Register a SA Forums Account here!
JOINING THE SA FORUMS WILL REMOVE THIS BIG AD, THE ANNOYING UNDERLINED ADS, AND STUPID INTERSTITIAL ADS!!!

You can: log in, read the tech support FAQ, or request your lost password. This dumb message (and those ads) will appear on every screen until you register! Get rid of this crap by registering your own SA Forums Account and joining roughly 150,000 Goons, for the one-time price of $9.95! We charge money because it costs us money per month for bills, and since we don't believe in showing ads to our users, we try to make the money back through forum registrations.
 
  • Post
  • Reply
FAUXTON
Jun 2, 2005

spero che tu stia bene

Discendo Vox posted:

classic example of trunk cost fallacy

trunkcaust

Adbot
ADBOT LOVES YOU

muckswirler
Oct 22, 2008

TK-42-1 posted:

only in an accident so while your head is being whipped backwards they deploy and launch your noggin into the fuckin stratosphere

Combat Theory
Jul 16, 2017


Hahahahahahahahahhaha
:tesla:

obstipator
Nov 8, 2009

by FactsAreUseless
https://youtu.be/9YBBc2eg27g

doesnt seem to show anything for traffic lights, signs, road markings, or speed bumps

CommieGIR
Aug 22, 2006

The blue glow is a feature, not a bug


Pillbug

obstipator posted:

https://youtu.be/9YBBc2eg27g

doesnt seem to show anything for traffic lights, signs, road markings, or speed bumps

"Firetruck acquired, you have 20 seconds to comply."

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.

obstipator posted:

https://youtu.be/9YBBc2eg27g

doesnt seem to show anything for traffic lights, signs, road markings, or speed bumps

it doesn't do that yet

it also still projects a driving path through buildings 20 feet ahead of it

Potato Salad
Oct 23, 2014

nobody cares


Squinky v2.0 posted:



don’t be a fre*akin assh*le
even if it’s not gore, tag it

Are there any brokers NOT making margin calls at this point?

PIZZA.BAT
Nov 12, 2016


:cheers:


guy working here was talking about buying a tesla. i showed him some of the recent headlines and told him to wait at least a quarter to ‘make sure the company is safe’ before buying. now he’s digging on his own and discovering the lol goldmine so mission accomplished

Alighieri
Dec 10, 2005


:dukedog:

They should just turn the seats in the 3rd row around and make it into a wagon out of the 80's/90's.

Jabor
Jul 16, 2010

#1 Loser at SpaceChem
turn the second row around and call it a limousine

Endless Mike
Aug 13, 2003



Alighieri posted:

They should just turn the seats in the 3rd row around and make it into a wagon out of the 80's/90's.
that's what they did in the model s

Nfcknblvbl
Jul 15, 2002

Endless Mike posted:

that's what they did in the model s

and it was awful. rich rebuilds did a cabin temperature test in the summer with the ac blowing in the car, and the cool air never made it to the third row so youre gonna cook your kids.

El_Elegante
Jul 3, 2004

by Jeffrey of YOSPOS
Biscuit Hider
now you’re just making me hungry

Arcteryx Anarchist
Sep 15, 2007

Fun Shoe
whenever i see that tesla camera video, it reminds me that they must be doing object detection, and that's still an area in machine vision that's been going through a massive amount of improvement in only more recent history

on the one hand, i can see tesla as being the type of company to just throw everything out for something hot and new-ish, but on the other hand i can also see them being a company that would throw thousands of hours trying to make a hot pile of poo poo methodology from a number of years ago optimized/horsepowered enough to work for video in that context

Just-In-Timeberlake
Aug 18, 2003
well eyes are good enough for people, cameras should be more than sufficient

*ignores the fact that eyes are good enough because there's a brain doing something with the input

FMguru
Sep 10, 2003

peed on;
sexually

Just-In-Timeberlake posted:

there's a brain doing something with the input
lol at actually thinking this in the tesla thread

Just-In-Timeberlake
Aug 18, 2003

FMguru posted:

lol at actually thinking this in the tesla thread

i'd trust anybody in these dead forums to drive me anywhere over a tesla on autopilot

poo poo, my dog is better at not running into walls than a tesla is

Nfcknblvbl
Jul 15, 2002

autopilot does some wacky poo poo, i bet tesla doesnt know why it makes weird decisions. autopilot randomly slams on the brakes when its overtaking trucks on the highway, and it makes no sense.

Arcteryx Anarchist
Sep 15, 2007

Fun Shoe
i feel like musk has gotten into it before, granted he's a moron that likes to make poo poo up, but I wonder how much of it is AI vs a more traditional approach

like the vision stuff is obviously being handled from an AI direction, but I wonder how much the overall driving decisions are

fack you
Sep 12, 2002

For Life

Nfcknblvbl posted:

autopilot does some wacky poo poo, i bet tesla doesnt know why it makes weird decisions. autopilot randomly slams on the brakes when its overtaking trucks on the highway, and it makes no sense.

please stop using autopilot on public roads you are going to kill somebody

El_Elegante
Jul 3, 2004

by Jeffrey of YOSPOS
Biscuit Hider
else. somebody else.

Lord Stimperor
Jun 13, 2018

I'm a lovable meme.

Combat Theory posted:

Hahahahahahahahahhaha
:tesla:

Hey CT was wondering if I could ask you something. Car stuff. You got some way I can message you? I mean I don't want to bother you though.

leftist heap
Feb 28, 2013

Fun Shoe

lancemantis posted:

i feel like musk has gotten into it before, granted he's a moron that likes to make poo poo up, but I wonder how much of it is AI vs a more traditional approach

like the vision stuff is obviously being handled from an AI direction, but I wonder how much the overall driving decisions are

what's the difference?

Just-In-Timeberlake
Aug 18, 2003
just if/then/else statements stretching to infinity

Arcteryx Anarchist
Sep 15, 2007

Fun Shoe
basically writing a program that more or less explicitly says "given these sensor inputs x, perform y" vs trying to learn "given these sensor inputs x, perform y" in the hope that it will be better generalized/more robust/etc

evilweasel
Aug 24, 2002

Just-In-Timeberlake posted:

well eyes are good enough for people, cameras should be more than sufficient

*ignores the fact that eyes are good enough because there's a brain doing something with the input

eyes are fine, as long as you have a brain with a large vision processing center that took hundreds of millions of years of evolution to develop and that still takes sixteen years of growth and feedback to mature sufficiently to be trusted to drive

Arcteryx Anarchist
Sep 15, 2007

Fun Shoe
the underlying philosophy of the current AI-centric approach to robot cars is pretty much, as ideal, "what if we could make a human being that didn't make mistakes drive the car"

where those mistakes at this point seem to be bounded to "won't get distracted in specific ways like having a conversation or looking at their phone while driving" and maybe having a larger simultaneous field of vision

Chalks
Sep 30, 2009

ai is currently just a fancy party trick. the real world requires way too much semantic knowledge to properly navigate.

in ideal conditions, you'll be better than a human driver, and outside of that you'll be worse way worse.

BobHoward
Feb 13, 2012

The only thing white people deserve is a bullet to their empty skull

Chalks posted:

ai is currently just a fancy party trick. the real world requires way too much semantic knowledge to properly navigate.

in ideal conditions, you'll be better than a human driver, and outside of that you'll be worse way worse.

this. teaching a *nn of the complexity we can build today to drive seems akin to training something lesser than an insect brain to drive. there isn’t much higher order understanding of the scene or object permanence

this is why it sometimes panic brakes or lane swaps. if the system identifies “obstacle ahead” there’s no higher layer filtering out cases like “nah that‘s just a trick of shadow, we could see nothing was there a moment ago, and the illusion will disappear a second from now”.

Combat Theory
Jul 16, 2017

Lord Stimperor posted:

Hey CT was wondering if I could ask you something. Car stuff. You got some way I can message you? I mean I don't want to bother you though.

you can hop into ol' man smarts discord and ill find you there and get you an invite to our goony place

E: should shoot me a message ITT too

Combat Theory fucked around with this message at 00:47 on May 2, 2019

big scary monsters
Sep 2, 2011

-~Skullwave~-

BobHoward posted:

this. teaching a *nn of the complexity we can build today to drive seems akin to training something lesser than an insect brain to drive. there isn’t much higher order understanding of the scene or object permanence

this is why it sometimes panic brakes or lane swaps. if the system identifies “obstacle ahead” there’s no higher layer filtering out cases like “nah that‘s just a trick of shadow, we could see nothing was there a moment ago, and the illusion will disappear a second from now”.
i mean neural networks can absolutely do that, see rnns for instance, but i'm not sure that's actually the issue. it's more likely a failure earlier in the processing. from working on similar vision applications i'd guess tesla's pipeline for an image frame goes something like
pre:
scene segmentation/object detection -> object recognition -> labelling -> neural network -> output action
so rather than feeding the big, expensive images to the nn it just gets labelled features with locations, does its magic there and spits out whatever the car should do next. maybe there are multiple nns, maybe it gets input integrated from all sensors, whatever

object trackers that can handle a few frames of missed detection without merging overlapping objects or labelling previously seen ones as new are hardly unusual, and why tesla's vision system doesn't have something like that isn't really clear to me. from the videos we've seen it seems like it doesn't - bounding boxes are jittering all over the places, detection stutters as it loses and reacquires objects, objects are relabelled from frame to frame. but otoh i have to assume it's calculating movement vectors or how could it predict where objects in the scene are going to be next frame (or in ten or a hundred frames) and react correctly? that goes hand in hand with tracking so either something is going on under the covers that isn't obvious in the video or the tesla engineers are doing something really mystifying. tesla is a garbage company but they had plenty of money until this year and you have to imagine they were hiring at least the occasional competent person for their flagship autopilot project so ???

Just-In-Timeberlake
Aug 18, 2003
i was driving the other day and if you're in any way an observant driver you're constantly assessing other drivers/pedestrians/whatever not on what they're doing, but on what they might do. i don't know how to explain it better, other than your mind flags stuff you see (kids are playing ball in the yard, watch out in case the ball ends up in the street and one of them runs out without looking) and you subconsciously anticipate it.

no way anybody's nn does any of that.

Silver Alicorn
Mar 30, 2008

𝓪 𝓻𝓮𝓭 𝓹𝓪𝓷𝓭𝓪 𝓲𝓼 𝓪 𝓬𝓾𝓻𝓲𝓸𝓾𝓼 𝓼𝓸𝓻𝓽 𝓸𝓯 𝓬𝓻𝓮𝓪𝓽𝓾𝓻𝓮
can't wait for people to buy the model Y and insist they love the car and all their friends use the third row

PIZZA.BAT
Nov 12, 2016


:cheers:


Just-In-Timeberlake posted:

i was driving the other day and if you're in any way an observant driver you're constantly assessing other drivers/pedestrians/whatever not on what they're doing, but on what they might do. i don't know how to explain it better, other than your mind flags stuff you see (kids are playing ball in the yard, watch out in case the ball ends up in the street and one of them runs out without looking) and you subconsciously anticipate it.

no way anybody's nn does any of that.

yeah this is what a lot of us have been saying since day one. to get a true AI that's capable of driving a car you have to get an AI that's capable of empathy because that's the only way you're going to be able to predict the actions of other drivers / pedestrians

big scary monsters
Sep 2, 2011

-~Skullwave~-
yeah totally, the feeling you get that something isn't quite right with the car in front so you slow down just before they slam on their brakes to take that offramp they nearly missed or whatever. i mean we're a way off but the whole idea is to emulate (simulate?) human driving by learning from human drivers so who knows, it's not entirely outside the realm of possibility that it might get there

the fun part of nns is that they're completely incomprehensible black boxes and you have little idea why they do anything at all. that whole deep dream thing google released wasn't just for making creepy pictures, they were trying to gain some small understanding of what their imagenet cnn was even doing at a feature level. that's one thing that's pretty concerning about machine learning being thrown at every problem - sure the performance looks great compared with boring old hand-written image processing algorithms, but with that kind of oldschool ai you can read the code, understand what's happening and why, and it's predictable and testable. nns can be deterministic so i guess testing is theoretically possible, but it seems like it would be incredibly costly and difficult. i'm a couple years out of working in that field so i don't know what the latest is, but i'm not aware that the kind of full coverage testing exists for these systems that you would want if you were, for instance, certifying software for flying planes or sending a spacecraft into orbit

but because no proper regulation exists for the software that drives cars you get to share the road with these unpredictable, unknowable, untestable ais that are in this case also dumb as a loving rock

big scary monsters fucked around with this message at 02:21 on May 2, 2019

Shipon
Nov 7, 2005

Rex-Goliath posted:

yeah this is what a lot of us have been saying since day one. to get a true AI that's capable of driving a car you have to get an AI that's capable of empathy because that's the only way you're going to be able to predict the actions of other drivers / pedestrians

until we have human level AI self-driving cars will never work as well as a human driver

and if we somehow do develop human-level AI, all bets are off because who the gently caress knows how society will ever accommodate that, the way labor even works will be changed

Agile Vector
May 21, 2007

scrum bored



Silver Alicorn posted:

can't wait for people to buy the model Y and insist they love the car and all their friends use the third row

and everyone sat down comfortable and clapped

Shipon posted:

until we have human level AI self-driving cars will never work as well as a human driver

and if we somehow do develop human-level AI, all bets are off because who the gently caress knows how society will ever accommodate that, the way labor even works will be changed

the real achievement will be when uber presses for ai citizenship so they can start selling new instances loans for the cars they live in

Shipon
Nov 7, 2005

Agile Vector posted:

and everyone sat down comfortable and clapped


the real achievement will be when uber presses for ai citizenship so they can start selling new instances loans for the cars they live in

time to organize the AI into a union

obstipator
Nov 8, 2009

by FactsAreUseless
to get a true ai, you also need to get the ai to participate in road rage

Adbot
ADBOT LOVES YOU

Endless Mike
Aug 13, 2003



tesla stock went up 5% from ol musky saying he doesn't know where they'll get money from in the future.

  • 1
  • 2
  • 3
  • 4
  • 5
  • Post
  • Reply