I think anything over 50ms is useless. 200ms at 60mph is almost 6 meters. It's possible with proposed 5g specs but internet latency would have to catch up too. This is for purposes of decision making. I think a network can share data about traffic conditions and update parameters on the fly.

The "learning" process is something similar to what we do for learning poker. That's the ProPokerTools where we calculate odds and plot out decision trees. The trees are then memorized/practiced so we make instantaneous decisions with simplified heuristics.

"Deep" learning is giving computers to update the decision tree itself with incoming data. This approach worked out great for a closed game like Go because you can perfectly define winning conditions. For driving, "good labelled data" as someone pointed out is going to be crucial. Tesla, especially if it opts to track even driver behavior without auto-pilot, is going to have more data to feed whatever program they have than anyone else.

I don't know if this is what's going to happen. But as technology stands I think people are working on online "deep learning" to update heuristics and then local "instantaneous decisions."

Oh. Noted expert TS didn't mention Baidu. They are rapidly scaling. And they poached Ng from Google. But none of that bears mentioning in his fantasy world of 150mp cameras.

Also, I neglected to mention that even with offboard processing. Any sort of latency issues will cause crashes, and then what happens when someone crashes the network, which obviously will happen. Do we just have 2 million cars just slamming the brakes and stranding everyone? I mean the idea just makes no sense from a dozen or so perspectives.

Keep in mind I'm talking about collision avoidance. Obviously non-critical computation like traffic data for optimal pathing/etc could be offloaded, but I just thought it was comical he thought a human would perform better than an onboard computer, which would be inferior to an offloaded processor. This is just the exact opposite of reality and makes absolutely no sense.

It's obvious that offloaded processing is a no go, for several very strong reasons. It's not even in the realm of possibility imo. The processing unit will be entirely in the car. NVidia's next unit will have 10x the bandwidth/processing power of their current cutting edge one. That's getting up to where we need.

The Lidar that Google are using, the HDL-64E typically runs at 10Hz. I assume that they have some other sensors and algorithms than sensor fusion, but I would guess that 100ms should be doable. Humans typically don't react faster than 1s anyway. Even when I have max focus on simple task I react in ~250ms at this game:
http://www.humanbenchmark.com/tests/reactiontime

The benchmark SDCs need to meet is not humans. They need to much much much much safer.

China has already forged a path to legalization. Other countries will follow. This is where Baidu has a big advantage.

Humans don't have uniform performance. Imo if they can drive equal to humans when sober and alert that is good enough. No need to beat The Stig every time.


Also Tesla updated the Model S today:

Good article on the bear thesis which I agree with

http://www.theverge.com/2016/4/8/113...sales-compared

Come on. Wrecks have to be at damn near the "unavoidable by anybody" level to be commercially viable at driving speeds, otherwise they're getting their asses sued off constantly, not to mention epic levels of negative press and pressure. It doesn't matter if they're safer overall- if they're tarding off in avoidable ways with any regularity, it's no good.

ya, has to reduce accidents by atleast 90%

Again. China. Baidu got that hot hot on Beijing. You feel me?

It easy will reduce it by that amount. Tesla is probably already ahead of the normal accidents / distance at this point. Would be interested in the numbers.

http://electrek.co/2016/04/13/tesla-...ance-settings/
Pretty neat.

Wow, now that I understand Tesla is just a "gimmicky side project" to Elon, it makes so much more sense why he went into negative personal net worth guaranteeing loans to save the company in December '08.

Thank you for your valuable contribution to this always, always super fact-based thread!

http://www.theverge.com/2016/4/13/11...nevada-factory

Noticed this. $1 billion seems quite big, and the governor went out there. Another competitor maybe oil is:

dead 😿 bounce

Could you explain how exactly he went into negative personal net worth?

You seem to repeat things from Elon or reported as said by Elon as facts, when he has a known history of mostly saying stuff ranging from nonsense to totally bat**** insane.

Energy has gone out of TSLA now. Market is at highs. OPEC meeting over the weekend is a wildcard, but if you want to re-short, now's a decent time at $253.

The underlined text in my post is a link.

So suffice to say that you do not understand anything about loans in America other than what you heard Elon say?

ETA: I guess Steve said it and Elon agreed.

Yeah I added about a week ago.

Will do a final add (all things considered) @275 to boost the avg price if it TSLA makes it that far.

I know its easy for results to affect prior judgment, especially for someone in the early stages, but that initial short was still horrendous.

http://www.motortrend.com/news/apple-car-exclusive/

If there's a future, we're in it. Or at least soon to be in it (estimated 2019 release date).

The Motor Trend article is all about the apple car. Street view looks awesome.

You realize this "Apple car" they're suggesting is pure fiction, right? Apple may come out with a car, but this is just the mockup of some graphic designer who knows nothing.

But yes, the future will be all glass surrounds in ultra ultra high definition that are seamlessly powered by your phone. 2019 is too soon though - think 2025. The technology simply isn't ready.

This Microsoft 2019 video (put out in 2009, pre-iPad) is pretty close to what the future will look like in terms of displays (including in your car):

Wait....you think that was real?

I'm aware that it was a pre-rendered design from a design team, not an actual Apple design. It's mentioned in the article.

But it's also a committed project set with a 2019 release date. There's just no actual pictures yet.