TSLA: When this baby hits 88 you're going to see some serious shit

The Something Awful Forums > Discussion > Serious Hardware/Software Crap > YOSPOS > TSLA: When this baby hits 88 you're going to see some serious shit

«‹›3418 »

fart simpson: Jul 2, 2005; DEATH TO AMERICA

infernal machines posted:

hey guys, guess what?

lol. feature regressions and no user communication regarding changes to life critical systems. they're not just a software company, they're a really bad software company

that�s our tesla :imunfunny:

# ? Mar 20, 2019 15:08

Adbot: ADBOT LOVES YOU

# ? Apr 24, 2024 08:26

infernal machines: Oct 11, 2012; we monitor many frequencies. we listen always. came a voice, out of the babel of tongues, speaking to us. it played us a mighty dub.

quote:

Since you want to be technical: It didn't pick the wrong lane, it chose to drive on the gore. It probably thought to be following the lane.

# ? Mar 20, 2019 15:12

infernal machines: Oct 11, 2012; we monitor many frequencies. we listen always. came a voice, out of the babel of tongues, speaking to us. it played us a mighty dub.

also,

quote:

Never be complacent with Autopilot. Longtime Tesla drivers are the most susceptible to getting into trouble. Remember that every corrective input you give is a data point to make the software better. As an owner, you are part of a half million person army (and growing) who are training a neural network AI every day. Data is king. It will keep getting better, but always remain vigilant. There has never been anything like this done in history, don't lose sight of the big picture and stay safe.

it doesn't, you aren't, it isn't, and it won't

# ? Mar 20, 2019 15:20

mystes: May 31, 2006

"Every person killed by autopilot is another data point we use to improve our software. Thanks for paying to be a guinea pig."

# ? Mar 20, 2019 15:23

infernal machines: Oct 11, 2012; we monitor many frequencies. we listen always. came a voice, out of the babel of tongues, speaking to us. it played us a mighty dub.

not only that, but they still believe fervently that their daily driving is somehow training this thing. there's no reason to believe that beyond some offhand bullshit musk said about a million years ago, and there's every reason to assume that's not the case since people have been tracking the vehicle's data usage for a half-decade via their home wifi networks. but whatever

# ? Mar 20, 2019 15:27

Zlodo: Nov 25, 2006

blood for the blood ai

# ? Mar 20, 2019 15:27

H.P. Hovercraft: Jan 12, 2004; one thing a computer can do that most humans can't is be sealed up in a cardboard box and sit in a warehouse; Slippery Tilde

infernal machines posted:

not only that, but they still believe fervently that their daily driving is somehow training this thing. there's no reason to believe that beyond some offhand bullshit musk said about a million years ago, and there's every reason to assume that's not the case since people have been tracking the vehicle's data usage for a half-decade via their home wifi networks. but whatever

the amount of data that would need to push is actually one of the problems with implementing autonomous vehicles

quote:

�Each car driving on the road will generate about as much data as about 3,000 people,� Krzanich says. And just a million autonomous cars will generate 3 billion people�s worth of data, he says.

Maps are just one example of the incoming data a car will need, and those maps won�t be a one-time Google map download, as one can perform now. They will have to be extremely detailed and timely�down to the nearest inch. They will be used for lane control and road hazards, among other things, so they will need to be continuously updated.

�You�re going to have to have data as much as any other kind of propulsion,� Krzanich says.

Krzanich splits the continually changing intelligence into three data sets:

The car will have to learn about such things as cones in the road and other hazards, which Krzanich calls technical data.

There will also be societal data, also called crowd-sourced data. It includes an automatic version of platforms such as Waze, for example. Waze is a community-based traffic awareness app that is heavily reliant on crowd-sourced traffic reports.

Personal data will make up the third classification. That includes locations and stop time..

Without data, your self-driving car "will have to deal with the world in a very manual way,� Krzanich says. �Data is the next oil.�

H.P. Hovercraft fucked around with this message at 16:44 on Mar 20, 2019

# ? Mar 20, 2019 16:42

Endless Mike: Aug 13, 2003

https://twitter.com/KlendathuCap/status/1108364223493668864

# ? Mar 20, 2019 16:55

infernal machines: Oct 11, 2012; we monitor many frequencies. we listen always. came a voice, out of the babel of tongues, speaking to us. it played us a mighty dub.

H.P. Hovercraft posted:

�Each car driving on the road will generate about as much data as about 3,000 people,� Krzanich says. And just a million autonomous cars will generate 3 billion people�s worth of data, he says.

Maps are just one example of the incoming data a car will need, and those maps won�t be a one-time Google map download, as one can perform now. They will have to be extremely detailed and timely�down to the nearest inch. They will be used for lane control and road hazards, among other things, so they will need to be continuously updated.

�You�re going to have to have data as much as any other kind of propulsion,� Krzanich says.

Krzanich splits the continually changing intelligence into three data sets:

The car will have to learn about such things as cones in the road and other hazards, which Krzanich calls technical data.

There will also be societal data, also called crowd-sourced data. It includes an automatic version of platforms such as Waze, for example. Waze is a community-based traffic awareness app that is heavily reliant on crowd-sourced traffic reports.

Personal data will make up the third classification. That includes locations and stop time..

Without data, your self-driving car "will have to deal with the world in a very manual way,� Krzanich says. �Data is the next oil.�

yes, unfortunately the bazingas' takeaway from this was "this is a thing that's real and happening right now"

# ? Mar 20, 2019 16:57

H.P. Hovercraft: Jan 12, 2004; one thing a computer can do that most humans can't is be sealed up in a cardboard box and sit in a warehouse; Slippery Tilde

lol he hasn't responded

https://twitter.com/danahull/status/1108362219538907136

# ? Mar 20, 2019 17:00

Shifty Pony: Dec 28, 2004; Up ta somethin'

also the crowdsourced data is meaningless without review and classification by a trained person. even if the car didn't crash and the driver didn't notice anything odd to report there's still no way to know if the AI operated correctly without manual review of the data it used.

there's a big difference between "successfully tracked the lane markers to follow the road through a curve" and "thought a tree shadows were the line but that happened to be going the right direction at this time of day so it still kept on the road."

a single driver can collect more data in a day than a trained worker can review and classify in a week.

Shifty Pony fucked around with this message at 17:06 on Mar 20, 2019

# ? Mar 20, 2019 17:02

Cybernetic Vermin: Apr 18, 2005

Shifty Pony posted:

also the crowdsourced data is meaningless without review and classification by a trained person. even if the car didn't crash and the driver didn't notice anything odd to report there's still no way to know if the AI operated correctly without manual review of the data it used.

there's a big difference between "successfully tracked the lane markers to follow the road through a curve" and "thought a tree shadows were the line but that happened to be going the right direction at this time of day so it still kept on the road."

a single driver can collect more data in a day than a trained worker can review and classify in a week.

yeah, my thinking is more and more getting to be that the workload involved in supervising the machine learning involved is actually a far more limiting factor than previously imagined. there was a sense in the field that all problem areas would be such that the fiddling would eventually be done and more data could be piled on, as it was for some earlier problems (e.g. a lot of nlp tasks are pretty well understood now), but it might not be playing out that way

it is especially notable that regressions get pretty common in complex systems, and testing them properly amounts to even more annotation by hand and supervision.

# ? Mar 20, 2019 17:08

Munkeymon: Aug 14, 2003; Motherfucker's got an
armor-piercing crowbar! Rigoddamndicu𝜆ous.

Lutha Mahtin posted:

e: also buses in metro MSP don't use the HOV lane. they get to drive on the shoulder of the road during rush hour past evvvvvverybody

sure they will if traffic is flowing. they also obey the speed limit which makes it hilarious when leadfoot shitbirds try to use the HOV lanes behind a bus and the driver just doesn't respond to their usual tailgating

# ? Mar 20, 2019 17:10

Endless Mike: Aug 13, 2003

remember when forza came out with "drivatars" that analyzed your bideo jame driving and would make an ai version to race online so you ended up with a bunch of cars crashing into you constantly?

what i am saying is this is where tesla got their data

# ? Mar 20, 2019 17:19

infernal machines: Oct 11, 2012; we monitor many frequencies. we listen always. came a voice, out of the babel of tongues, speaking to us. it played us a mighty dub.

Shifty Pony posted:

a single driver can collect more data in a day than a trained worker can review and classify in a week.

but also the cars ain't collecting poo poo and no one is reviewing anything

# ? Mar 20, 2019 17:24

evilweasel: Aug 24, 2002

Mr. Nice! posted:

the judge is going to fine him and tell him to comply with the order. this may include periodic reporting on the pre-approval status.

that would be the default but he's so vigorously flipped the bird to the court and to the SEC by completely ignoring the settlement and boasting about doing so I have significant doubts he's getting away that easy

he could not have ignored the settlement more than he did, and tesla knew he was ignoring it and let him do it

# ? Mar 20, 2019 17:56

evilweasel: Aug 24, 2002

Shifty Pony posted:

also the crowdsourced data is meaningless without review and classification by a trained person. even if the car didn't crash and the driver didn't notice anything odd to report there's still no way to know if the AI operated correctly without manual review of the data it used.

there's a big difference between "successfully tracked the lane markers to follow the road through a curve" and "thought a tree shadows were the line but that happened to be going the right direction at this time of day so it still kept on the road."

a single driver can collect more data in a day than a trained worker can review and classify in a week.

im not entirely sure this is correct, at least in theory. as long as it is reasonable to assume that a manual override indicates that there is a high probability that what the system did was wrong, you have had some level of "review and classification by a trained person". it's not nearly as good as you'd want it to be and the data will be quite noisy - in particular you can't assume all the uncorrected behavior was right, merely that it is more likely to be right than the corrected behavior was - but at least in theory you could use it to improve the system in an automated way.

further, you could (again, in theory) use it to determine what requires review and classification by a trained person: your review is going to want to focus on the data where there was a manual correction, allowing you to focus on the most important data in your flood of driving data, and if you can figure out what kinds of situations create abnormally high manual corrections you've identified even more relevant data.

# ? Mar 20, 2019 18:04

a few DRUNK BONERS: Mar 25, 2016

Shifty Pony posted:

also the crowdsourced data is meaningless without review and classification by a trained person. even if the car didn't crash and the driver didn't notice anything odd to report there's still no way to know if the AI operated correctly without manual review of the data it used.

there's a big difference between "successfully tracked the lane markers to follow the road through a curve" and "thought a tree shadows were the line but that happened to be going the right direction at this time of day so it still kept on the road."

a single driver can collect more data in a day than a trained worker can review and classify in a week.

and if you're not google it's a little hard to get people to click on pictures of bridges for free millions of times a day

# ? Mar 20, 2019 18:33

mystes: May 31, 2006

H.P. Hovercraft posted:

the amount of data that would need to push is actually one of the problems with implementing autonomous vehicles

Not to say they're actually doing it, but it probably wouldn't be impossible in this case if they only sent the data for a few seconds around when people override autopilot.

Heck, they could just collection locations where autopilot is overridden and then send someone over to drive a test vehicle in places where it happened several times.

# ? Mar 20, 2019 18:45

Luigi Thirty: Apr 30, 2006; Emergency confection port.

https://twitter.com/consequence/status/1108399347438358529

# ? Mar 20, 2019 19:23

psiox: Oct 15, 2001; Babylon 5 Street Team

cool myspace photo

imagining elon thirstily commenting

# ? Mar 20, 2019 19:53

Lutha Mahtin: Oct 10, 2010; Your brokebrain sin is absolved...go and shitpost no more!

Munkeymon posted:

sure they will if traffic is flowing. they also obey the speed limit which makes it hilarious when leadfoot shitbirds try to use the HOV lanes behind a bus and the driver just doesn't respond to their usual tailgating

which MSP bus routes do this? the only routes i know that drive on the shoulder do so by driving on the shoulder next to the rightmost lane. if a bus route normally uses the shoulder, it would make no sense for them to go from the far right side of the road all the way over to the HOT lane on the far left side

e: also this type of maneuver wouldn't even be possible on one-third of the roads that have HOT lanes (394)

Lutha Mahtin fucked around with this message at 20:22 on Mar 20, 2019

# ? Mar 20, 2019 20:15

indigi: Jul 20, 2004; ^{how can we not talk about family

when family's all that we got?}

they should just make remotely piloted drone cars/buses

# ? Mar 20, 2019 21:08

Plank Walker: Aug 11, 2005

indigi posted:

they should just make remotely piloted drone cars/buses

and then all the remote drivers should team up and crash at a set time and date, solving transportation once and for all

# ? Mar 20, 2019 21:31

Silver Nitrate: Oct 17, 2005; WHAT

Lutha Mahtin posted:

e: also buses in metro MSP don't use the HOV lane. they get to drive on the shoulder of the road during rush hour past evvvvvverybody

this absolutely terrified the last person I brought into town.

# ? Mar 20, 2019 21:47

wargames: Mar 16, 2008; official yospos cat censor

why we can't have nice things

https://i.imgur.com/UMqDC14.mp4

# ? Mar 20, 2019 21:51

Arcteryx Anarchist: Sep 15, 2007; Fun Shoe

lol its always a fat guy in a sports car

# ? Mar 20, 2019 21:55

Chalks: Sep 30, 2009

maybe he's planning on working out by getting in and out of the car

# ? Mar 20, 2019 22:04

Lightbulb Out: Apr 28, 2006; slack jawed yokel

i'm a skinny guy and had a hell of a time getting in and out of an i8

# ? Mar 20, 2019 22:22

Luigi Thirty: Apr 30, 2006; Emergency confection port.

https://twitter.com/dhh/status/1108480192975011841

# ? Mar 20, 2019 22:29

indigi: Jul 20, 2004; ^{how can we not talk about family

when family's all that we got?}

I usually like BMWs but the i8 is a really stupid car

# ? Mar 20, 2019 22:29

My Linux Rig: Mar 27, 2010; Can't post for 6 years!

H.P. Hovercraft posted:

the amount of data that would need to push is actually one of the problems with implementing autonomous vehicles

easy solution: just follow the musk model and remove all the sensors except gps and camera

what could go wrong?

Luigi Thirty posted:

https://twitter.com/dhh/status/1108480192975011841

but the autopilot was doing its job correctly! if it wasn�t interrupted, the driver would have died as intended

My Linux Rig fucked around with this message at 22:41 on Mar 20, 2019

# ? Mar 20, 2019 22:37

Lightbulb Out: Apr 28, 2006; slack jawed yokel

indigi posted:

I usually like BMWs but the i8 is a really stupid car

i got to drive one for a day and it was super fun and definitely sold me on a cool electric hybrid. it's just really impractical.

i beat on it for about 250 miles and got 50mpg

# ? Mar 20, 2019 22:50

indigi: Jul 20, 2004; ^{how can we not talk about family

when family's all that we got?}

Lightbulb Out posted:

i got to drive one for a day and it was super fun and definitely sold me on a cool electric hybrid. it's just really impractical.

i beat on it for about 250 miles and got 50mpg

does it feel like driving a BMW? I get >50mpg on my Prius (usually doing 75-80mph on my commute) but it doesn't feel like it has much power once you're over about 70

# ? Mar 20, 2019 22:57

Mr. Nice!: Oct 13, 2005; c-spam cannot afford

i've been averaging high 30s low 40s for the lifetime of my vw jetta. it's got a 1.4l turbocharged engine and will flat loving move if i put the gas down but will also get insanely good gas mileage both city and highway. i average 45mpg on the highway and can easily pass anyone or match speed with the maniacs on the florida turnpike and i75 without trouble.

i drove my friend's prius to texas and back and it got around the same mileage (it was an older model) but it did not have near the get up and go of the jetta. its crazy that this has one of the smallest engine sizes i've ever owned (including motorcycles) but it's also one of the fastest.

# ? Mar 20, 2019 23:02

drgitlin: Jul 25, 2003; luv 2 get custom titles from a forum that goes into revolt when its told to stop using a bad word.

The i8 is a weird one. the coup� blew me away when I tested one in 2015. It felt 200kg lighter than it is, and 150hp more powerful than the specs. and even if you drove it like you stole it it would still do 35mpg.

Then they sent me the i8 Roadster at the end of last year and it was utterly dire. Creaked and rattled over every bump and barely had enough power to get out of its own way.

# ? Mar 20, 2019 23:02

indigi: Jul 20, 2004; ^{how can we not talk about family

when family's all that we got?}

Mr. Nice! posted:

drove my friend's prius to texas and back and it got around the same mileage (it was an older model) but it did not have near the get up and go of the jetta.

yeah you really have to stand on it if you want decent acceleration once you're going highway speeds. I figured BMW would be able to do better but

drgitlin posted:

Then they sent me the i8 Roadster at the end of last year and it was utterly dire. Creaked and rattled over every bump and barely had enough power to get out of its own way.

# ? Mar 20, 2019 23:28

theflyingorc: Jun 28, 2008; ANY GOOD OPINIONS THIS POSTER CLAIMS TO HAVE ARE JUST PROOF THAT BULLYING WORKS; Young Orc

Why is autopilot legal

# ? Mar 20, 2019 23:53

EVGA Longoria: Dec 25, 2005; Let's go exploring!

theflyingorc posted:

Why is autopilot legal

air bud law

# ? Mar 20, 2019 23:58

Adbot: ADBOT LOVES YOU

# ? Apr 24, 2024 08:26

Not a Children: Oct 9, 2012; Don't need a holster if you never stop shooting.

theflyingorc posted:

Why is autopilot legal

because it could theoretically make rich people a lot of money

# ? Mar 21, 2019 00:00

The Something Awful Forums > Discussion > Serious Hardware/Software Crap > YOSPOS > TSLA: When this baby hits 88 you're going to see some serious shit

«‹›3418 »