Thoughts on TESLA

[NW-Bound]

A neuron is far more sophisticated than a transistor.

From an article describing an IBM project and a separate EU project to simulate the human brain:

The trouble is that at the moment, no computer is powerful enough to run a program simulating the brain. One reason is the brain’s interconnected nature. In computing terms, the brain’s nerve cells, called neurons, are the processors, while synapses, the junctions where neurons meet and transmit information to each other, are analogous to memory. Our brains contain roughly 100 billion neurons; a powerful commercial chip holds billions of transistors. Yet a typical transistor has just three legs, or connections, while a neuron can have up to 10,000 points of connection, and a brain has some 100 trillion synapses. “There’s no chip technology which can represent this enormous amount of wires,” says Diesmann.

Diesmann is one of the lead scientists of the Human Brain Project of the EU.

The article went on to say German scientists used 24 bytes of memory to simulate each synapse. So, that's 24 x 8 x 100 trillions = 19,200 trillion bits or transistors (1.92 x 10^16), just for the synapses.

The above is the number of transistors in 1.6 million NVIDIA processors.

And they did not even know if that was enough to model the neurons and synapses.

Neurons have many characteristics and properties. Any simulation can represent only a few of them, and hope this makes for a reasonably realistic model. “We don’t know which level [of description] is the correct one,” says Diesmann.

See: Why we’re a long way from computers that really work like the human brain.

The Human Brain Project has this Web site: https://www.humanbrainproject.eu/en/.

 

[meierlde]

If you read the Singularity is Near by Kurzweil, he estimates (assuming Moore's law continues to hold) that it will be the 2030s before a single computer will have the power of the human brain. The assumption about Moore's law is a big one however. If you look at object recognition for humans it takes a couple of years for a baby to be able to reliably recognize a lot of things, and get the things name. Then consider that to cross streets it takes child until 8-10 to be able to handle busy streets. It seems that driving is legal in the case of farm kids at 14 (in rural areas). So that suggests 14 years although farm kids can drive a tractor when they can see over the wheel and reach the pedals. (on private land)
Just looking at the amount of time it takes a human to set up the visual recognition process suggests how complicated it really is.

 

[NW-Bound]

They first have to teach the computer to break these common CAPTCHA's that are widely used today on the Web.

 

[Totoro]

Just looking at the amount of time it takes a human to set up the visual recognition process suggests how complicated it really is.

Basic object recognition is quickly becoming a solved problem. You can train a simple network in a few hours that distinguishes hundreds of animals from each other, as an example. It takes just one programmer and a decent GPU.

It's recognizing and dealing with context that makes things difficult.

 

[Totoro]

A neuron is far more sophisticated than a transistor.

As I mentioned, connectivity is the caveat

The Human Brain Project is not the same as building capabilities on par with a human. That is about emulating a brain. not simulating its functions. If you want to copy the brain exactly, I fully agree, we're a long way off.

Most signs however point to it not being necessary for building something that outperforms us in every relevant metric. We didn't duplicate the bird to learn how to fly, we built an airplane inspired by the bird.

To balance it out further:

• A transistor is millions of times faster in switching speeds.
• There's a reason to believe the actual target is the 20 Billion neurons in the neo-cortex as a complexity indicator, not the total of 100 billion in the entire brain (or 85B according to some estimates).
• To represent the state of a neuron, 16 bits or less might be enough. In some lossy approximations (neural networks) 8 bits sometimes get used.

Even if you would follow the logic of exactly copying the architecture of the brain with your numbers: NVIDIA shipped over 50 million GPU units last year. That's less than two weeks of production time to get to 1.2 million GPUs.

I really believe it's about figuring the architecture at this point, not the raw capabilities.

 

[NW-Bound]

As I mentioned, connectivity is the caveat

The Human Brain Project is not the same as building capabilities on par with a human. That is about emulating a brain. not simulating its functions. If you want to copy the brain exactly, I fully agree, we're a long way off.

Most signs however point to it not being necessary for building something that outperforms us in every relevant metric. We didn't duplicate the bird to learn how to fly, we built an airplane inspired by the bird...

Fair enough. A computer to drive a car does not need to have other human's abilities, such as being able to appreciate music or poetry, nor debate on a forum about AI.

So, it does not need to have full-blown human's capability.

Even if you would follow the logic of exactly copying the architecture of the brain with your numbers: NVIDIA shipped over 50 million GPU units last year. That's less than two weeks of production time to get to 1.2 million GPUs.

Sure. But how large a building to house, and how much power to feed 1.2 million GPUs? And has anybody figured out how to connect and program that many processors in parallel? Lots of work to equal one human.

I really believe it's about figuring the architecture at this point, not the raw capabilities.

I agree with this. AI has been shown to work in specific applications. Robot car is harder than anything that has been done, and they will get there.

But for now, they will need the lidar as a crutch, because computer vision has not advanced as far and as fast as Musk expected.

 

[jazz4cash]

Just listened to the TSLA conference call. Bizarre. Musk was rambling, incoherent on the analyst questions. Very uninspiring. Then he took a YouTube call asking about aspirational plans and he went into Tedtalk mode. I wonder if he'll move to another role.

 

[dirt_dobber]

I listened as well - ... boring, bonehead CEOs are not cool, incinerating investors cash is not cool, being unprepared for a conference call and wasting people’s valuable time is not cool and acting like a spoiled child with no sense of responsibility is not cool...
Hubris can be deadly

 

[bobandsherry]

Musk acted just as he is, not really surprising. Gotta guess that he just said what other CEO's would like to say on these calls. Current environment seems to reward those who act in a less than professional manner.

 

[ChiliPepr]

Musk acted just as he is, not really surprising. Gotta guess that he just said what other CEO's would like to say on these calls. Current environment seems to reward those who act in a less than professional manner.

The start or Enron's downfall was when Skilling, the Enron CEO, used vulgarity in an attack on a fund manager.

 

[Breedlove]

Tesla = Enron

 

[Texas Proud]

Tesla = Enron

Not quite as bad, but I think it will have the same outcome...

 

[Breedlove]

Didn't he get tons of money from the Federal Government a few years ago ? I think the Tesla cars , the ones that I have seen around here are really nice . Time will tell...

 

[bobandsherry]

Musk has only one thing in mind, how to increase the market value of Tesla. Anything else doesn't matter. If he can get enough people to continue to drive up the price of the stock then he wins. Empty promises help to drive people to do silly things when it comes to investing. And the Board did no one a favor when they approved such a odd compensation target back in March.

$2.6 Billion Pay Approved

 

[bobandsherry]

Didn't he get tons of money from the Federal Government a few years ago ? I think the Tesla cars , the ones that I have seen around here are really nice . Time will tell...

According to this article, most of the subsidies Tesla has received came from California and Nevada.

Tesla has gobbled up $5.3 billion in taxpayer money: a $3 billion factory subsidy from California, another $1.3 billion tax incentive from Nevada, a $45 million discounted Department of Energy loan, $517 million in the sale of California and other regulatory credits, $284 million in federal income tax credits given to Model S buyers and $257 million in other incentives.

Enough of the Elon Musk and Tesla Inc Snake Oil

That said, many companies, including other automakers, receive mountains of subsidies.

 

[jazz4cash]

Musk acted just as he is, not really surprising. Gotta guess that he just said what other CEO's would like to say on these calls. Current environment seems to reward those who act in a less than professional manner.

Did you listen to the call?

 

[bobandsherry]

Did you listen to the call?

Had no need to listen in as I have no horse in the race. I did listen to relevant portions of the call from various sources (just due to my curiosity). Why?

 

[jazz4cash]

Had no need to listen in as I have no horse in the race. I did listen to relevant portions of the call from various sources (just due to my curiosity). Why?

I was curious to know if your comment was based on actually listening vs responding to the reports of the call. His behavior on the call was widely reported but his demeanor was not. He sounded like he was medicated or perhaps exhausted from sleeping on the factory floor. Not good. I have no horse in the race either but I find it interesting/entertaining and maybe a bit scary for the overall market.

 

[bobandsherry]

I was curious to know if your comment was based on actually listening vs responding to the reports of the call. His behavior on the call was widely reported but his demeanor was not. He sounded like he was medicated or perhaps exhausted from sleeping on the factory floor. Not good. I have no horse in the race either but I find it interesting/entertaining and maybe a bit scary for the overall market.

I listened to various parts of the conference call, my take on his demeanor is he's tired of answering questions that focus on short term while he's focused on the long term. He forgets he's taken money from people who are investing in his company and has accountability to them. He seems he still wants to run his company as though it's private and then has a lot more latitude with setting the milestones further out. He's a lot like other wealthy people, thinking they are superior to everyone else and discussions with those people is just a waste of time. He wants to be told how great he is and anything that is contrary to that is an irritant to him. That's what I took from his demeanor.

 

[ERD50]

I have more reading to catch up on, but I saw this, and it really has me scratching my head. This is what Musk offers up as one of their 'obstacles'?

https://arstechnica.com/cars/2018/0...model-3s-weekly-pins-progress-to-consistency/

"We did go too far on the automation front and automated some very silly things," he said. One example the CEO offered: Originally the Model 3 included "fiber glass mats" of fluff on the top of the battery pack and the company had a "FluffBot" that would pick up fluff and place it on the battery pack. "Machines are not good at picking up fluff, human hands are very well suited for that," Musk said, "FluffBot would frequently just fail to pick up the fluff." Tesla ended up testing whether the fluff made any difference in cabin noise, found that it did not, and dispensed with the FluffBot.

OK, I realize I'm armchair quarterbacking here, but I do have some experience in these things (second hand with this type of automation), and this seems trivially simple to solve. You just have the supplier attach 3" circles of thin, adhesive backed plastic on to the fiberglass mat in each corner, maybe a few in the middle depending on the size of this mat. Then the robot picks it up with suction handlers. I'm sure I've seen this sort of thing done before, just to give the robot an area to 'grab'.

I'm super impressed with his SpaceX stuff, but I wonder if the Tesla, power-cells, and solar roof tiles is the house-of-cards ready to collapse?

And though I'm not a fan of EVs, I do think Musk has brought some creativity to the car market, so I kinda want the company to survive, if only to make other cars better.

-ERD50

 

[JoeWras]

He sounded like he was medicated or perhaps exhausted from sleeping on the factory floor.

CNBC morning show suggested he change the call to mornings to avoid happy hour. That's all they said. I got it...

I listened to various parts of the conference call, my take on his demeanor is he's tired of answering questions that focus on short term while he's focused on the long term. He forgets he's taken money from people who are investing in his company and has accountability to them. He seems he still wants to run his company as though it's private and then has a lot more latitude with setting the milestones further out. He's a lot like other wealthy people, thinking they are superior to everyone else and discussions with those people is just a waste of time. He wants to be told how great he is and anything that is contrary to that is an irritant to him. That's what I took from his demeanor.

Silicon Valley, tech in general. Working in it, I've seen the change over the last 2 decades. Hard questions and honest criticism were encouraged. It comes fromm the roots of the scientific method.

In recent years, the C-suite leaders seem to be more worried about the plebs kissing feet and coronating them. Money has always been good if your company hits. And in recent years, the money is beyond good, it is insane. This puts you on a different plane of existence. No time for these silly analyst boring questions, ya know.

 

[Music Lover]

According to Bloomberg, the Model 3 is now being built in a temporary tent in the parking lot:

“It’s preposterous,” Bernstein’s Warburton said. “I don’t think anyone’s seen anything like this outside of the military trying to service vehicles in a war zone. I pity any customer taking delivery of one of these cars. The quality will be shocking.”

https://www.bloomberg.com/news/articles/2018-06-25/the-future-of-tesla-hinges-on-this-gigantic-tent

 

[ERD50]

According to Bloomberg, the Model 3 is now being built in a temporary tent in the parking lot:

“It’s preposterous,” Bernstein’s Warburton said. “I don’t think anyone’s seen anything like this outside of the military trying to service vehicles in a war zone. I pity any customer taking delivery of one of these cars. The quality will be shocking.”

https://www.bloomberg.com/news/articles/2018-06-25/the-future-of-tesla-hinges-on-this-gigantic-tent

I'm no Tesla fan, but that article is largley hubris.

It’s a tent the size of two football fields that Musk calls “pretty sweet” and that manufacturing experts deride as, basically, nuts.

“Words fail me. It’s insanity,” said Sanford C. Bernstein & Co.’s Max Warburton, who benchmarked auto-assembly plants around the world before becoming a financial analyst.

As a retired manufacturing 'expert', getting assembly set up in a tent isn't necessarily a prelude to poor quality. It really depends on what is being assembled, and what steps were taken to qualify the line. It's not a silicon fab plant and clean room. If it is bringing finished sub-assemblies together, it could be a pretty basic, easily controlled process.


I'd be more worried about what appears to be the swing between 'robots for everything', to sudden shifts of throwing people at those tasks. That just doesn't sound well thought out, and sets the stage for all kinds of issues - but that doesn't mean they will happen.


Oh well, I wasn't going to buy one anyhow, and don't own the stock (other than any in my index funds), so it's just a passing interest for me. But if I were going to purchase one, hmmm, yes, I would be concerned about buying a product that was produced in the middle of a difficult ramp up. I've seen products go from 'production hell' and poor field quality, to smooth sailing and good field quality, in less than a year. Sometimes the 'kinks' are just 'kinks' that get worked out. Other times they are just so much a part of poor design, that the product will always be poor quality - but those die off in a sort of consumer form of Darwinism.


-ERD50

 

[Chuckanut]

Tent is a bad name for the structure as it brings up images of a cloth shelter held up by light weight poles that connect together with little clips, and it wobbles easily and can collapse in very bad weather. Actually it is something made by a company called Sprung, and it really quite an interesting design. Apparently, their structures have survived hurricanes.

http://www.sprung.com/

OTOH, as much as I would love to drive an electric vehicle, Mr. Musk is skating on very thin financial ice with Tesla these days. Granted there are hoards of short sellers who need Tesla to fail and they have a vested interest in bad-mouthing the company. But I do wish Musky was a bit more practical at times.

Call the structure what you will, it's the people and equipment inside it that will determine how many and how well the cars are made.

 

[Running_Man]

Any company that requires a $1,000 deposit to order something that will come, sometime and takes 450,000 orders in backlog and then requires another $2500 this time non-refundable if you want to get a place in line to get a car, when they have orders for 90 months of peak production they have been unable to hit, has a good salesman for a leader and a terrible production company.