 6:53 Car stopped in the middle of the turn without discernible reason

I think you are shamelessly full of it and should stop spamming this thread with baseless theories.

 

[CaliBear1]

Is Acura's NSX hybrid supercar struggling amid slow early sales? - Roadshow

Interesting article on sluggish sales of Acura's new hybrid NSX. The author explores a number of possible explanations for the glacial pace of sales. Yet, he leaves out the possibility that many people willing to spend $150k or more on a "super" car may be gravitating to the Tesla P100D. I have no evidence to back that claim, but it would not surprise me in the slightest.

 

[RobStark]

Is Acura's NSX hybrid supercar struggling amid slow early sales? - Roadshow

Interesting article on sluggish sales of Acura's new hybrid NSX. The author explores a number of possible explanations for the glacial pace of sales. Yet, he leaves out the possibility that many people willing to spend $150k or more on a "super" car may be gravitating to the Tesla P100D. I have no evidence to back that claim, but it would not surprise me in the slightest.

BMW i8 hybrid is killing it. Still. This is a direct competitor to the NSX unlike the P100D 4 Door Super Sedan.

More likely it is a branding issue. The RLX, their next most expensive car, starts at $54k, then NSX starts at $156k.

Acura is a step up brand like Buick not a full luxury brand like Lexus, Mercedes or Porsche.

 

[AcesDealt]

Pic of my prior story

 

[schonelucht]

In fact, to note the obvious, being able to train the brand new software to do a drive without disengagement over just 550 miles is a very impressive result.

After 550 miles with 168 disengagments their AI is good enough to do a 20 mile uninterrupted run but 4 months later they still can't lane keep without bouncing, read traffic signs or avoid phantom braking in TACC. Everyone can choose to believe what they want, but my money is on their engineers special casing a lot of the travel in that video rather than their model being good enough after that many miles to pick up where there is a stop sign, a traffic light, what speed restriction there is, etc...

 

[AcesDealt]

I welcome the skepticism. Tesla is the duck, calm above and underneath paddle like hell. Fsd is not a solved problem, but it is within ten years of being a solved problem. In the meantime everyone bluffs superiority.

 

[vgrinshpun]

After 550 miles with 168 disengagments their AI is good enough to do a 20 mile uninterrupted run but 4 months later they still can't lane keep without bouncing, read traffic signs or avoid phantom braking in TACC. Everyone can choose to believe what they want, but my money is on their engineers special casing a lot of the travel in that video rather than their model being good enough after that many miles to pick up where there is a stop sign, a traffic light, what speed restriction there is, etc...

Where did you get 20 mile uninterrupted run? The slowed down video is less than 10 min long, with speed varying 20 - 35 mph, so it is probably 5 miles long or so.

You are making progress - from claiming that video was doctored to saying the particular route was "special cased". Well, once again, "special casing" as in training neural network software accumulating software driven miles over particular route.

Nothing to be suspicious about - this IS the model of how the software is going to advance and as such IS representative of what to expect from it in not too distant future. So it is totally legitimate for Tesla to demonstrate software's capabilities using this video.

You are, of course is right about believing what one chooses. I see the demo as indicative of what the software they are working on is capable of when it is properly trained and tested.

 

[schonelucht]

Where did you get 20 mile uninterrupted run? The slowed down video is less than 10 min long, with speed varying 20 - 35 mph, so it is probably 5 miles long or so.

20 miles comes straight from the official report.

You are making progress - from claiming that video was doctored to saying the particular route was "special cased". Well, once again, "special casing" as in training neural network software accumulating software driven miles over particular route.

To be clear: I didn't write that the video was doctored. I believe the ride happened exactly as recorded. My claim is that the demo was doctored by hand programming constraints (like possibly where it can stop, where it must stop, what speed it can go, where it can cross lanes, what road to follow exactly etc...) for the particular route chosen instead of acquiring the information through real time sensor data. Given how bad their current solution is, even well past when they expected they'd be at parity with what they had, I still think that is the most plausible explanation.

 

[Ovulation]

20 miles comes straight from the official report.



To be clear: I didn't write that the video was doctored. I believe the ride happened exactly as recorded. My claim is that the demo was doctored by hand programming constraints (like possibly where it can stop, where it must stop, what speed it can go, where it can cross lanes, what road to follow exactly etc...) for the particular route chosen instead of acquiring the information through real time sensor data. Given how bad their current solution is, even well past when they expected they'd be at parity with what they had, I still think that is the most plausible explanation.

Hmmm Tesla had IMO no reason to do such desperate things, what would be different when the didn't came up with their bold self driving car claim?

 

[Vitold]

After 550 miles with 168 disengagments their AI is good enough to do a 20 mile uninterrupted run but 4 months later they still can't lane keep without bouncing, read traffic signs or avoid phantom braking in TACC. Everyone can choose to believe what they want, but my money is on their engineers special casing a lot of the travel in that video rather than their model being good enough after that many miles to pick up where there is a stop sign, a traffic light, what speed restriction there is, etc...

Not necessarily so. Roads may look all the same to you but to the AP2 they look different and 'unfamiliar' road features throw it off until it drives long enough to learn. Calibration and alignment issues add to systems 'confusion'.

 

[vgrinshpun]

20 miles comes straight from the official report.

But 20 miles was not an uniterrupted run as you claimed. Report must have mentioned this in different context. Link?

To be clear: I didn't write that the video was doctored. I believe the ride happened exactly as recorded. My claim is that the demo was doctored by hand programming constraints (like possibly where it can stop, where it must stop, what speed it can go, where it can cross lanes, what road to follow exactly etc...) for the particular route chosen instead of acquiring the information through real time sensor data. Given how bad their current solution is, even well past when they expected they'd be at parity with what they had, I still think that is the most plausible explanation.

Well, not to bicker about who said what, but you did say "doctored demo"...

I do not see your claim as plausible. If video would be a result of programming constraints by hand, rather than demonstrating what neural net software can do, namely learn, there would not be so many bugs/errors in it, as I outlined in this post.

I do not believe calling "their current solution" "bad" is justified. We see exactly the same pattern that was observed with AP1 - it is progressively getting better, even without OTAs, which is again very reassuring, as it confirms what company is telling the world - they have narrow AI at work, in every car and at the mother ship. As investors, we should be all very encouraged by this, as this is huge differentiator with the competition.

Having said all of the above, seeing the sausage made would not please everyone...

 

[-=buzz=-]

Where did you get 20 mile uninterrupted run? The slowed down video is less than 10 min long, with speed varying 20 - 35 mph, so it is probably 5 miles long or so.

According to this, the route was about 7miles long.

 

[vgrinshpun]

According to this, the route was about 7miles long.

Thanks! I think that the demo did not use the 280 (limited access), but used the Junipero Sierra Blvd instead, and this route, as I've estimated is exactly 5 miles.

My point in noting this again is that the fact that all of the people trashing the demo are so off about details of it, underlines, IMO, the main problem with their strange claims: they seem to have formulated their opinion independently from the actual information available to an impartial observer.

 

[T3slaTulips]

Any other reports of this 10 day retooling?



Owners delivery stories, photos and first impressions

Caveat: I'd wait to hear from more than one source before trusting this story. Poster created their account Feb 1. Story said they got an incorrectly made 70D in December (but they didn't create their forum account then). I'm not saying it's not true, just saying I'd get a second/third source (which should be available for a 10 day shutdown) before I'd trust it.

 

[Bgarret]

To be blunt. The fact that we are comparing this:

To THIS:

Is.....wait for it.....Ludicrous. One is a rolling test mule funded by one of the largest, deepest pocketed companies in the world that has been refining by iterations on the same test platforms for years.

The other is a - quite sexy, and fast - a Production car that people can buy and drive (or soon choose not to drive) made by a startup with limited resources and has been out in the wild for a nanosecond.

There is no comparison - may as well compare an orange peels to ear wax.

Both companies are trying to accomplish similar goals by widely disparate methods. Google, surprisingly, is taking the old school, OEM approach of working on components to reduce marginal costs until they have a sensor suite and AI to sell a product at a profit and maybe bundle Android into the car somehow.

Tesla is getting product out into the hands of people to break, then designing, fixing, breaking, re-designing - A good example of their methodology is the recent crash test information. Before results are even announced, Tesla has a re-designed seatbelt in testing and is working on the headlights. Meanwhile Tesla is accumulating BILLIONS of miles of data....data needed for proof of concept to regulators and to train AI. The race for autonomy was not decided over 535 miles driven in 2016 - it will be decided over the Billions of miles driven in the next 2-5 years. Place your bets on the winner accordingly.

Always fun to see Vgrin, SBenson and Shonelucht mix it up, but let's move on...

Ear wax vs. Orange peels......Go!

 

[Bgarret]

I'm curious how Tesla will handle fleet sales of the Model III. It's not something they've had to deal with before. I can imagine demand for a fleet Model III to be robust, especially in urban areas. I don't believe Tesla will discount, as their demand far exceeds their supply. It would be interesting if a "fleet order" option were included in the next reservation round.

Scotland Yard is considering Tesla and other electric automakers to convert London’s police car fleet