It's not that humans are not adaptable, it's that parallel computing is hard for humans to figure out. Linear execution lends itself to all kinds of easy abstractions: loops, branches, methods, etc. Parallel computing, not so much. Mutexes are awful. The best we've got is message passing and functional programming, but even that is hard to design correctly to be both understandable and exploit inherent parallelism.
Y'know what's even harder to design? Analog computing. Holy cow. Remember, digital computing was invented by Touring before we even had built a computer. It's easy to visualize how it works. My brain explodes though trying to imagine a fuzzy-logic analog equivalent of a touring machine.
I used to think that AI research combined with neuroscience would figure out a simple solution to this problem, but it's increasingly seeming like, no, it's even complicated in the brain.
So people can pine for analog memristor computation, and analog optical computing all they want, but the hardware is the easy part here. Get the software side solved, and if you build it they will come. But it's not because we aren't used to these problems, it's because these problems are really really hard.
From the GPL:
"a) You must cause the modified files to carry prominent notices stating that you changed the files and the date of any change."
So it is required for them to list their changes.
Speech recognition isn't too CPU intensive, but it's *massively* memory intensive. It's not unreasonable for speech recognition engines to eat up a gig of ram, and the 4S only has 512mb. However, push it to a server with lots of ram and it can handle lots and lots of simultaneous speech recognition queries. It's tailor made to be a server-side task. At least until phones have gigs of free memory that aren't needed.
Even with transistors the same size, there are so many avenues to explore in processor design.
Just off the top of my head, how about a memristor-based analog co-processor for tasks like facial detection or language/speech recognition.
How about processors with asynchronous clocks, or clockless designs. Sure, they're harder to build, but once transistor sizes fixate, might as well spend the effort because designs will have a much longer lifecycle.
Let me preface this answer by revealing that I no longer work in the video game industry, as I did not enjoy it enough to stay.
A lot of people cut their teeth on writing Windows stuff for fun, maybe working on mods, but a fair amount of developers worked their way up from QA. At least where I worked, it seemed like there were way too many people wanting to get into the video games industry, and once they did get in, they worked their asses off. People would learn to code due to their love of games, not because they liked coding. There seemed to be a lot of very bright high-school guys who, instead of doing the whole computer-science thing at a university, would work QA, and then progress up to be a developer. These people were highly respected because of their commitment.
There was another group of people who formed the more senior developers who got started in academia. People who worked on the engines ususally had PHD's in computer science with an emphasis on graphics. I would think graduate work on game theory or AI would put you in this group.
Being an old-school linux hacker who cut his teeth by contributing to OSS projects, I felt a bit out of place. Most of the guys in the industry don't leave because the idea of working on something other than video games is distasteful. Me, I find lots of engineering problems satisfying.
I used to work for Sony developing PS2 games. The number of people I met that cut their teeth writing code on the linux kit before getting into the business was exactly 0. I might have been the only person I knew who even had a modchipped PS2, everybody else just didn't care since they had the PS2Tool on their desk to do development. Sony is probably discontinuing offering Linux because it didn't spark the development push that they had hoped for.
Still, I would think this would limit the number of supercomputer clusters that use PS3's. You'd think the marketing benefits of being a platform in the top 100 supercomputers would be valuable, but perhaps Sony is still willing to work with academic institutions to make this possible still.
The reason that memristors were so quick to market is that they were discovered while researching new fabrication processes. This isn't like, say, carbon nanotubes where they're able to make one or two in a lab. For memristors, they had a fabrication process from day one.
I heard a report on the Diane Rehm show that that's exactly what they're planning to do. The other approach that is forthcoming is convoys where multiple shipping vessels get together with one armed gunboat to protect them. Blackwater has come forth as a potential candidate for the security contract of future convoys.
Polls indicate that Obama will win the election with a sizable Democratic majority in both houses as well. This gives him two options to lead the country:
1) He leads from the left (or the center of his party) and the Democrats ram their ajenda past the hapless Republicans in the legislature. This is the easiest approach, but I fear that it will alienate the 45% of the population who didn't vote Democrat, in the same way that Bush has alienated Democrats. The result is a productive, but unpopular president.
2) He leads from the center (or the right of his party) and tries to be post-partisan even when he doesn't have to. This approach is going to be fairly challenging, because the legislature can tell him to bugger off and adopt method #1 without him. However, if he's successful, he could end up being one of the best presidents in history. With all of the gigantic issues coming to the fore in his presidency, if Obama can actually muster the bipartisan government he claims to support, the resulting solutions to energy, healthcare, taxes, and the budget deficit will be triumphant.
It's not that her phone company refused, it's that they can't do anything about it. A local phone carrier can't find out who's using that caller-id if the call is made from some other exchange. Phone companies don't do egress filtering of caller-id, so there's nothing stopping you from spoofing your caller-id as a New York number from your phone in California. When the old lady in New York calls her local phone carrier, there's nothing they can do to help her.
Short sales can be used in reliable investing strategies. For example, say you purchased the S&P 500 index for a long-term investment, but you feel that the auto manufacturers are a bad play for the next few years. You can short an equivalent amount of auto stocks to effectively remove that component from your S&P holdings.
The weird thing about Jabber Inc. is how irrelevant they are to the XMPP scene. There's a huge array of jabber servers and clients out there, and Jabber Inc. doesn't really have anything to do with any of them. Then there's the whole branding shift from calling it XMPP instead of Jabber. I'm not quite sure what Jabber Inc. brings to the table for Cisco to buy them.
like the way they "stereoscopically" create a depth-map from a _single_ still photograph
TFV said they were using video frames to do stereoscopic depth-mapping. Since the source footage changed perspective, they can build a depth map based on the relative shift of each object in the video, and then project the high-quality photograph on top of the derived 3D structure
Transponders aren't enough. Unless you make sure that people only throw couches into the road that also include transponders, you're gonna need real obstacle detection.
Heaven forbid a sinkhole opens up, swallowing 100's of robocars like lemmings to a cliff.
That doesn't mean there's not a software solution to this problem. Organisms like lizards do quite well at visually detecting obstacles with brain power that's reasonably close to modern computers. While it might be a hugely difficult software problem, it's at least conceivably doable with today's hardware.
Ok, so it might not be as extreme as all that, but have you seen the inside of a taxi? It's thrashed, and I think the only reason it's not more thrashed is that there's an taxi driver who would beat you up if you did something stupid.
TFAuthor says that people might want to rent their robocars out while they're at work. Like hell I would! The last thing I need is some jackass with a spiked belt ripping a hole in my leather seats.
If robocars become practical, and energy costs rise, it's possible that the author's vision will be inevitable. Still, it's gonna suck to find that some bum puked in my robotaxi right when I'm late for work.
Maybe we can engineer robocops to sit in every robotaxi to prevent the vandals from ruining it for all of us.
Wow, that's a good point. And since it actually stores it's bits in an analog storage medium, you could in theory recover data from the memristors even after they've been written over, just like they do with magnetic drives.
Every single example you list above is based on the transistor. Sure, there's lots of variations (MRAM using magnets, PRAM using chalcogenide glass, FeRAM using a ferroelectric layer, etc.), but these are all basically: glue stuff on a transistor to store data.
Memristor-based RRAM is different. It doesn't use transistors at all. This is truly a departure from all of the exsting RAM technologies, and while the prospect of RRAM storage is pretty cool, the possibility of analog computing using memristors is even neater. I'm cautiously optimistic that this technology is going to take computing in some interesting directions.
TFA states that the attorneys that brought the case are demanding 1.3 million in legal fees, way more than the 2,676 * (max $35) = $93,660 settlement fees that Take Two will have to pay.
The difference between a memristor and FeRAM is that because the memristor is constructed without using any transistors, it can be used as a kind of analog memory. Instead of just storing 1's and 0's, it's resistance is an analog value anywhere in the range of on and off. Of course you can still use it to store digital data, but the real fun will come when you interconnect these things to emulate the analog behavior of the brain. This is where the claim of pattern recognition and facial recognition come in. They're not actually talking about software there but the actual analog capabilities of circuitry built with memristors.
The other amazing thing about memristors is how small they are. The articles state that you can emulate a transistor by connecting a few memristors, and that transistor is smaller than any we have today. Also it states that the memristor actually performs better at smaller sizes. This really is neat stuff.
I've actually been trying to hire superstars for my company recently, and I find they fit into two categories:
1) Superstars who are content where they are are purely fought-over by their former colleagues. They never look for work because they're constantly being recruited by their peers. If they do look for work, they just apply directly to Google, or wherever else they might want to work.
2) Superstars who are more ambitious quickly advance to a point where they only really work for companies they found themselves. At this point, even if you do know them, it's hard to hire them unless you're going in on a joint venture.
With these two facts in mind, the only real way to hire superstars is to make them. Find younger talent, preferably straight out of college and groom them for 3-4 years. If you're near a university, try to strike up a dialog with the professors, see if you can get access to their TAs or star students. Talented programmers tend to hang out with each other, so if you do land a superstar, you've got a good chance at hiring their buddies too.
27 years is a long time to project for technology. For example, Ray Kurzweil bet $10,000 that computers will have passed the turing test by 2029. Even if you think Kurzweil is an optimistic hack, 27 years is 18 iterations of Moore's law. If that continues, we'll have computers with 200,000 cores and 32 petabyte hard drives by 2035. I'm not saying that will happen, my point is just that it's probably not prudent to make such long-term plans wrt defense technology, because it's quite likely that technological advancements will make most of your plans obsolete by the time you get that far out.
Perhaps he can also uncover why this reviewer thinks that a $60 aftermarket DVD power cable somehow affects it's digital video output. From the review:
Colours of the individual vehicles come out much richer, and the all-important skin tone (she shows quite a bit of it too...) is more natural. Edges are more defined, which makes it easy to make out the shapes and movement of vehicles far below. The biggest improvement, though, was in terms of contrast, and it was easier to make out details on areas of shadow than before.
Too bad they left out Intel's newest low-end core2 duo proc, the e2160. It's basically an e4300 with less l2 cache and 40 bucks cheaper. The e2160 starts to put a lot of pressure on AMD's low cost dual-core offerings.
It would be unfortunate to lay all of the blame for this on the corporations providing the support. When you combine an incredibly complex piece of machinery such as a computer and it's OS with a customer base that refuses to pay extra for support, this is what happens. The field of questions that a general PC tech support rep is expected to answer is insanely broad. You can't pay the bare minimum for tech-support staff and expect them to be able to field these questions with any competancy, and scripting will only get you so far. But consumers have a lot of trouble paying for intangibles like "good support". So retailers have to bake-in the support costs into their product and only provide the bare minimum of help.
This is especially evident in the open-source world. Even when people get the software for free, they feel entitled to help from the project owners. However, when people try to charge for support services, customers balk at paying. Only when you get to very-large corporations with very deep support requirements does the support/services gig become profitable.
Slashdot Mirror
It's not that humans are not adaptable, it's that parallel computing is hard for humans to figure out. Linear execution lends itself to all kinds of easy abstractions: loops, branches, methods, etc. Parallel computing, not so much. Mutexes are awful. The best we've got is message passing and functional programming, but even that is hard to design correctly to be both understandable and exploit inherent parallelism.
Y'know what's even harder to design? Analog computing. Holy cow. Remember, digital computing was invented by Touring before we even had built a computer. It's easy to visualize how it works. My brain explodes though trying to imagine a fuzzy-logic analog equivalent of a touring machine.
I used to think that AI research combined with neuroscience would figure out a simple solution to this problem, but it's increasingly seeming like, no, it's even complicated in the brain.
So people can pine for analog memristor computation, and analog optical computing all they want, but the hardware is the easy part here. Get the software side solved, and if you build it they will come. But it's not because we aren't used to these problems, it's because these problems are really really hard.
From the GPL: "a) You must cause the modified files to carry prominent notices stating that you changed the files and the date of any change." So it is required for them to list their changes.
Speech recognition isn't too CPU intensive, but it's *massively* memory intensive. It's not unreasonable for speech recognition engines to eat up a gig of ram, and the 4S only has 512mb. However, push it to a server with lots of ram and it can handle lots and lots of simultaneous speech recognition queries. It's tailor made to be a server-side task. At least until phones have gigs of free memory that aren't needed.
Even with transistors the same size, there are so many avenues to explore in processor design. Just off the top of my head, how about a memristor-based analog co-processor for tasks like facial detection or language/speech recognition. How about processors with asynchronous clocks, or clockless designs. Sure, they're harder to build, but once transistor sizes fixate, might as well spend the effort because designs will have a much longer lifecycle.
Agreed. Star Wars very well could have had a medieval setting and it would have made no real difference to the plot..
I think that movie was called Willow
Let me preface this answer by revealing that I no longer work in the video game industry, as I did not enjoy it enough to stay. A lot of people cut their teeth on writing Windows stuff for fun, maybe working on mods, but a fair amount of developers worked their way up from QA. At least where I worked, it seemed like there were way too many people wanting to get into the video games industry, and once they did get in, they worked their asses off. People would learn to code due to their love of games, not because they liked coding. There seemed to be a lot of very bright high-school guys who, instead of doing the whole computer-science thing at a university, would work QA, and then progress up to be a developer. These people were highly respected because of their commitment.
There was another group of people who formed the more senior developers who got started in academia. People who worked on the engines ususally had PHD's in computer science with an emphasis on graphics. I would think graduate work on game theory or AI would put you in this group.
Being an old-school linux hacker who cut his teeth by contributing to OSS projects, I felt a bit out of place. Most of the guys in the industry don't leave because the idea of working on something other than video games is distasteful. Me, I find lots of engineering problems satisfying.
I used to work for Sony developing PS2 games. The number of people I met that cut their teeth writing code on the linux kit before getting into the business was exactly 0. I might have been the only person I knew who even had a modchipped PS2, everybody else just didn't care since they had the PS2Tool on their desk to do development. Sony is probably discontinuing offering Linux because it didn't spark the development push that they had hoped for. Still, I would think this would limit the number of supercomputer clusters that use PS3's. You'd think the marketing benefits of being a platform in the top 100 supercomputers would be valuable, but perhaps Sony is still willing to work with academic institutions to make this possible still.
The reason that memristors were so quick to market is that they were discovered while researching new fabrication processes. This isn't like, say, carbon nanotubes where they're able to make one or two in a lab. For memristors, they had a fabrication process from day one.
I heard a report on the Diane Rehm show that that's exactly what they're planning to do. The other approach that is forthcoming is convoys where multiple shipping vessels get together with one armed gunboat to protect them. Blackwater has come forth as a potential candidate for the security contract of future convoys.
Polls indicate that Obama will win the election with a sizable Democratic majority in both houses as well. This gives him two options to lead the country:
1) He leads from the left (or the center of his party) and the Democrats ram their ajenda past the hapless Republicans in the legislature. This is the easiest approach, but I fear that it will alienate the 45% of the population who didn't vote Democrat, in the same way that Bush has alienated Democrats. The result is a productive, but unpopular president.
2) He leads from the center (or the right of his party) and tries to be post-partisan even when he doesn't have to. This approach is going to be fairly challenging, because the legislature can tell him to bugger off and adopt method #1 without him. However, if he's successful, he could end up being one of the best presidents in history. With all of the gigantic issues coming to the fore in his presidency, if Obama can actually muster the bipartisan government he claims to support, the resulting solutions to energy, healthcare, taxes, and the budget deficit will be triumphant.
I'm not optimistic that he'll be able to pull off #2, but if you look at his balanced approach at the Harvard Law Review, there's some hope that he'll at least try.
It's not that her phone company refused, it's that they can't do anything about it. A local phone carrier can't find out who's using that caller-id if the call is made from some other exchange. Phone companies don't do egress filtering of caller-id, so there's nothing stopping you from spoofing your caller-id as a New York number from your phone in California. When the old lady in New York calls her local phone carrier, there's nothing they can do to help her.
Short sales can be used in reliable investing strategies. For example, say you purchased the S&P 500 index for a long-term investment, but you feel that the auto manufacturers are a bad play for the next few years. You can short an equivalent amount of auto stocks to effectively remove that component from your S&P holdings.
The weird thing about Jabber Inc. is how irrelevant they are to the XMPP scene. There's a huge array of jabber servers and clients out there, and Jabber Inc. doesn't really have anything to do with any of them. Then there's the whole branding shift from calling it XMPP instead of Jabber. I'm not quite sure what Jabber Inc. brings to the table for Cisco to buy them.
like the way they "stereoscopically" create a depth-map from a _single_ still photograph
TFV said they were using video frames to do stereoscopic depth-mapping. Since the source footage changed perspective, they can build a depth map based on the relative shift of each object in the video, and then project the high-quality photograph on top of the derived 3D structure
Transponders aren't enough. Unless you make sure that people only throw couches into the road that also include transponders, you're gonna need real obstacle detection.
Heaven forbid a sinkhole opens up, swallowing 100's of robocars like lemmings to a cliff.
That doesn't mean there's not a software solution to this problem. Organisms like lizards do quite well at visually detecting obstacles with brain power that's reasonably close to modern computers. While it might be a hugely difficult software problem, it's at least conceivably doable with today's hardware.
Ok, so it might not be as extreme as all that, but have you seen the inside of a taxi? It's thrashed, and I think the only reason it's not more thrashed is that there's an taxi driver who would beat you up if you did something stupid.
TFAuthor says that people might want to rent their robocars out while they're at work. Like hell I would! The last thing I need is some jackass with a spiked belt ripping a hole in my leather seats.
If robocars become practical, and energy costs rise, it's possible that the author's vision will be inevitable. Still, it's gonna suck to find that some bum puked in my robotaxi right when I'm late for work.
Maybe we can engineer robocops to sit in every robotaxi to prevent the vandals from ruining it for all of us.
Wow, that's a good point. And since it actually stores it's bits in an analog storage medium, you could in theory recover data from the memristors even after they've been written over, just like they do with magnetic drives.
Every single example you list above is based on the transistor. Sure, there's lots of variations (MRAM using magnets, PRAM using chalcogenide glass, FeRAM using a ferroelectric layer, etc.), but these are all basically: glue stuff on a transistor to store data.
Memristor-based RRAM is different. It doesn't use transistors at all. This is truly a departure from all of the exsting RAM technologies, and while the prospect of RRAM storage is pretty cool, the possibility of analog computing using memristors is even neater. I'm cautiously optimistic that this technology is going to take computing in some interesting directions.
TFA states that the attorneys that brought the case are demanding 1.3 million in legal fees, way more than the 2,676 * (max $35) = $93,660 settlement fees that Take Two will have to pay.
The difference between a memristor and FeRAM is that because the memristor is constructed without using any transistors, it can be used as a kind of analog memory. Instead of just storing 1's and 0's, it's resistance is an analog value anywhere in the range of on and off. Of course you can still use it to store digital data, but the real fun will come when you interconnect these things to emulate the analog behavior of the brain. This is where the claim of pattern recognition and facial recognition come in. They're not actually talking about software there but the actual analog capabilities of circuitry built with memristors.
The other amazing thing about memristors is how small they are. The articles state that you can emulate a transistor by connecting a few memristors, and that transistor is smaller than any we have today. Also it states that the memristor actually performs better at smaller sizes. This really is neat stuff.
I've actually been trying to hire superstars for my company recently, and I find they fit into two categories:
1) Superstars who are content where they are are purely fought-over by their former colleagues. They never look for work because they're constantly being recruited by their peers. If they do look for work, they just apply directly to Google, or wherever else they might want to work.
2) Superstars who are more ambitious quickly advance to a point where they only really work for companies they found themselves. At this point, even if you do know them, it's hard to hire them unless you're going in on a joint venture.
With these two facts in mind, the only real way to hire superstars is to make them. Find younger talent, preferably straight out of college and groom them for 3-4 years. If you're near a university, try to strike up a dialog with the professors, see if you can get access to their TAs or star students. Talented programmers tend to hang out with each other, so if you do land a superstar, you've got a good chance at hiring their buddies too.
27 years is a long time to project for technology.
For example, Ray Kurzweil bet $10,000 that computers will have passed the turing test by 2029.
Even if you think Kurzweil is an optimistic hack, 27 years is 18 iterations of Moore's law. If that continues, we'll have computers with 200,000 cores and 32 petabyte hard drives by 2035.
I'm not saying that will happen, my point is just that it's probably not prudent to make such long-term plans wrt defense technology, because it's quite likely that technological advancements will make most of your plans obsolete by the time you get that far out.
Too bad they left out Intel's newest low-end core2 duo proc, the e2160.
It's basically an e4300 with less l2 cache and 40 bucks cheaper.
The e2160 starts to put a lot of pressure on AMD's low cost dual-core offerings.
It would be unfortunate to lay all of the blame for this on the corporations providing the support. When you combine an incredibly complex piece of machinery such as a computer and it's OS with a customer base that refuses to pay extra for support, this is what happens. The field of questions that a general PC tech support rep is expected to answer is insanely broad. You can't pay the bare minimum for tech-support staff and expect them to be able to field these questions with any competancy, and scripting will only get you so far. But consumers have a lot of trouble paying for intangibles like "good support". So retailers have to bake-in the support costs into their product and only provide the bare minimum of help.
This is especially evident in the open-source world. Even when people get the software for free, they feel entitled to help from the project owners. However, when people try to charge for support services, customers balk at paying. Only when you get to very-large corporations with very deep support requirements does the support/services gig become profitable.