Open source software and dataset to verify its effectiveness but restrict it for commercial purpose would help build confidence in auto pilot.
Apple engineer killed in Tesla crash had previously complained about autopilot
231–240 of 919 posts
Re: Apple engineer killed in Tesla crash had previously complained about autopilot
#232Tesla has some blame in this for sure but Caltrans (maintainers of the 101) are equally if not more to blame. California is rich in capital but so poor infrastructure. Moving here from South Carolina(where taxes are much lower and roads are paltry) I was shocked at how dismal a state many of the roads are and at the sheer amount of time it takes to complete. Patrick Collison of Stripe actually wrote a blog piece(1) c…
Many US citizens I talk to, seem to have the stance that they do not want to pay any taxes and that taxation is theft.
I am sure California and others could do more and be held accountable (every service provider should and must be), but that attitude has to change too.
Re: Apple engineer killed in Tesla crash had previously complained about autopilot
#233I avoid new technology, exactly because I'm an engineer. I wonder if that is just me, but when new technology is introduced and hyped, I usually take a quick look at implementations, research, and talks just to get an idea of what the state of the art really is like. As a result, I have become the late adopter among my group of friends because the first iteration of any new technology usually just isn't worth the iss…
I don't know where you got these papers but you clearly got an awful sampling and I don't think you're giving computer vision a fair evaluation. Granted, we're not quite ready for self driving, but there's no question that the neural network subfield of ML has absolutely exploded in the last 5-10 years and is bursting with productionizable, application ready architectures that are already solving real world problems.
You sound like an NVIDIA salesperson trying to sell me on a $3000 Titan ;)
There is no doubt in my mind that an AI with billions of parameters will be excellent at memorizing stuff.
I also have no doubt that research activity has exploded, which might be related to the generous hardware grants being handed out...
But all that research has produced surprisingly little progress over algorithms from 2004 in the field of optical flow.
The papers I looked at were the top-ranked optical flow algorithms on Sintel and KITTY. So those were the AIs that work best in practice and better than 99% of the other AI solutions.
Re: Apple engineer killed in Tesla crash had previously complained about autopilot
#234There are going to be Autopilot crashes and people should disable it if they uncomfortable with putting their lives in the hands of Tesla's software. I would be. But I see no reason to not believe to not believe what they have on https://www.tesla.com/VehicleSafetyReport It shows Teslas have much fewer accidents than the average car, and that Teslas with autopilot enabled are substantially safe per mile than with aut…
Teslas are more expensive than the average car (hence might be driven more carefully), and autopilot is more likely to be enabled in predictable driving environments like highways.
hence might be driven more carefully
I don't know where you get that idea from, when I'm out cycling it's generally the people in expensive cars that don't pay much attention.
Re: Apple engineer killed in Tesla crash had previously complained about autopilot
#235I avoid new technology, exactly because I'm an engineer. I wonder if that is just me, but when new technology is introduced and hyped, I usually take a quick look at implementations, research, and talks just to get an idea of what the state of the art really is like. As a result, I have become the late adopter among my group of friends because the first iteration of any new technology usually just isn't worth the iss…
It’s been a while for me, but I’m pretty sure there are techniques for approximating gradients for step functions, although the paper may not have mentioned them (perhaps for secret sauce)
Re: Apple engineer killed in Tesla crash had previously complained about autopilot
#236I avoid new technology, exactly because I'm an engineer. I wonder if that is just me, but when new technology is introduced and hyped, I usually take a quick look at implementations, research, and talks just to get an idea of what the state of the art really is like. As a result, I have become the late adopter among my group of friends because the first iteration of any new technology usually just isn't worth the iss…
Yes. Current self-driving might be safer than the avarage driver, but what most people forget is how unsafe it is when it fails. It fails harder than most avarage drivers and even bad drivers.
In my eyes self-driving should still be called driving-assistance.
Re: Apple engineer killed in Tesla crash had previously complained about autopilot
#237I avoid new technology, exactly because I'm an engineer. I wonder if that is just me, but when new technology is introduced and hyped, I usually take a quick look at implementations, research, and talks just to get an idea of what the state of the art really is like. As a result, I have become the late adopter among my group of friends because the first iteration of any new technology usually just isn't worth the iss…
I don't know where you got these papers but you clearly got an awful sampling and I don't think you're giving computer vision a fair evaluation. Granted, we're not quite ready for self driving, but there's no question that the neural network subfield of ML has absolutely exploded in the last 5-10 years and is bursting with productionizable, application ready architectures that are already solving real world problems.
> Granted, we're not quite ready for self driving
And yet it's literally in cars on the road.
I'm not saying you're wrong because of that. I just wonder how far from "ready" we are, and how much of a gamble manufacturers are taking, and how much risk that presents for not just their customers, but everyone else their customers may drive near.
Re: Apple engineer killed in Tesla crash had previously complained about autopilot
#238Tesla says autopilot works but that driver should be ready to intervene. This is terrifying - like having a kid riding shotgun that might just reach out and turn the wheel while you are on the highway. Anyone who has driven with autopilot, how quickly might it react to a perceived obstacle? Would it take a hard turn into a median faster than a person could reasonably react if it thought there was something in the roa…
I use autopilot daily. It drives me, on a twisty road, from my house to work with intervention when I get into town. > how quickly might it react to a perceived obstacle? I've had it head towards a curbed median in the middle of the road and within 10 feet swerve to the right to miss it (in a spot that I've driven by hundreds of times, but this time it kinda decided to head towards the curbed median). Auto pilot can…
Thanks! I haven't read many firsthand accounts.
Re: Apple engineer killed in Tesla crash had previously complained about autopilot
#239I avoid new technology, exactly because I'm an engineer. I wonder if that is just me, but when new technology is introduced and hyped, I usually take a quick look at implementations, research, and talks just to get an idea of what the state of the art really is like. As a result, I have become the late adopter among my group of friends because the first iteration of any new technology usually just isn't worth the iss…
Yes. Current self-driving might be safer than the avarage driver, but what most people forget is how unsafe it is when it fails. It fails harder than most avarage drivers and even bad drivers. In my eyes self-driving should still be called driving-assistance.
Can you cite any sources that AI fails harder? I'm not exactly sure what you are referring to.
Re: Apple engineer killed in Tesla crash had previously complained about autopilot
#240I avoid new technology, exactly because I'm an engineer. I wonder if that is just me, but when new technology is introduced and hyped, I usually take a quick look at implementations, research, and talks just to get an idea of what the state of the art really is like. As a result, I have become the late adopter among my group of friends because the first iteration of any new technology usually just isn't worth the iss…
It’s been a while for me, but I’m pretty sure there are techniques for approximating gradients for step functions, although the paper may not have mentioned them (perhaps for secret sauce)
In the case I evaluated, the authors later admitted to using supervised training for initializing their supposedly unsupervised learning methods.
Also, I was reviewing their public source code release and there was no approximation. That part of their loss function had simply never worked, but they had not noticed prior to the release of their paper.
And due to them training slightly different from what they described in the paper, the AI worked competitively well nonetheless.