|
other tampa goons back me up on this because i don't remember the details very much but we had like, a sting operation against uber drivers in our city and so the state government disbanded the entire city-level regulatory agency that regulates that poo poo and passed a law saying nobody can restrict uber stuff on a city level, under the guise of passing "state-level regulations" for how uber must certify drivers for "consistency across the state," which, surprise!, were basically what uber was already doing
|
# ? Nov 20, 2017 20:10 |
|
|
# ? Jun 9, 2024 19:20 |
|
http://www.clickhole.com/article/je...SocialMarketingquote:
|
# ? Nov 20, 2017 20:12 |
infernal machines posted:they have magic. I read a journal article that was basically the opposite: entirely focused on just how goddamn hard the processing and communications arrangement would be because it is basically race conditions all the way down, clock sync is nearly impossible, and packets aren't the only thing that can collide. I seem to recall it also concluded that complexity goes up faster than exponentially with vehicle and lane count and that attempting to scale by adding computation units was likely going to be counterproductive due to the additional race condition checks that would be required. I wish I had saved it. I think it was in one of the IEEE journals.
|
|
# ? Nov 20, 2017 20:14 |
|
ate all the Oreos posted:p much. lots of people, even smart people who should know better, think machine learning is some exponential process that one day will just hit the part of the curve that flies off to infinity and suddenly we'll have god-AI. https://twitter.com/danevandyck/status/751447473407684609 (he cannot)
|
# ? Nov 20, 2017 20:14 |
|
Cybernetic Vermin posted:this'd be a huge derail to get into discussing very deeply, but this is a wide-ranging assumption that it is somewhat invisible to people; deep learning techniques have had a of success when getting data thrown at them, but there is no actual argument or proof why they must necessarily keep going towards perfection in every scenario we imagine, such as self-driving cars it's not, "they perfectly self-drive", it's, "they self-drive 10x better than peeps" not a low bar, not an impossibly high bar. w/o rain, I think googcar is getting close (and only googcar) we knew that as soon as you poke a single goddamn thing combinatorial explosions fall out in neural net land, and we've known that since the 60's. but backprop already is a O(n^(2r)) to O(rn^2) speed improvement anyhow per minibatch so you know
|
# ? Nov 20, 2017 20:15 |
|
H.P. Hovercraft posted:don't forget the best part: magical intersection management goddamn think about how terrifying that would be to experience
|
# ? Nov 20, 2017 20:15 |
|
qirex posted:66% of downtown sf traffic citations are issued to tnc drivers [estimated to be 20% of traffic], the bulk of them are for stopping where they shouldn't and driving in the bus/taxi lane holy moly
|
# ? Nov 20, 2017 20:16 |
|
Shifty Pony posted:I read a journal article that was basically the opposite: entirely focused on just how goddamn hard the processing and communications arrangement would be because it is basically race conditions all the way down, clock sync is nearly impossible, and packets aren't the only thing that can collide. I seem to recall it also concluded that complexity goes up faster than exponentially with vehicle and lane count and that attempting to scale by adding computation units was likely going to be counterproductive due to the additional race condition checks that would be required. yeah, it pretty obviously doesn't work for low-latency realtime poo poo like organizing traffic flow, but i've seen this poo poo come up every year or so like clockwork since the darpa grand challenge was a thing
|
# ? Nov 20, 2017 20:16 |
|
C.H.O.M.E posted:goddamn think about how terrifying that would be to experience the grannykiller intersection design
|
# ? Nov 20, 2017 20:18 |
|
ate all the Oreos posted:p much. lots of people, even smart people who should know better, think machine learning is some exponential process that one day will just hit the part of the curve that flies off to infinity and suddenly we'll have god-AI. in constraint satisfaction problems, they even have a phase diagram spelled out like in physics and poo poo, with a region in the phase diagram where Nature basically says "lol go gently caress yourself"
|
# ? Nov 20, 2017 20:19 |
|
ate all the Oreos posted:p much. lots of people, even smart people who should know better, think machine learning is some exponential process that one day will just hit the part of the curve that flies off to infinity and suddenly we'll have god-AI. it will reach a limit, and then the smarty smart mans at the computer factory will write a new code and it will be better. i think even flawed computer driving machines will kill fewer people than people currently do.
|
# ? Nov 20, 2017 20:19 |
|
...And You Will Know Us by the Trail of the Dead would be a good motto for Tesla
|
# ? Nov 20, 2017 20:19 |
|
https://twitter.com/danevandyck/status/751490901252329472 no poo poo bob dobbs is dead posted:it's not, "they perfectly self-drive", it's, "they self-drive 10x better than peeps" when you can stop qualifying that statement it'll stop being an impossibly high bar.
|
# ? Nov 20, 2017 20:19 |
|
hifi posted:the type of people that eat pico de gallo. look at how stupid you are
|
# ? Nov 20, 2017 20:19 |
|
qirex posted:if you know anything about the sfpd they pretty much don't enforce traffic laws like ever so the reality is far, far worse about the only time i see them do enforcement is like hov lane near bryant at 4pm, or the harrison-3rd st shitshow usually has a traffic cop. tho you can p much drive in the bus-only lanes there, since there's like 100 cars infront of you and after you that they aren't going to pull you over
|
# ? Nov 20, 2017 20:20 |
|
crabrock posted:it will reach a limit, and then the smarty smart mans at the computer factory will write a new code and it will be better. this is a religious belief at this point. it's a statement of pure faith
|
# ? Nov 20, 2017 20:21 |
|
crabrock posted:i think even flawed computer driving machines will kill fewer people than people currently do. hope they can see through all inclement weather and understand missing pavement delineation and damaged signs like a human can or else it's a total nonstarter b/c right now it looks like you'd need strong ai to do those things lol
|
# ? Nov 20, 2017 20:22 |
|
C.H.O.M.E posted:goddamn think about how terrifying that would be to experience that's not even the scariest computer driving gif, I remember seeing a different one that tried to find a way to eliminate red lights entirely by tweaking follow distances so that intersecting traffic would be able to zip between vehicles without anyone involved slowing down
|
# ? Nov 20, 2017 20:22 |
|
infernal machines posted:
i saw a googcar self-driving around last week in mountain view during the moderate rain. it's not like they gently caress off at the sight of rain like they did 4 years ago https://arstechnica.com/cars/2017/10/5-things-we-learned-from-waymos-big-self-driving-car-report/ e: generally, I believe that there exist impossible limits to computation the impossible limits to computation are not, "can drive a car", otherwise people wouldn't be able to do it more like, "3-sat at alpha=4.8whatever in polytime"
|
# ? Nov 20, 2017 20:23 |
|
oh and im suddenly reminded, with respect to the whole argument about self driving cars improving traffic, that the efficiency of perfectly coordinated traffic vs independent actors is already a topic of study and the improvement is barely enough to matter to everyday people because traffic will still exist, there's just a like a slight improvement to throughput, and they're mad about having to sit in traffic at all
|
# ? Nov 20, 2017 20:27 |
|
bob dobbs is dead posted:i saw a googcar self-driving around last week in mountain view during the moderate rain. it's not like they gently caress off at the sight of rain like they did 4 years ago i don't think anyone here believes that self-driving cars are impossible, just that the timeline for them becoming reality on a commercial scale is more like 30 years rather than the 3 to 5 musk et al seem to think
|
# ? Nov 20, 2017 20:28 |
|
bob dobbs is dead posted:i saw a googcar self-driving around last week in mountain view during the moderate rain. it's not like they gently caress off at the sight of rain like they did 4 years ago lol that is "moderate rain" for mountain view not even umbrella weather for anywhere that gets real rainfall intensities also it's low speeds on millimeter-mapped surface streets and it still entirely relies upon pavement marking detection which water on a roadway fucks up royally, assuming it's properly maintained in the first place the best was the intersection improvements goin on at two san antonio cross streets right near the googcar facility: they were building new signal mast arms so for a couple of months there were two sets of signals up w/ the new ones taped over as you can imagine the self driving cars had a really hard time w/ this - i saw one sitting in a left turn lane and spend the entire protected turn cycle thinking about what it was lookin at while all the cars behind it lost their poo poo
|
# ? Nov 20, 2017 20:28 |
|
infernal machines posted:this is a religious belief at this point. it's a statement of pure faith i'm carrying around a computer in my pocket that is cheaper and more powerful than the one i bought in high school. why are you so sure it's impossible? I'm not saying there won't be a ton of death/failure along the way, but somebody will eventually get it good enough.
|
# ? Nov 20, 2017 20:29 |
|
ate all the Oreos posted:p much. lots of people, even smart people who should know better, think machine learning is some exponential process that one day will just hit the part of the curve that flies off to infinity and suddenly we'll have god-AI. self-awareness immediately precedes self-destruction
|
# ? Nov 20, 2017 20:32 |
|
crabrock posted:i'm carrying around a computer in my pocket that is cheaper and more powerful than the one i bought in high school. why are you so sure it's impossible? I'm not saying there won't be a ton of death/failure along the way, but somebody will eventually get it good enough. it needs to be able to beat humans at the things humans are good at: intuitive understanding of things it has never seen before in less than ideal conditions for its sensors, including missing huge amounts of crucial identifying information if you can do that you can entirely replace humans at many other things than driving
|
# ? Nov 20, 2017 20:32 |
|
prediction: nobody who is currently qualified to drive a car will be relieved of that responsibility by a car-driving computer except perhaps for someone testing a car-driving computer who gets run over and can't drive anymore but htat's not really where i was going with that
|
# ? Nov 20, 2017 20:34 |
|
we don't even automate trains
|
# ? Nov 20, 2017 20:35 |
|
H.P. Hovercraft posted:if you can do that you can entirely replace humans at many other things than driving which, again, is probably not impossible, it's just decades away
|
# ? Nov 20, 2017 20:35 |
|
also since you obliquely brought up moore's law with the whole computer in your pocket thing i might as well mention that it's gonna be real interesting to see what happens when we finally slam up against the physics wall at the end of silicon in a few years
|
# ? Nov 20, 2017 20:37 |
|
ate all the Oreos posted:which, again, is probably not impossible, it's just decades away I don't think its something we will see in our lifetime beyond maybe some novelty expensive poo poo
|
# ? Nov 20, 2017 20:38 |
|
trains also don't kill 35,000 people a year in the US, so there's not a ton of need... but also there are systems in place that automate parts of driving the train
|
# ? Nov 20, 2017 20:38 |
|
H.P. Hovercraft posted:we don't even automate trains lots of automated lines in copenhagen, barcelona, paris, sao paulo metros, where the peeps in the train are for customer service, not doing anything there's also 2 orders of magnitude less cost savings, something like that. how many train drivers do you know? how many car drivers?
|
# ? Nov 20, 2017 20:39 |
|
crabrock posted:trains also don't kill 35,000 people a year in the US, so there's not a ton of need... but also there are systems in place that automate parts of driving the train they probably would [kill tons of people] if they didn't have a ton of safety mechanisms and strict training and the stuff that we very explicitly don't do for cars but could that would be much more immediate and much more effective at saving lives
|
# ? Nov 20, 2017 20:39 |
|
ate all the Oreos posted:also since you obliquely brought up moore's law with the whole computer in your pocket thing i might as well mention that it's gonna be real interesting to see what happens when we finally slam up against the physics wall at the end of silicon in a few years doped diamond transistors most likely. That stuff is still in its infancy but its working really well in high power applications where performance actually improves as internal temps increase
|
# ? Nov 20, 2017 20:40 |
|
crabrock posted:i'm carrying around a computer in my pocket that is cheaper and more powerful than the one i bought in high school. why are you so sure it's impossible? why do you think that's in any way related to whether autonomous vehicles will be practicable as envisioned?
|
# ? Nov 20, 2017 20:41 |
|
we actually pretty much hit end of moore's law for cpu's gpu's still skipping along fine
|
# ? Nov 20, 2017 20:41 |
|
ate all the Oreos posted:which, again, is probably not impossible, it's just decades away it's all popular science zeplins and rocketplanes and gyrojets are the future style bullshit. by the time the technology exists to make it work as envisioned it'll be solving a problem that no longer exists or has changed som much as to make it an impractical solution
|
# ? Nov 20, 2017 20:42 |
|
ate all the Oreos posted:also since you obliquely brought up moore's law with the whole computer in your pocket thing i might as well mention that it's gonna be real interesting to see what happens when we finally slam up against the physics wall at the end of silicon in a few years nah that wasn't the point, more of a "people will find a way to do what seems impossible now, at some point." I mean you're arguing that having a computer drive around streets is like trying to travel faster than the speed of light. It's a task that could be accomplished even with technology that exists right now. It wouldn't be pretty and it's definitely not ready to be used in that way, but if some totalitarian state dictated that only driverless cars were allowed from now on, we'd make it work. also the answer to the moore's law thing is "more cores! totallycountscanthearyoulalala" personal air travel is pretty much always gonna be future poo poo because of all the problems you cite +3d
|
# ? Nov 20, 2017 20:42 |
|
ate all the Oreos posted:which, again, is probably not impossible, it's just decades away it's not even realistically achievable w/ our current understanding of science, which is why google et al attempted to brute force it via 3d mapping you could accurately claim that we're closer to figuring out cold fusion
|
# ? Nov 20, 2017 20:44 |
|
|
# ? Jun 9, 2024 19:20 |
|
crabrock posted:nah that wasn't the point, more of a "people will find a way to do what seems impossible now, at some point." as i said earlier nobody here thinks it can't be done, just that the technology for it to be done well (let alone on a large cost-effective scale) won't be available for decades at best.
|
# ? Nov 20, 2017 20:45 |