[Musk] he's the dog that caught the car, but the car has fsd and is running him over

The Something Awful Forums > Discussion > C-SPAM > [Musk] he's the dog that caught the car, but the car has fsd and is running him over

«‹›4544 »

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.

zegermans posted:

lol it doesn't just ignore trailers, it actively suicides you into them

the car decides it'll fit under, and it will

you however...

# ? Sep 25, 2020 13:51

Adbot: ADBOT LOVES YOU

# ? Apr 29, 2024 18:04

Zerg Mans: Oct 19, 2006

infernal machines posted:

the car decides it'll fit under, and it will

you however...

for sale, rare Tesla Convertible!

# ? Sep 25, 2020 13:54

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.

there are at least two of them

# ? Sep 25, 2020 13:56

Nothus: Feb 22, 2001; Buglord

infernal machines posted:

https://www.wired.com/story/tesla-a...hy-crash-radar/ posted:
Raj Rajkumar, who researches autonomous driving at Carnegie Mellon University, thinks those assumptions concern one of Tesla's key sensors. �The radars they use are apparently meant for detecting moving objects (as typically used in adaptive cruise control systems), and seem to be not very good in detecting stationary objects," he says.

That's not nearly as crazy as it may seem. Radar knows the speed of any object it sees, and is also simple, cheap, robust, and easy to build into a front bumper. But it also detects lots of things a car rolling down the highway needn't worry about, like overhead highway signs, loose hubcaps, or speed limit signs. So engineers make a choice, telling the car to ignore these things and keep its eyes on the other cars on the road: They program the system to focus on the stuff that's moving.

They mean focus on the stuff that's not moving very fast, right? Isn't movement detected as being relative to the radar unit? So other cars are moving more slowly than signs and utility poles flying by?

# ? Sep 25, 2020 13:59

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.

yes, that's not worded very well. iirc, they also make an assumption that any vehicle they identify while moving is also moving, because they're on a highway (at least back when AP was only for use on limited access highways). of course it's very obvious how this assumption fails, but that's never slowed them down any.

# ? Sep 25, 2020 14:07

Xakura: Jan 10, 2019; A safety-conscious little mouse!

infernal machines posted:

https://twitter.com/greentheonly/status/1105996443658129408

and

Holy loving poo poo, stationary objects go in the notch filter? Of course they do. :psyduck:

# ? Sep 25, 2020 14:18

WhyteRyce: Dec 30, 2001

Just-In-Timeberlake posted:

leveled with stacks of nickels

Woah Grimes is installing the cameras in the car

# ? Sep 25, 2020 14:34

moist turtleneck: Jul 17, 2003; Represent.; Dinosaur Gum

drat I had to go check and grimes is following musk again on twitter

# ? Sep 25, 2020 14:36

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

I still maintain that the "drivable space under a truck" thing is because the system can't reliably tell the difference between a truck and a billboard and they had to add in the exception to avoid the system slamming on the brakes on interstates.

# ? Sep 25, 2020 15:46

Just-In-Timeberlake: Aug 18, 2003

Shifty Pony posted:

I still maintain that the "drivable space under a truck" thing is because the system can't reliably tell the difference between a truck and a billboard and they had to add in the exception to avoid the system slamming on the brakes on interstates.

Every self driving car has had to have its rules relaxed otherwise the rides were herky jerky and unbearable.

# ? Sep 25, 2020 16:00

Just-In-Timeberlake: Aug 18, 2003

https://www.usnews.com/news/technology/articles/2020-09-15/back-up-driver-in-arizona-2018-fatal-uber-self-driving-car-crash-charged-in-death

quote:

WASHINGTON (Reuters) - The back-up safety driver behind the wheel of a self-driving Uber Technologies test vehicle that struck and killed a woman in Tempe, Arizona, in 2018 was charged with negligent homicide, prosecutors said on Tuesday.

Rafael Vasquez, age 46, who is also known as Rafaela, pleaded not guilty on Tuesday after being charged in the death of Elaine Herzberg on Aug. 27, court records show. She was released pending trial set for February 2021.

Herzberg died after she was struck while walking a bicycle across a street at night. The first recorded death involving a self-driving vehicle prompted significant safety concerns about the nascent autonomous vehicle industry.

Uber declined comment. A lawyer for Vasquez did not immediately respond to a request to comment.

A Tempe police report said Vasquez was repeatedly looking down instead of keeping her eyes on the road. Prosecutors in March 2019 said Uber was not criminally liable in the crash.

�Distracted driving is an issue of great importance in our community,� said Maricopa County Attorney Allister Adel. "When a driver gets behind the wheel of a car, they have a responsibility to control and operate that vehicle safely."

Police said previously the crash was "entirely avoidable" and that Vasquez was streaming "The Voice" TV program at the time of the crash.

In November, the National Transportation Safety Board (NTSB) faulted Vasquez's inactions and Uber for inadequate attention to safety and decisions in the company's autonomous vehicle development.

The NTSB said the probable cause was Vasquez's failure to monitor the driving environment �because she was visually distracted throughout the trip by her personal cell phone." She was supposed to act in the event of an emergency.

Uber made a series of development decisions that contributed to the crash's cause, the NTSB said. The software in the modified Volvo XC90 did not properly identify Herzberg as a pedestrian and did not address "operators� automation complacency."

Uber deactivated the automatic emergency braking systems in the Volvo XC90 vehicle and precluded the use of immediate emergency braking, relying instead on the back-up driver.

read that last line, they did that to stop it from braking all the time.

also lol at Uber not bearing any responsibility

# ? Sep 25, 2020 16:04

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

Just-In-Timeberlake posted:

Every self driving car has had to have its rules relaxed otherwise the rides were herky jerky and unbearable.

the google self driving cars used to tootle around my neighborhood and they drove like the worlds oldest grandma but braked like they gave control of that system to a squirrel that just did their entire body's weight worth of cocaine.

# ? Sep 25, 2020 16:04

ekuNNN: Nov 27, 2004; by Jeffrey of YOSPOS

Shifty Pony posted:

the google self driving cars used to tootle around my neighborhood and they drove like the worlds oldest grandma but braked like they gave control of that system to a squirrel that just did their entire body's weight worth of cocaine.

That's what happens when you actually care about safety I'd guess :v:

# ? Sep 25, 2020 16:12

Communist Zombie: Nov 1, 2011

Cross posting from the Space Thread

DJ Burette posted:

DoD confirming what everyone suspected about Starlink's business case; they'll pay enough initially to ensure that the US has the first and largest working LEO constellation until there's enough customer demand to pay the costs.

https://twitter.com/thesheetztweetz/status/1309250704389931008?s=09

So it seems Telsa is atleast up to par with the MIC, which kinda makes sense since theyre both not supposed to deliver products but siphon money from the buyer.

# ? Sep 25, 2020 18:00

yippee cahier: Mar 28, 2005

Who wants to start the scam company that gets the contract for studying the feasibility of blanketing the Pacific Ocean in ground stations with me?

# ? Sep 25, 2020 18:05

Gods_Butthole: Aug 9, 2020; Can't post for 8 years!

Vincent Van Goatse posted:

Trouble is most if not all volcanic activity is on Hawai'i and most of the population is on O'ahu and there's a lot of ocean between the two making power transmission something of an issue.

Just set up a wireless microwave energy transmission system. Bing bong it's so simple!

# ? Sep 25, 2020 18:21

Gods_Butthole: Aug 9, 2020; Can't post for 8 years!

Karl Sharks posted:

no it's fine, they'll do an over the air update and it'll work

Just download more RAM, nbd

# ? Sep 25, 2020 18:24

Lady Militant: Apr 8, 2020; The history of all hitherto existing society is the history of class struggles.

infernal machines posted:

the car decides it'll fit under, and it will

you however...

it honestly looks better with the top off lol

# ? Sep 25, 2020 18:27

Lady Militant: Apr 8, 2020; The history of all hitherto existing society is the history of class struggles.

it'd be perfectly in character with WHACKY ENGINEERING MAN to release a convertable line of your car cause you saw how good it looked after the top got sheared off but elon is as terrible of an actor as he is a engineer

# ? Sep 25, 2020 18:28

Gods_Butthole: Aug 9, 2020; Can't post for 8 years!

Samovar posted:

This man seems to have severe reading problems.

You don't get to be a rich savvy investor by reading, you get there by leading.

# ? Sep 25, 2020 18:30

Three Olives: Apr 10, 2005; I FUCKING HATE POOR PEOPLE BUT I LOVE BEING FUCKED IN THE ASS and having two dishwashers in my CONDO!!!!!!!!!!!!!!!!!!!!

Lady Militant posted:

it honestly looks better with the top off lol

There are a few coach builders that will turn basically anything into a convertible, and yes, that includes a few Teslas. They look like poo poo though.

https://insideevs.com/news/325710/135000-tesla-model-s-convertible-for-sale/

# ? Sep 25, 2020 19:58

Ytlaya: Nov 13, 2005

Three Olives posted:

What? It's a great car. You are insane if you buy one brand new because of the deprecation but leasing/buying used is a great deal.

It's nuts how that car costs around the same (and realistically less if buying used and accounting for most Model 3's being sold at a much higher price than their "base price") than a Model 3 while actually looking pretty good and not like a cheap piece of poo poo.

# ? Sep 25, 2020 20:24

pigz: Jul 12, 2004; Nearly as overlooked as Joe Mauer

Sudden Loud Noise posted:

It's just a giant manually ordered if-then-else decision tree. Every single time a new variable is introduced (like an overpass or someone has a bike with training wheels) everything has to be rethought. What if there is someone in the middle of the road but they are under an overpass, am I allowed to hit them? It's impossible to know.

It's a great system.

yup, and I'm not sure like 99% of the people realize this.

The alternative would likely be some sort of Reinforcement Learning setup that took in various sensors and state information and output car control surface displacements. Doing this would still require heuristics for processing input and output regardless for obvious reasons.

That says nothing that no one in the world would be able to tell you why any specific output came out of any model. And, depending on the input data, methods and complexity of your output data, they may not even to be able to tell you the range of data that *could* come out.

Consider if you somehow made a ml fake noise crowd simulator.. You have the raw crowd noise from the nba for every game ever televised , so you train it up and bring it to espn and the first question they ask you is "will this accidently output swears or other offensive things" and the answer will be 'probably' and they'd be like well get back to us when it doesn't.

# ? Sep 25, 2020 20:45

Divot: Dec 23, 2013

I'm sure I'm correct in assuming that the whole pitch of 'Tesla has an advantage for developing self-driving because their whole fleet is constantly transmitting data that continuously trains their model and helps the cars learn' and all that is total hogwash right?

Like the little I know about how ML stuff actually works it doesn't seem to make any sense. Stuff that was posted earlier showing their software being only vision-based classifiers seems to confirm this.

Like what could the cars / software be 'learning' by all this data that's supposedly being used to train their poo poo in real time?

Seems like total fiction meant to convince the less technically literate that their poo poo is the best and brightest.

Robotaxi fleet my rear end.

Divot has issued a correction as of 21:05 on Sep 25, 2020

# ? Sep 25, 2020 20:52

Spatial: Nov 15, 2007

Yes it's bullshit. Most of what Tesla says is bullshit, they have excellent marketing and tailor it very carefully for their target audience: people who argue in the comments of android phone reviews every single day.

# ? Sep 25, 2020 22:06

Just-In-Timeberlake: Aug 18, 2003

Lady Militant posted:

it honestly looks better with the top off lol

Simone Giertz turned her Tesla into a pickup (more like a Holden ute or El Camino) and Tesla asked her to not bring it to the Cybertruck reveal because it�s better than that hunk of poo poo in every way

https://insideevs.com/news/385216/tesla-truckla-cybertruck-update-video/

# ? Sep 25, 2020 23:29

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

Divot posted:

I'm sure I'm correct in assuming that the whole pitch of 'Tesla has an advantage for developing self-driving because their whole fleet is constantly transmitting data that continuously trains their model and helps the cars learn' and all that is total hogwash right?

Like the little I know about how ML stuff actually works it doesn't seem to make any sense. Stuff that was posted earlier showing their software being only vision-based classifiers seems to confirm this.

Like what could the cars / software be 'learning' by all this data that's supposedly being used to train their poo poo in real time?

Seems like total fiction meant to convince the less technically literate that their poo poo is the best and brightest.

Robotaxi fleet my rear end.

99.99% horseshit.

the hard part of getting good training data is that you need to to have someone go through and classify everything in it. that requires the full video which as far as anyone can tell Teslas do not collect.

the 0.01% truth is that they do record the times and locations of where autopilot disengaged or the driver took over manually. that lets them see spots where the system is loving up. a good company would then send out a team to check out what's unique about that spot, Tesla just disables whichever classifier was causing the disengagement.

even that's not all that useful because you miss all the times when the system should have disengaged but didn't.

# ? Sep 25, 2020 23:50

Just-In-Timeberlake: Aug 18, 2003

Shifty Pony posted:

99.99% horseshit.

the hard part of getting good training data is that you need to to have someone go through and classify everything in it. that requires the full video which as far as anyone can tell Teslas do not collect.

the 0.01% truth is that they do record the times and locations of where autopilot disengaged or the driver took over manually. that lets them ~~see spots where the system is loving up. a good company would then send out a team to check out what's unique about that spot, Tesla just disables whichever classifier was causing the disengagement.~~ avoid all liability by pinning the blame on the driver

even that's not all that useful because you miss all the times when the system should have disengaged but didn't.

# ? Sep 25, 2020 23:56

fits my needs: Jan 1, 2011; Grimey Drawer

https://twitter.com/imjuliebrown/status/1309591619629318144?s=20

# ? Sep 26, 2020 00:28

Karl Sharks: Feb 20, 2008; The Immortal Science of Sharksism-Fininism

fits my needs posted:

https://twitter.com/imjuliebrown/status/1309591619629318144?s=20

quote:

SFGATE: I�ve read about some Tesla batteries catching fire, and California is obviously having some issues in that department. Is any of that a concern?

MORRIS: I�ve been stuck inside for the better part of three weeks because of poor air quality, and we�ve had friends who�ve had to evacuate their homes, so it�s a huge threat.

For the first eight months, we had teams of people and machinery on the property clearing as much dead brush and hazardous material as possible, creating firebreaks, and installing water storage tanks that Cal Fire can use. Cal Fire has been instrumental in guiding our efforts with fire mitigation. We even bought our own fire truck for the property � it�s pretty awesome.

"so teslas catch fire easily, and the battery is in the bottom, wtf?"
"dw we talked to fire folks and cleared bushes out"

# ? Sep 26, 2020 00:56

thatfatkid: Feb 20, 2011; by Azathoth

I like how the only reason teslas don't have lidar for the self driving is because its expensive. Musk is such a typical fuckwit engineer/manager.

# ? Sep 26, 2020 02:51

Spazzle: Jul 5, 2003

thatfatkid posted:

I like how the only reason teslas don't have lidar for the self driving is because its expensive. Musk is such a typical fuckwit engineer/manager.

Lidars are currently both expensive and unreliable long term on moving vehicles.

# ? Sep 26, 2020 03:09

comedyblissoption: Mar 15, 2006

pigz posted:

yup, and I'm not sure like 99% of the people realize this.

The alternative would likely be some sort of Reinforcement Learning setup that took in various sensors and state information and output car control surface displacements. Doing this would still require heuristics for processing input and output regardless for obvious reasons.

That says nothing that no one in the world would be able to tell you why any specific output came out of any model. And, depending on the input data, methods and complexity of your output data, they may not even to be able to tell you the range of data that *could* come out.

Consider if you somehow made a ml fake noise crowd simulator.. You have the raw crowd noise from the nba for every game ever televised , so you train it up and bring it to espn and the first question they ask you is "will this accidently output swears or other offensive things" and the answer will be 'probably' and they'd be like well get back to us when it doesn't.

i watched some starcraft 2 ai neural network videos where it played on the ladder, and the ai would sometimes as part of its build order start a building, cancel it very shortly after, then start building it again in the same spot

people want this type of opaque system with bizarre behaviors to drive cars lol

# ? Sep 26, 2020 05:33

Hillary 2024: Nov 13, 2016; by vyelkin

https://twitter.com/elonmusk/status/1309390263333289986?s=20
https://twitter.com/elonmusk/status/1240754657263144960?s=20

# ? Sep 26, 2020 05:43

Stefan Prodan: Jan 7, 2002; I deeply respect you as a human being... Some day I'm gonna make you *Mrs* Buck Turgidson!; Grimey Drawer

comedyblissoption posted:

i watched some starcraft 2 ai neural network videos where it played on the ladder, and the ai would sometimes as part of its build order start a building, cancel it very shortly after, then start building it again in the same spot

people want this type of opaque system with bizarre behaviors to drive cars lol

https://www.youtube.com/watch?v=Q70ulPJW3Gk

the first quarter of this video but with driving lol

# ? Sep 26, 2020 05:45

Waffle House: Oct 27, 2004; You follow the path
fitting into an infinite pattern.

Yours to manipulate, to destroy and rebuild.

Now, in the quantum moment
before the closure
when all become one.

One moment left.
One point of space and time.

I know who you are.

You are Destiny.

Spazzle posted:

Lidars are currently both expensive and unreliable long term on moving vehicles.

LIDAR is also currently not so great with inclement weather.

Lady Militant posted:

it honestly looks better with the top off lol

I am unironically hoping they do a droptop at some point.

# ? Sep 26, 2020 06:03

The Atomic Man-Boy: Jul 23, 2007

Yeah, neural network especially are prone to getting stuck in �local maximums,� where they do something less than ideal and can�t make the jump to doing something �smarter.�

Plus they completely poo poo the bed when they come across data that is completely unlike any data they�ve been trained on.

# ? Sep 26, 2020 08:19

The Atomic Man-Boy: Jul 23, 2007

Also since they are just long chains of matrices, were every number is interlinked with every other number, going back and figuring out how or why they gently caress up is just about impossible.

# ? Sep 26, 2020 08:35

Three Olives: Apr 10, 2005; I FUCKING HATE POOR PEOPLE BUT I LOVE BEING FUCKED IN THE ASS and having two dishwashers in my CONDO!!!!!!!!!!!!!!!!!!!!

The Atomic Man-Boy posted:

Also since they are just long chains of matrices, were every number is interlinked with every other number, going back and figuring out how or why they gently caress up is just about impossible.

This may or may not be true, I read this anecdote a million years ago but it was about how early the USPS got incredibly good at OCR of handprinted addresses and why other OCR just still completely sucks despite the fact that the USPS system can turn what even looks like gibberish to humans into actual addresses.

Basically the answer was the USPS was using a neural net to recognize addresses and literally no one had any loving clue how it worked, not even the developers, much less how it applied to other OCR applications. Just a huge loving black box that turned handwriting into addresses.

# ? Sep 26, 2020 09:28

Adbot: ADBOT LOVES YOU

# ? Apr 29, 2024 18:04

Hillary 2024: Nov 13, 2016; by vyelkin

infernal machines posted:

https://twitter.com/greentheonly/status/1105996443658129408

Don't worry, after 2-300 iterations the Tesla AI will learn that it can't drive under trailers and will start driving around them.

# ? Sep 26, 2020 10:00

The Something Awful Forums > Discussion > C-SPAM > [Musk] he's the dog that caught the car, but the car has fsd and is running him over

«‹›4544 »