Fuck AT&T. I don't tether currently. I didn't cringe when I got charged $26 per line for "activation". I didn't cringe at signing a 2-year contract to get a phone for $300. I didn't even cringe at an "unlimited" data plan that limits downloads to 10MB files (which, coincidentally, is smaller than most of the apps on the "approved" app store).
Why is Apple sticking with these people. The overall user experience of an "approved" iPhone is significantly worse because of AT&T's behavior as greedy little fucktards.
Hrm? Even with 3G, I have an unlimited data plan for $40/month. Granted there are all sorts of "gotchas" but for the most part, it's fast (in the city I live, it's ~3mbit), stable and I don't have to hop from network to network when driving.
If 4G is an improvement -- and the cell providers decide to stop being greedy little fucksticks -- it could easily kill city-wide WiFi.
Of course, if the cell providers remain greedy little fucksticks, there could still be a demand for it.
I would imagine (or hope) that the cache is write-through. But yes, even with flash, there is a chance that the latest file written hasn't finished writing to disk before the machine crashes.
While many are arguing against this from the point of view of economic and social success, I really question whether this is at all healthy for the development of a human being. I mean, to never be told "no". To never be made to do something you don't want to? That sounds like it'll raise one hell of a whiny, never-satisfied child.
Then again, I'm typing this from the work even though I'd rather be at home with a beer in my hand....
Your OS doesn't always have time to shut down properly. Don't think anyone's fond of the idea of having their last couple of saves go poof because Windows crashed.
Intel's SSD drives already have 32MB of DRAM. But it's not used to buffer data because of reliability issues.
That seemed odd as well. But it does follow the current trend. Path delays aren't decreasing as much as feature sizes shrink. Gates at 45nm aren't *that* much faster than they were at 90nm if you designed both for speed. They are, however, a lot less power-consuming.
Tunneling isn't completely random. Hell, without it, modern flash memory wouldn't be possible. The trick is to stop thinking in terms of absolutes. Modern EDA tools at 45nm and below already treat gate-delay as a probability function rather than absolute min/max. A chain of gates would produce a probability distribution and you'd simply design it to be ~98% inside your timing envelope.
Except that wasn't the point. Star Trek wasn't about inter-species conflict. The Klingons weren't just ridged headed aliens (and originally, they weren't). Star Trek was political allegory. The Klingons were the Soviets; the Federation was the U.S. The point of the whole "we won't fight directly but we'll both bully smaller planets to join our side to fight against their side" was the common theme. The result was that a higher being (the Organians) came in, bitchslapped their stupid asses and said "behave".
Almost every story and every alien world (save for filler episodes) were an allegory for modern-day problems. Everything form how we treat veterans to racism to ruthless imperialism (Cardassian occupation of Bajor) and the moral ambiguities of those situations.
They are extremely good given the limitations of biology, but not nearly as good as man-made sensors like CCDs
Not really. The dynamic range of the best CCD's out there are around 13.5 stops. The human eye (on average) can see a bout 20 stops at any given time. The eye also has a natural non-linear saturation filter. Some CCD's nowadays have been made with resistor networks between pixel clusters to try to mimic this but the added noise produces something nowhere near as good as human visual perception.
Your point about gesture recognition underscores what I'm saying. You shouldn't need to have any training period at all in a gesture recognition system: It should work the first time.
My point was that a computer has to do this out of the box with very little information. Take your average human. If he were unfamiliar with the gestures of a person or even of what a mouse pointer was, it'd take him weeks if not months to learn to recognize different gestures. This is why AI algorithms aren't modeled after humans; they have to work under different circumstances.
Even when they're tuned, they still make errors a 3 year old would never make. The approach I'm advocating, growing from a seed in connection with a rich environment, is something that happens before deployment time so to speak.
It would take 3 years or more to "grow" that AI assuming you could provide it with the stimulus that teaches it constantly like an infant has access to. It'd also be a hell of a lot of information to package with the software.
How much training time do you need to understand the speech of someone you just met?
It took me about 10 years before I could understand about 90% of most people's speech. That's not even taking into account dialect and accents.
What Moore's Law will not solve is the question of how do we properly build these systems. My point is that real intelligence needs to be grown, not designed.
That approach may be preferable one day but not today. The data contained by a human capable of piloting a drone or understand language is vast. It would take Moore's law quite some time before the amount of data you can ship with typical AI applications is sufficient and even then, it will only work with the type of stuff you can train it for in the lab.
Find me an algorithm that can design an iPhone or prove Poincare's conjecture, and I might change my mind.
Most of the iPhone was probably designed by an algorithm. I very much doubt Samsung had their engineers hand-wire the CPU or map out the logic. Logic synthesis, behavorial synthesis, automated place-and-route and physical design tools are integral to designing all electronic devices nowadays.
The PCB layout was probably semi-automated with some engineer creating bus and wire guides and an auto-route software.
Design-automation tools are getting more and more powerful by the day. This year's DAC demonstrated quite a few large advances in C-to-hardware synthesis.
That's because most AI research isn't aimed at creating lovable but emotionally troubled human-like intelligence. Most of the commercial applications for AI involve much different circumstances than humans. The gesture recognition in Firefox or an iPhone, for example, simply doesn't have years to learn a particular person's pointer movements. It has to work practically out-of-the-box with maybe a small (few hours worth of use) training period.
The thing about reproducing the type of functions the human brain can do is that it involves a huge amount of information. Everything that person has witnessed using their sense since birth. That is a lot of information due to just how sensitive human senses are. The human eye's visual perception is orders of magnitude better than any CMOS image sensor and provides far more robust information. The human ear is far more sensitive than any microphone.
Most AI applications simply don't have the time or ability to transfer all of the "life experience" information. So the goal is to make algorithms that can use very little information (e.g. a single image) and make decisions based on that. Take a combat drone with an auto-pilot system. Real human pilots have decades worth of information in him/her about what different objects look like. You can't store all of that inside a small drone.
No. A memsistor is mathematically a function of charge and how much has flowed through the device in the pass. It varies its resistance depending on how much charge has flowed through previously.
It's the reconfigurable nature of the human brain that's unique and powerful. If all you did was take one person, listed all of the skills of that person -- all of the things he knew; all of the skills in smell, touch, sight and taste; all of the cognitive reasoning ability -- then you could create a chip to simulate those skills. Algorithms for image recognition, feature extraction, speech recognition, etc. are all available that are very very close to what humans can do.
But the thing that separates humans is that it didn't take hundreds of years of mathematical development to come up with these algorithms. The human brain develops these algorithms through changes in its structure from birth. At about age 10, speech recognition specialized and tailored to the dialect, language and tones that the person hears has developed on its own.
That type of structural formation and learning is what would need to happen in silicon to make a truly intelligent machine. Neuron clusters emulated using transistors would need to be able to dynamically form connections to other neuron clusters. There'd have to be some type of distributed learning algorithm encoded in the operation of each individual neuron.
Speech recognition is easy. Image recognition is easy. Developing a distributed, scalable, self-modifying architecture that can learn all of those and more on its own with nothing more than training samples is the difficult part.
I don't think that's entirely true. While I'm sure there is a stigma against non-profits, teaching is generally considered a social nobility. The adage "those who can't teach" was more a commentary on the poor quality and low qualification-for-entry of teachers in the current U.S. education system -- a criticism that is for the most part true.
Anecdotal but to the point: during college, the College of Education was a laughing joke amongst the other departments. It was literally a free "A" for anyone who took a College of Education course. One of our math professors taught a class for the CoE and tried to actually teach them math -- that is, make them understand the concept instead of just letting them coast and handing out a free "A". The class went ape-shit; they couldn't believe they had to actually work for their grade.
A degree from my university in Education was considered fairly prestigious and I knew many who got their Master's and went on to teach in very high-income public school areas. Not one of them -- despite being calculus teachers -- can actually explain to you the concept of a limit beyond the rules in the textbook.
So before you blame the industry -- and I'm not saying the CEO-worshipping trogs are innocent in all of this -- perhaps a look at government and how (poorly) it compensates teachers -- and therefore attract low-quality applicants.
There is also risk that a male will get a better job... so how is that any different?
You can't offer a pregnant mother-to-be a salary increase to not have her baby. You also don't have to re-hire the guy if his other job goes south. Once he leaves the company, that's it.
I don't disagree with maternity leave but either:
1. It should be a perk -- something you sign up for like premium health insurance -- upon hire in return for a lower salary. 2. Something mandated by the government but there be a provision of allowing lower salaries for those who qualify.
Not really. The average marrying age for girls used to be 13-20. If you made it past then, you were either on your second marriage already or were desperately trying to find a husband.
Today's standards of beauty follow the same premise. The difference is, women are getting married later in life -- due to how much preparation is given before they are considered adults -- and thus, desire to be unnaturally skinny in order to resemble young girls in the 13-20 range.
Biologically, this makes sense. Human males are attracted to just post-pubescent girls as that is their most fertile time. I'd say that it isn't so much the standards of beauty that are changing, but rather social views of when a woman can be viewed as sexually attractive.
If you mention to someone in the 1920's that one can get arrested for photoshoping a pic of Miley Cyrus (16 years old), they'd look at you funny.
They used old photographs in the study of people from past generations and their method of "objective" measurement of beauty were to have modern-day people judge them.
It seems almost a foregone conclusion that people in modern times would find the women of modern times more attractive; standards of beauty change.
Nobody ever learned to program a GPU because there is a very very nice standardized -- two of them actually -- API abstraction layer; DirectX and OpenGL. Common graphics functions that are coupled with hardware by the library.
IMO, this is exactly what IBM/Sony should be doing with the Cell -- create a standard library as well as have a staff dedicated to consulting/developing low-level functions and farm those guys out to game developers.
While true, the vast majority of the reason developers say the 360's PPC is "faster" is that it's easier to program. In terms of theoretical flops, the CELL inside the PS3 is far more capable.
If you just strung together a bunch of code and hit "compile" on GCC, the 360 will be lightyears ahead. If you have 2-3 really good engineers writing sequences for the CELL, the vertex computational power will be far superior.
No. In fact the biggest improvement of this car appears to be the nanophosphate battery. It doesn't use the chemicals inside traditional li-ions that become heated when overcharged (lithium particles start leaking across to the anode).
Slashdot Mirror
Fuck AT&T. I don't tether currently. I didn't cringe when I got charged $26 per line for "activation". I didn't cringe at signing a 2-year contract to get a phone for $300. I didn't even cringe at an "unlimited" data plan that limits downloads to 10MB files (which, coincidentally, is smaller than most of the apps on the "approved" app store).
Why is Apple sticking with these people. The overall user experience of an "approved" iPhone is significantly worse because of AT&T's behavior as greedy little fucktards.
Hrm? Even with 3G, I have an unlimited data plan for $40/month. Granted there are all sorts of "gotchas" but for the most part, it's fast (in the city I live, it's ~3mbit), stable and I don't have to hop from network to network when driving.
If 4G is an improvement -- and the cell providers decide to stop being greedy little fucksticks -- it could easily kill city-wide WiFi.
Of course, if the cell providers remain greedy little fucksticks, there could still be a demand for it.
I would imagine (or hope) that the cache is write-through. But yes, even with flash, there is a chance that the latest file written hasn't finished writing to disk before the machine crashes.
And they'll be fuck his wife while he's at work hating every second of it.
While many are arguing against this from the point of view of economic and social success, I really question whether this is at all healthy for the development of a human being. I mean, to never be told "no". To never be made to do something you don't want to? That sounds like it'll raise one hell of a whiny, never-satisfied child.
Then again, I'm typing this from the work even though I'd rather be at home with a beer in my hand....
Your OS doesn't always have time to shut down properly. Don't think anyone's fond of the idea of having their last couple of saves go poof because Windows crashed.
Intel's SSD drives already have 32MB of DRAM. But it's not used to buffer data because of reliability issues.
Nobel prize, here I come.
That seemed odd as well. But it does follow the current trend. Path delays aren't decreasing as much as feature sizes shrink. Gates at 45nm aren't *that* much faster than they were at 90nm if you designed both for speed. They are, however, a lot less power-consuming.
Tunneling isn't completely random. Hell, without it, modern flash memory wouldn't be possible. The trick is to stop thinking in terms of absolutes. Modern EDA tools at 45nm and below already treat gate-delay as a probability function rather than absolute min/max. A chain of gates would produce a probability distribution and you'd simply design it to be ~98% inside your timing envelope.
Really? You thought "Cylons started with an emo teen" was a good plot?
Except that wasn't the point. Star Trek wasn't about inter-species conflict. The Klingons weren't just ridged headed aliens (and originally, they weren't). Star Trek was political allegory. The Klingons were the Soviets; the Federation was the U.S. The point of the whole "we won't fight directly but we'll both bully smaller planets to join our side to fight against their side" was the common theme. The result was that a higher being (the Organians) came in, bitchslapped their stupid asses and said "behave".
Almost every story and every alien world (save for filler episodes) were an allegory for modern-day problems. Everything form how we treat veterans to racism to ruthless imperialism (Cardassian occupation of Bajor) and the moral ambiguities of those situations.
They are extremely good given the limitations of biology, but not nearly as good as man-made sensors like CCDs
Not really. The dynamic range of the best CCD's out there are around 13.5 stops. The human eye (on average) can see a bout 20 stops at any given time. The eye also has a natural non-linear saturation filter. Some CCD's nowadays have been made with resistor networks between pixel clusters to try to mimic this but the added noise produces something nowhere near as good as human visual perception.
Your point about gesture recognition underscores what I'm saying. You shouldn't need to have any training period at all in a gesture recognition system: It should work the first time.
My point was that a computer has to do this out of the box with very little information. Take your average human. If he were unfamiliar with the gestures of a person or even of what a mouse pointer was, it'd take him weeks if not months to learn to recognize different gestures. This is why AI algorithms aren't modeled after humans; they have to work under different circumstances.
Even when they're tuned, they still make errors a 3 year old would never make. The approach I'm advocating, growing from a seed in connection with a rich environment, is something that happens before deployment time so to speak.
It would take 3 years or more to "grow" that AI assuming you could provide it with the stimulus that teaches it constantly like an infant has access to. It'd also be a hell of a lot of information to package with the software.
How much training time do you need to understand the speech of someone you just met?
It took me about 10 years before I could understand about 90% of most people's speech. That's not even taking into account dialect and accents.
What Moore's Law will not solve is the question of how do we properly build these systems. My point is that real intelligence needs to be grown, not designed.
That approach may be preferable one day but not today. The data contained by a human capable of piloting a drone or understand language is vast. It would take Moore's law quite some time before the amount of data you can ship with typical AI applications is sufficient and even then, it will only work with the type of stuff you can train it for in the lab.
Find me an algorithm that can design an iPhone or prove Poincare's conjecture, and I might change my mind.
Most of the iPhone was probably designed by an algorithm. I very much doubt Samsung had their engineers hand-wire the CPU or map out the logic. Logic synthesis, behavorial synthesis, automated place-and-route and physical design tools are integral to designing all electronic devices nowadays.
The PCB layout was probably semi-automated with some engineer creating bus and wire guides and an auto-route software.
Design-automation tools are getting more and more powerful by the day. This year's DAC demonstrated quite a few large advances in C-to-hardware synthesis.
That's because most AI research isn't aimed at creating lovable but emotionally troubled human-like intelligence. Most of the commercial applications for AI involve much different circumstances than humans. The gesture recognition in Firefox or an iPhone, for example, simply doesn't have years to learn a particular person's pointer movements. It has to work practically out-of-the-box with maybe a small (few hours worth of use) training period.
The thing about reproducing the type of functions the human brain can do is that it involves a huge amount of information. Everything that person has witnessed using their sense since birth. That is a lot of information due to just how sensitive human senses are. The human eye's visual perception is orders of magnitude better than any CMOS image sensor and provides far more robust information. The human ear is far more sensitive than any microphone.
Most AI applications simply don't have the time or ability to transfer all of the "life experience" information. So the goal is to make algorithms that can use very little information (e.g. a single image) and make decisions based on that. Take a combat drone with an auto-pilot system. Real human pilots have decades worth of information in him/her about what different objects look like. You can't store all of that inside a small drone.
No. A memsistor is mathematically a function of charge and how much has flowed through the device in the pass. It varies its resistance depending on how much charge has flowed through previously.
Memsistors are nothing like neurons....
Neurons are incredibly complex nodes with a built-in structural formation algorithm; an algorithm that's not understood at all.
Memsistors store current values.
It's the reconfigurable nature of the human brain that's unique and powerful. If all you did was take one person, listed all of the skills of that person -- all of the things he knew; all of the skills in smell, touch, sight and taste; all of the cognitive reasoning ability -- then you could create a chip to simulate those skills. Algorithms for image recognition, feature extraction, speech recognition, etc. are all available that are very very close to what humans can do.
But the thing that separates humans is that it didn't take hundreds of years of mathematical development to come up with these algorithms. The human brain develops these algorithms through changes in its structure from birth. At about age 10, speech recognition specialized and tailored to the dialect, language and tones that the person hears has developed on its own.
That type of structural formation and learning is what would need to happen in silicon to make a truly intelligent machine. Neuron clusters emulated using transistors would need to be able to dynamically form connections to other neuron clusters. There'd have to be some type of distributed learning algorithm encoded in the operation of each individual neuron.
Speech recognition is easy. Image recognition is easy. Developing a distributed, scalable, self-modifying architecture that can learn all of those and more on its own with nothing more than training samples is the difficult part.
I assume they have the activation key on hand and will request that Microsoft deactivate it.
I don't think that's entirely true. While I'm sure there is a stigma against non-profits, teaching is generally considered a social nobility. The adage "those who can't teach" was more a commentary on the poor quality and low qualification-for-entry of teachers in the current U.S. education system -- a criticism that is for the most part true.
Anecdotal but to the point: during college, the College of Education was a laughing joke amongst the other departments. It was literally a free "A" for anyone who took a College of Education course. One of our math professors taught a class for the CoE and tried to actually teach them math -- that is, make them understand the concept instead of just letting them coast and handing out a free "A". The class went ape-shit; they couldn't believe they had to actually work for their grade.
A degree from my university in Education was considered fairly prestigious and I knew many who got their Master's and went on to teach in very high-income public school areas. Not one of them -- despite being calculus teachers -- can actually explain to you the concept of a limit beyond the rules in the textbook.
So before you blame the industry -- and I'm not saying the CEO-worshipping trogs are innocent in all of this -- perhaps a look at government and how (poorly) it compensates teachers -- and therefore attract low-quality applicants.
There is also risk that a male will get a better job ... so how is that any different?
You can't offer a pregnant mother-to-be a salary increase to not have her baby. You also don't have to re-hire the guy if his other job goes south. Once he leaves the company, that's it.
I don't disagree with maternity leave but either:
1. It should be a perk -- something you sign up for like premium health insurance -- upon hire in return for a lower salary.
2. Something mandated by the government but there be a provision of allowing lower salaries for those who qualify.
Not really. The average marrying age for girls used to be 13-20. If you made it past then, you were either on your second marriage already or were desperately trying to find a husband.
Today's standards of beauty follow the same premise. The difference is, women are getting married later in life -- due to how much preparation is given before they are considered adults -- and thus, desire to be unnaturally skinny in order to resemble young girls in the 13-20 range.
Biologically, this makes sense. Human males are attracted to just post-pubescent girls as that is their most fertile time. I'd say that it isn't so much the standards of beauty that are changing, but rather social views of when a woman can be viewed as sexually attractive.
If you mention to someone in the 1920's that one can get arrested for photoshoping a pic of Miley Cyrus (16 years old), they'd look at you funny.
They used old photographs in the study of people from past generations and their method of "objective" measurement of beauty were to have modern-day people judge them.
It seems almost a foregone conclusion that people in modern times would find the women of modern times more attractive; standards of beauty change.
Nobody ever learned to program a GPU because there is a very very nice standardized -- two of them actually -- API abstraction layer; DirectX and OpenGL. Common graphics functions that are coupled with hardware by the library.
IMO, this is exactly what IBM/Sony should be doing with the Cell -- create a standard library as well as have a staff dedicated to consulting/developing low-level functions and farm those guys out to game developers.
While true, the vast majority of the reason developers say the 360's PPC is "faster" is that it's easier to program. In terms of theoretical flops, the CELL inside the PS3 is far more capable.
If you just strung together a bunch of code and hit "compile" on GCC, the 360 will be lightyears ahead. If you have 2-3 really good engineers writing sequences for the CELL, the vertex computational power will be far superior.
No. In fact the biggest improvement of this car appears to be the nanophosphate battery. It doesn't use the chemicals inside traditional li-ions that become heated when overcharged (lithium particles start leaking across to the anode).