The Uber system detected miscategorized the cyclist as roadside debris-- like a plastic bag-- after categorizing objects correctly the vast majority of the time.
Look, Uber clearly had all kinds of software quality and process problems. But I guarantee you even the leading implementations like Waymo are learning a whole shit-ton from real road stuff-- exactly about this kind of misclassification problem. Synthetic test conditions reflect experimenters' bias.
Tesla has a bit of an advantage with all the video they're sampling from customers' driving habits, but even so the selection of that video and annotation and review for potential problems doesn't have the strength of real-world testing because it's not a priori (researchers' own biases, classification issues, and selection issues can come into play) and not as rigorous.
How do you propose we know when a technology is "safe enough" to be tested on a public road, since we can expect that public road testing will reveal new defects and weaknesses?
I'm not quibbling that Uber really fucked up. I just think that it's reasonable to think that we can't get the technologies to be perfectly safe relying on recorded data of other driving and test-track scenarios.
> Hold up. While it's tempting to think this case is the bar we're setting. Consider the released footage of the driver for a second. [theguardian.com] Imagine any industry where you look away from a machine in motion for that long, yeah, you're at fault for your reckless behavior. Like if I was watching a show on my phone while operating a table saw, yeah, I really wouldn't have a strong case for an injury lawsuit.
You have a point, but conversely: people are really, really, really bad at vigilance tasks, where they are supposed to monitor something that almost always goes right, look for something that almost always doesn't appear, etc. Asking someone to be a vigilant backup driver for 4 hours may be a lot harder than driving for 4 hours.
Almost all of the Canadian population is in a narrow southwards populated band, though, with a much higher density. That is, Canada has a bunch of land with basically no people in it.
> The cables and controllers sure aren't compatible, you need a converter from USB-x to USB-2 which typically includes a full controller on-chip.
USB3 type A has a few extra contacts, but you can plug USB1.0 type A cables into it and use USB1.0 devices, or anything inbetween...
> Even USB-3 to USB-2, the controllers aren't backwards compatible and thus older software that talks to OHCI/EHCI won't talk to xHCI or beyond.
Yah, you need an appropriate driver and OS that knows how to talk to xHCI, but actual end-user software that just wants to do USB2 things it looks identical.
> My solution? Call different technologies by different names. Make a different connector for different electrical protocols.
How do I benefit from this? I have a limited footprint for connectors in my laptop. Better to have electrical/pinout magic to make ports work with "slow" (but still pretty capable) devices, and multiplexer magic to allow multiple kinds of high speed signalling over the remaining pins.
These ports with multiple functions aren't too expensive, and I am usually better off with 4 "USB" ports than 3 ports good for keyboard, mice, and slow thumb drives; 2 ports good for fast SSD or PCIe expansion, 2 ports good for external monitors, etc-- both in cost, physical footprint, and ease of use.
The power argument isn't really super true anymore-- Atom got pretty good, etc. But ARM already had a foothold, and... perhaps most important, it's not too onerous to be an ARM licensee. So there's a lot of SOCs now featuring ARM and with a superior degree of integration and nice things to design a phone with.
And then the same network effects hold true-- if 99% of users are Android-on-ARM then if you use Android-on-X86 you're going to have weird experiences at time.
NY was anteing a few billion in the short term, and hoping to get $30B in tax revenue in the long term. The models for "future revenue based on project" tend to be complete bollocks-- assuming that everyone who works for Amazon is new to the area and becomes a new income tax filer that would be paying $0 otherwise, etc.
> 25,000 new jobs for a 10% discount on taxes, seems like an OK deal to me
Possibly 15,000-25,000 new jobs vs. $3bln in subsidies-- $120k/job. Also, time value of money, yada yada yada.
Oh, come on. The personnel may be uneven, but nation states both have very, very nasty toolkits to assist spread, concealment, and information extraction... and they buy zero-days.
An initial foothold on the network can only be prevented by A) inability to be exploited by undisclosed exploits and B) perfect end-user practices to not inadvertently cooperate with attackers. Otherwise, we're in the scenario described above.
Like the $1.2B in refundable tax credits that Amazon demanded to move to NY, and $500M in capital grants to build facilities with. It takes a long, long time of a couple percent of property tax to pay these things back, let alone pay for the services they're consuming in the meantime.
> And as far as I know, they don't pull bags for passengers that miss a connecting flight. I've missed a connecting flight before due to a late aircraft and a long walk + train ride to the gate and still had my bags go on without me.
They are supposed to. Otherwise it's a great way to blow up a plane without being on it.
(note, I'm aware of the Essential Air Service subsidies at Reno, but they are for specific low volume routes, not something someone would be likely to get on from SkipLagged. and they ain't $300!)
On through flights, they count the passengers remaining on the plane and adjust the manifest if someone is missing. Usually counting them all multiple times.
I'd guess that it's more like 10 years from the start of a tipping point. By then about 20% of the cars will be electric, at that point gas will start to get cheap (with the removal of 20% of the demand), but banks and investors will stop financing oil projects (which require lots of money just to keep going). In about 20 years gasoline would be very expensive and pull the rest of the car market into full electric.
It's more complicated than this. Only about half of oil goes to gasoline. So even if gasoline demand falls significantly, oil itself still won't be a specialty product.
I'm not exactly a fan of Amazon, but it's rational for them to dedicate resources to the communities where they will have a significant presence. If they don't go to New York, and go somewhere else instead, then resources they were going to spend on the community in New York will instead go somewhere else.
Are you troll-y, or just have very fluid opinions since you're bipolar?
>I like my AMD more than my GTX 680 I replaced. The drivers are much better quality and I have much less issues and no latency lag or Windows 10 bugs related to the Nvidia drivers.
> I think the tide has turned and the shitty AMD catalyst drivers were retired years ago. They are totally redone.
It is power inefficient, but it is hardly slower than a 1080ti. What, did you find a single benchmark somewhere that said that? What about the GPGPU benchmarks where it's blowing the doors off the 2080, or the others where it's neck and neck?
No, but murder victim's sketchy ex-boyfriend who did DV once to someone else is not enough to get a search warrant or require him to provide DNA. And there might be 10 people in the victim's past and relations that have some kind of reason for minor suspicion. Too many dead ends to investigate in depth, and not enough cause to forcibly collect DNA samples from them.
Then you have a familial DNA match-- you find the person who is closest related to the actual criminal, and then there's 1000 people in the world who are the right relation distance away. Of whom 12 live in the surrounding areas. And the intersection of these 12 and the other 10 people yields one person. Suddenly it's worth sneaking around to collect that dude's DNA or getting a court order.
You're right. And this is a key reason why minorities and the lowest 25% of incomes are dying more of cancer: incidence in the poor of cancer is much higher.
Yes, cancer detection and treatment is better as income increases, but this effect is smaller than you might think. The poor are exposed to so many cancer risk factors, from smoking, ambient air exposure, excessively processed foods, etc, so the baseline risk is significantly higher even before you get to any difference in treatment.
That sounds like a lot of money, huh? But A) it wasn't all cash, and B) a leading-edge process node for processors costs $2.5B or more. Design costs for an microprocessor are a couple billion.
Slashdot Mirror
Yah, what a ridiculous idea he offers. If I can call on the phone, watch TV, and surf on the web, I *totally* want to go to prison.
The Uber system detected miscategorized the cyclist as roadside debris-- like a plastic bag-- after categorizing objects correctly the vast majority of the time.
Look, Uber clearly had all kinds of software quality and process problems. But I guarantee you even the leading implementations like Waymo are learning a whole shit-ton from real road stuff-- exactly about this kind of misclassification problem. Synthetic test conditions reflect experimenters' bias.
Tesla has a bit of an advantage with all the video they're sampling from customers' driving habits, but even so the selection of that video and annotation and review for potential problems doesn't have the strength of real-world testing because it's not a priori (researchers' own biases, classification issues, and selection issues can come into play) and not as rigorous.
How do you propose we know when a technology is "safe enough" to be tested on a public road, since we can expect that public road testing will reveal new defects and weaknesses?
I'm not quibbling that Uber really fucked up. I just think that it's reasonable to think that we can't get the technologies to be perfectly safe relying on recorded data of other driving and test-track scenarios.
> Hold up. While it's tempting to think this case is the bar we're setting. Consider the released footage of the driver for a second. [theguardian.com] Imagine any industry where you look away from a machine in motion for that long, yeah, you're at fault for your reckless behavior. Like if I was watching a show on my phone while operating a table saw, yeah, I really wouldn't have a strong case for an injury lawsuit.
You have a point, but conversely: people are really, really, really bad at vigilance tasks, where they are supposed to monitor something that almost always goes right, look for something that almost always doesn't appear, etc. Asking someone to be a vigilant backup driver for 4 hours may be a lot harder than driving for 4 hours.
Almost all of the Canadian population is in a narrow southwards populated band, though, with a much higher density. That is, Canada has a bunch of land with basically no people in it.
???
> The cables and controllers sure aren't compatible, you need a converter from USB-x to USB-2 which typically includes a full controller on-chip.
USB3 type A has a few extra contacts, but you can plug USB1.0 type A cables into it and use USB1.0 devices, or anything inbetween...
> Even USB-3 to USB-2, the controllers aren't backwards compatible and thus older software that talks to OHCI/EHCI won't talk to xHCI or beyond.
Yah, you need an appropriate driver and OS that knows how to talk to xHCI, but actual end-user software that just wants to do USB2 things it looks identical.
> My solution? Call different technologies by different names. Make a different connector for different electrical protocols.
How do I benefit from this? I have a limited footprint for connectors in my laptop. Better to have electrical/pinout magic to make ports work with "slow" (but still pretty capable) devices, and multiplexer magic to allow multiple kinds of high speed signalling over the remaining pins.
These ports with multiple functions aren't too expensive, and I am usually better off with 4 "USB" ports than 3 ports good for keyboard, mice, and slow thumb drives; 2 ports good for fast SSD or PCIe expansion, 2 ports good for external monitors, etc-- both in cost, physical footprint, and ease of use.
?? in what jurisdiction? Pretty much everywhere I've been, when you return stuff, you get the sales tax back too.
Indeed, on the California Sales and Use Tax return, Section B, item 3--- returned taxable merchandise is a deduction.
The "front brake" is almost certainly regenerative braking on the motor, while the rear brake is a mechanical brake.
There is really relatively few choices besides the front motor effort (including regenerative effort) to be under software/firmware control.
The power argument isn't really super true anymore-- Atom got pretty good, etc. But ARM already had a foothold, and... perhaps most important, it's not too onerous to be an ARM licensee. So there's a lot of SOCs now featuring ARM and with a superior degree of integration and nice things to design a phone with.
And then the same network effects hold true-- if 99% of users are Android-on-ARM then if you use Android-on-X86 you're going to have weird experiences at time.
NY was anteing a few billion in the short term, and hoping to get $30B in tax revenue in the long term. The models for "future revenue based on project" tend to be complete bollocks-- assuming that everyone who works for Amazon is new to the area and becomes a new income tax filer that would be paying $0 otherwise, etc.
> 25,000 new jobs for a 10% discount on taxes, seems like an OK deal to me
Possibly 15,000-25,000 new jobs vs. $3bln in subsidies-- $120k/job. Also, time value of money, yada yada yada.
Oh, come on. The personnel may be uneven, but nation states both have very, very nasty toolkits to assist spread, concealment, and information extraction... and they buy zero-days.
An initial foothold on the network can only be prevented by A) inability to be exploited by undisclosed exploits and B) perfect end-user practices to not inadvertently cooperate with attackers. Otherwise, we're in the scenario described above.
> Most of the rest goes for government giveaways.
Like the $1.2B in refundable tax credits that Amazon demanded to move to NY, and $500M in capital grants to build facilities with. It takes a long, long time of a couple percent of property tax to pay these things back, let alone pay for the services they're consuming in the meantime.
> And as far as I know, they don't pull bags for passengers that miss a connecting flight. I've missed a connecting flight before due to a late aircraft and a long walk + train ride to the gate and still had my bags go on without me.
They are supposed to. Otherwise it's a great way to blow up a plane without being on it.
(note, I'm aware of the Essential Air Service subsidies at Reno, but they are for specific low volume routes, not something someone would be likely to get on from SkipLagged. and they ain't $300!)
> The real problem is when an airport like (say) Reno is paying $300 for every ticket sold to the airline
Citation needed.
On through flights, they count the passengers remaining on the plane and adjust the manifest if someone is missing. Usually counting them all multiple times.
I'd guess that it's more like 10 years from the start of a tipping point. By then about 20% of the cars will be electric, at that point gas will start to get cheap (with the removal of 20% of the demand), but banks and investors will stop financing oil projects (which require lots of money just to keep going). In about 20 years gasoline would be very expensive and pull the rest of the car market into full electric.
It's more complicated than this. Only about half of oil goes to gasoline. So even if gasoline demand falls significantly, oil itself still won't be a specialty product.
I'm not exactly a fan of Amazon, but it's rational for them to dedicate resources to the communities where they will have a significant presence. If they don't go to New York, and go somewhere else instead, then resources they were going to spend on the community in New York will instead go somewhere else.
One more comment:
> and has inferior AMD drivers.
https://slashdot.org/comments....
Are you troll-y, or just have very fluid opinions since you're bipolar?
>I like my AMD more than my GTX 680 I replaced. The drivers are much better quality and I have much less issues and no latency lag or Windows 10 bugs related to the Nvidia drivers.
> I think the tide has turned and the shitty AMD catalyst drivers were retired years ago. They are totally redone.
Big swing of opinion in 2 months.
Do you shill^H^H^H^H^H work at Nvidia?
It is power inefficient, but it is hardly slower than a 1080ti. What, did you find a single benchmark somewhere that said that? What about the GPGPU benchmarks where it's blowing the doors off the 2080, or the others where it's neck and neck?
> AMD will be out of business or the graphics market soon anyway
How's that work with $356M in net income over the past years and steadily improving financials?
No, but murder victim's sketchy ex-boyfriend who did DV once to someone else is not enough to get a search warrant or require him to provide DNA. And there might be 10 people in the victim's past and relations that have some kind of reason for minor suspicion. Too many dead ends to investigate in depth, and not enough cause to forcibly collect DNA samples from them.
Then you have a familial DNA match-- you find the person who is closest related to the actual criminal, and then there's 1000 people in the world who are the right relation distance away. Of whom 12 live in the surrounding areas. And the intersection of these 12 and the other 10 people yields one person. Suddenly it's worth sneaking around to collect that dude's DNA or getting a court order.
You're right. And this is a key reason why minorities and the lowest 25% of incomes are dying more of cancer: incidence in the poor of cancer is much higher.
Yes, cancer detection and treatment is better as income increases, but this effect is smaller than you might think. The poor are exposed to so many cancer risk factors, from smoking, ambient air exposure, excessively processed foods, etc, so the baseline risk is significantly higher even before you get to any difference in treatment.
That sounds like a lot of money, huh? But A) it wasn't all cash, and B) a leading-edge process node for processors costs $2.5B or more. Design costs for an microprocessor are a couple billion.