Honestly, the comparison of percentage success is of minimal concern here. What I'm more curious about is, what are their testing methodologies. Did they record someone and play the corresponding queries at the same volume and distance from each of the products' microphones while keeping the acoustics the same?
If they didn't then this test should be taken with a serious grain of salt, since enunciation and environment could be the biggest contributor to the differences.
Currently the D-Wave device is able to beat a few classical algorithms, but if multiple algorithms are used in combination, then there is no contest. https://www.dwavesys.com/sites...
There are some solvers that make use of the D-Wave's limited connectivity (which will improve in the near future) like the HFS solver. These will become less effective as the connectivity improves
Overall, there may be a chance there isn't any single problem that a properly designe algorithm couldn't solve in an equivalent time with decently fast hardware.
Sounds more like someone who doesn't understand that computer science isn't a degree in programming. It's a degree in the theory of programming, not in programming itself. So, to say that it's bad that a computer science teacher doesn't know how to code means nothing bad about them if they know the theory. Pretty much the only reason CS majors learn to code is to implement the theoretical algorithms and structures in a way that is concrete.
Depending on the institution many of the CS departments came directly out of the mathematics Department. Which is one big reason why many of them are highly math oriented.
Isn't the system nearly identical to the automated calling/automated systems that are in place in many businesses at the moment just on a significantly more personalised level? I understand there may be some confusion if the person on the other end doesn't know they're talking with an AI and it may be worrying if the call is recorded (really, how do the wire tapping laws work when one party isn't a human but is speaking on behalf of one, or better yet, if both parties are machines?).
Duplex, and other systems like it that will most definitely be developed, are just an extension of what businesses are using in their calling centres already.
Do hate to say this, but it is possible to get a statistically significant result from a sample that is a small percentage of a population. It's not a matter of the percentage size of the population that the sample is, but the number of elements in the sample. The confidence in the measurements decrease as the inverse square root of the number of elements in the sample. It doesn't matter if the sample is 1% or 0.0001% of the population (of course as the sample size converges to the population size the sample indicates the true values of the population). So, we can still make statistically significant conclusions from the data we have right now.
This only gives a person's work history? Far less of an issue than getting a loan in another person's name. Unless someone can show me a hack that makes use of this information that's worse than getting a credit line... Many places are also making a switch to transparent salaries anyway. Again, why is this a big deal?
There is probably the possibility of it happening in the reasonably far future in decently large cities. There are a couple problems with making a giant mesh network. A couple have been mentioned above.
Routing tables would be fairly hard to figure out how to handle effectively, especially if each device in the mesh network could move around ( like if we were using a cell phone as a node in the network ). The routes would have to update extremely fast with extreme variability in each nodes transfer speeds, latency, reliability, etc. Not something we can currently do fast enough. Even if we limit ourselves to stationary nodes there may be a large number of hops, which means fairly long latency. So we'd need a trunk, which almost sounds like your typical ISP.
Then we have the issue of distance between nodes. Depending on population density, each node may be out of range of any other node. Making the idea fall apart for any place with reasonably low population density. I'm imagining this will be a mostly wireless type mesh network ( as the OP seems to indicate asking about frequency allocations ). So, we could increase transmission power, but then we'll have problems with collision avoidance algorithms increasing latency since we'd have to have confirmation that no other devices will be communicating over a fairly large radius of transmission assuming people will want to pay enough for such a large transmitter/receiver, which also means certain nodes would become trunk nodes, fairly similar to what an ISP already will do.
Since we have such a large amount of distance between nodes in certain areas we'll most likely have disconnected networks. Especially over oceans and uninhabitable mountainous regions.
HAM radio is actually a possibility to build a network like this along with bluetooth and wifi over short/medium distances. So there does exist some spectrum already for experimental/commercial devices. But there's another problem, if this network does get built, there would need to be a sharing of some of the spectrum for this network, as only certain frequencies have the capability of being transmitted long enough distances, or being capable of anything other than line-of-sight type communication.
I'm sure I'm missing some problems with creating a world wide mesh network. These issues would severely limit the feasibility of any global type mesh network, unfortunately.
This article reminds me of using a quantum monte carlo algorithm vs a classical monte carlo. The quantum monte carlo scales linearly with the number of trotter slices, but really a real quantum algorithm should use 2^n slices each coupled to one another. It definitely sounds like the people who are doing this research are at least 30 years behind the technological times. The biggest reason for not having all the memory directly accessible in a single processor computer is technological feasibility and cost. Hence caching. A GPU can do massive parallel calculations just like this neural network seems to be doing. Fault tolerance would just be deactivating the various processors that are no longer functioning properly. Seems like this problem is already known and solved.
I remember reading an article about how dragonflies were using stones to tap down their nests making it harder for predators to find. The result was a reclassification on what constituted a tool removing the dragonflies from being classified as tool users.
Many animals use tools. So, I don't really see how this is news worthy other than that the bird learned to build them on its own without help from other birds.
I think you missunderstood the last post. I was refering to the word "Calibrating" in the parent. As it was posted using my phone autocorrect decided to change "jailbreaking" into "Calibrating". This change wasn't what I had intended, and, while admitedly, I should have checked the post to ensure that none of those really obvious substitutions was present, it was there and I attempted to make the post less obsfuscated. I suppose an assignment would have been more appropriate to help with understanding, or perhaps reposting the entirety of it, however, that may have resulted in more of the same errors decreasing readability and increasing my annoyance with my telephone.
Aren't jailbreaking and unlocking different? Calibrating refers to being able to execute cystine programs on your phone, whereas unlocking allows one to use any carrier. I think this article is referring to jailbreaking.
It's not even a generalization. It's the exact same algorithms. Once you can solve a 4x4x4 there is no extra algorithms needed at all to solve any cubes of any higher degree.
The typical idea is to solve the cube to the point of being a 3x3x3 with all the centers and edges solved. When solving the edges if you're trying to get the edges in place any portion of the edges can be grouped to look like a cube with a smaller degree.
Someone could argue that the 5x5x5 and 4x4x4 algorithms are needed, because of the extra center edge pieces, but with proper edge pairing that's usually a moot issue. But, I could always be wrong, the 4x4x4 and 5x5x5 algorithms might be needed for any cube larger than a 3x3x3. The real point is that anything above a certain point is the exact same algorithms.
Nailing, even by hand, isn't old-school manual construction.
Nailguns are mostly only useable in certain circumstances, like when doing framing, or woodwork finishing. When it comes to things like putting metal siding, spiking concrete, or putting in drywall beads nail guns don't have the accuracy, nor the light touch required.
Also, if you're meaning screws are the wave of the future, they don't have the type of give that are needed in many situations. For instance, if a building was put up with only screws in an earthquake the screws would tend to break whereas the nails would allow for some room to move and maintain their hold. Either way, nailing is still new school.
I'm pretty sure that not being paid would instroduce more bias than having some kind of standard rate. The reason I say this is because many of the reviewers will just briefly look over the paper and give some kind of default opinion on it without really reading it well. With monetary incentive they may do a better job of reviewing it as their time is paid for.
Then again, it could also introduce a problem where people still just give a default review and take the money.
Perhaps a study needs to be done on this so that the journal can be locked behind a paywall never to be read by anyone as the price would be too high.
I'm all for having more expert computers in the fields, as it means we can potentially get better treatment. But, people tend to be lazy, and I'm sure, given as many patients as some doctors have to see in a day, that some of the people that are supposed to be doing what we hire doctors to do will be just asking watson for the treatment with probably only a half-assed attempt at verifying how good those results may be. It seems like Isabel might have a bit more promise in these markets for safety.
Also, if Isabel doesn't understand non-jargon, why not develop a way to use Watson to "translate" into technical speak for Isabel. Then again, that might just be a google translate style accident waiting to happen.
I believe that the Netherlands was originally called Holland by many outsiders due to the confusion that originally came from the Dutch language itself. As Hollands is one of the old terms used to describe the Dutch language. To make matters worse, things that were considered "Dutch", at least in old Dutch, were called Hollandse. These two issues most likely contributed to the confusion about Holland vs Netherlands.
Another issue is that the Netherlands, about 200 years ago or so, consisted of, at least in part (I don't recall if it was the entirety or not), of Belgium, and Belgium then was known as South Netherlands.
It would appear that there is a strong push to try and make Netherlands to be the official term and uniform term for the Netherlands and it's language, as the language is now known as "Nederlands" in Dutch, and the land is refered to as "Nederland."
I remember one of my teachers telling us about a project he worked on where they needed to keep track of the people who were working on it. They came onto the bright idea of just giving everyone a cell phone. Simple, as the employees would be happy to have a cool new phone to use, and they could easily keep track of them.
There was, I believe, a service that you could call that would allow you to get GPS like service before GPS became popular. This service was also used by emergency service to determine where you were as well.
Having multiple booleans in a single word definitely makes sense to keep the true == 1 convension. I suspect, however, that in a case like the voyager probe they would likely have had multiple copies of the same memory to prevent radiation problems.
However, as Dog-Cow pointed out, true is defined as !false, which means that any value, including -1, would be classified as true. So, in at least the cases where one were trying to go for software that could counteract possible hardware problems of that likeness it would be a safer idea to go with -1 instead of 1.
Would probably be even more expensive given the number of features implemented in Windows.
The voyager probe code was most likely entirely purpose written, which is much easier to manage than something like windows which is designed for third party programs and tries to allow for general purpose computing.
Although, there are a few things that could be done to improve some reliability, even in old c++. Although I'm sure there might be difficulties with platforms other than x86. Like using -1 instead of 1 as "true". Thereby having all 1's set instead of a single bit which, in theory, could change.
But in reality, there's probably more issue with programmer's bugs than hardware issues.
One major question that I have had is why has the standard library not been expanded over the years. Code reuse is one of the best ways to reduce working bugs in pretty much all code. Although that's probably a result of many patent and IP issues.:/
Slashdot Mirror
If they didn't then this test should be taken with a serious grain of salt, since enunciation and environment could be the biggest contributor to the differences.
And the device has been shown to demonstrate some qualities in simulating quantum materials https://www.dwavesys.com/sites...
There are some solvers that make use of the D-Wave's limited connectivity (which will improve in the near future) like the HFS solver. These will become less effective as the connectivity improves
Overall, there may be a chance there isn't any single problem that a properly designe algorithm couldn't solve in an equivalent time with decently fast hardware.
Depending on the institution many of the CS departments came directly out of the mathematics Department. Which is one big reason why many of them are highly math oriented.
Duplex, and other systems like it that will most definitely be developed, are just an extension of what businesses are using in their calling centres already.
Do hate to say this, but it is possible to get a statistically significant result from a sample that is a small percentage of a population. It's not a matter of the percentage size of the population that the sample is, but the number of elements in the sample. The confidence in the measurements decrease as the inverse square root of the number of elements in the sample. It doesn't matter if the sample is 1% or 0.0001% of the population (of course as the sample size converges to the population size the sample indicates the true values of the population). So, we can still make statistically significant conclusions from the data we have right now.
It may be a clone of openworm.org.
I'm curious then, how many people are giving their potential employers their SSN and DOB before actually having the job?
This only gives a person's work history? Far less of an issue than getting a loan in another person's name. Unless someone can show me a hack that makes use of this information that's worse than getting a credit line... Many places are also making a switch to transparent salaries anyway. Again, why is this a big deal?
Routing tables would be fairly hard to figure out how to handle effectively, especially if each device in the mesh network could move around ( like if we were using a cell phone as a node in the network ). The routes would have to update extremely fast with extreme variability in each nodes transfer speeds, latency, reliability, etc. Not something we can currently do fast enough. Even if we limit ourselves to stationary nodes there may be a large number of hops, which means fairly long latency. So we'd need a trunk, which almost sounds like your typical ISP.
Then we have the issue of distance between nodes. Depending on population density, each node may be out of range of any other node. Making the idea fall apart for any place with reasonably low population density. I'm imagining this will be a mostly wireless type mesh network ( as the OP seems to indicate asking about frequency allocations ). So, we could increase transmission power, but then we'll have problems with collision avoidance algorithms increasing latency since we'd have to have confirmation that no other devices will be communicating over a fairly large radius of transmission assuming people will want to pay enough for such a large transmitter/receiver, which also means certain nodes would become trunk nodes, fairly similar to what an ISP already will do. Since we have such a large amount of distance between nodes in certain areas we'll most likely have disconnected networks. Especially over oceans and uninhabitable mountainous regions.
HAM radio is actually a possibility to build a network like this along with bluetooth and wifi over short/medium distances. So there does exist some spectrum already for experimental/commercial devices. But there's another problem, if this network does get built, there would need to be a sharing of some of the spectrum for this network, as only certain frequencies have the capability of being transmitted long enough distances, or being capable of anything other than line-of-sight type communication. I'm sure I'm missing some problems with creating a world wide mesh network. These issues would severely limit the feasibility of any global type mesh network, unfortunately.
This article reminds me of using a quantum monte carlo algorithm vs a classical monte carlo. The quantum monte carlo scales linearly with the number of trotter slices, but really a real quantum algorithm should use 2^n slices each coupled to one another. It definitely sounds like the people who are doing this research are at least 30 years behind the technological times. The biggest reason for not having all the memory directly accessible in a single processor computer is technological feasibility and cost. Hence caching. A GPU can do massive parallel calculations just like this neural network seems to be doing. Fault tolerance would just be deactivating the various processors that are no longer functioning properly. Seems like this problem is already known and solved.
I remember reading an article about how dragonflies were using stones to tap down their nests making it harder for predators to find. The result was a reclassification on what constituted a tool removing the dragonflies from being classified as tool users.
Many animals use tools. So, I don't really see how this is news worthy other than that the bird learned to build them on its own without help from other birds.
I think you missunderstood the last post. I was refering to the word "Calibrating" in the parent. As it was posted using my phone autocorrect decided to change "jailbreaking" into "Calibrating". This change wasn't what I had intended, and, while admitedly, I should have checked the post to ensure that none of those really obvious substitutions was present, it was there and I attempted to make the post less obsfuscated. I suppose an assignment would have been more appropriate to help with understanding, or perhaps reposting the entirety of it, however, that may have resulted in more of the same errors decreasing readability and increasing my annoyance with my telephone.
Stupid phone. Calibrating == jailbreaking.
Aren't jailbreaking and unlocking different? Calibrating refers to being able to execute cystine programs on your phone, whereas unlocking allows one to use any carrier. I think this article is referring to jailbreaking.
Is that regular avian delivery? or with QOS?
It's not even a generalization. It's the exact same algorithms. Once you can solve a 4x4x4 there is no extra algorithms needed at all to solve any cubes of any higher degree.
The typical idea is to solve the cube to the point of being a 3x3x3 with all the centers and edges solved. When solving the edges if you're trying to get the edges in place any portion of the edges can be grouped to look like a cube with a smaller degree.
Someone could argue that the 5x5x5 and 4x4x4 algorithms are needed, because of the extra center edge pieces, but with proper edge pairing that's usually a moot issue. But, I could always be wrong, the 4x4x4 and 5x5x5 algorithms might be needed for any cube larger than a 3x3x3. The real point is that anything above a certain point is the exact same algorithms.
Nailing, even by hand, isn't old-school manual construction.
Nailguns are mostly only useable in certain circumstances, like when doing framing, or woodwork finishing. When it comes to things like putting metal siding, spiking concrete, or putting in drywall beads nail guns don't have the accuracy, nor the light touch required.
Also, if you're meaning screws are the wave of the future, they don't have the type of give that are needed in many situations. For instance, if a building was put up with only screws in an earthquake the screws would tend to break whereas the nails would allow for some room to move and maintain their hold. Either way, nailing is still new school.
I'm pretty sure that not being paid would instroduce more bias than having some kind of standard rate. The reason I say this is because many of the reviewers will just briefly look over the paper and give some kind of default opinion on it without really reading it well. With monetary incentive they may do a better job of reviewing it as their time is paid for.
Then again, it could also introduce a problem where people still just give a default review and take the money.
Perhaps a study needs to be done on this so that the journal can be locked behind a paywall never to be read by anyone as the price would be too high.
I'm surprised noone has said it yet, but do they have a predefined kill limit? So, if they get out of hand we can send wave after wave of men at them?
I'm all for having more expert computers in the fields, as it means we can potentially get better treatment. But, people tend to be lazy, and I'm sure, given as many patients as some doctors have to see in a day, that some of the people that are supposed to be doing what we hire doctors to do will be just asking watson for the treatment with probably only a half-assed attempt at verifying how good those results may be. It seems like Isabel might have a bit more promise in these markets for safety.
Also, if Isabel doesn't understand non-jargon, why not develop a way to use Watson to "translate" into technical speak for Isabel. Then again, that might just be a google translate style accident waiting to happen.
You call that a bag? This is a bag.
That's not a bag, that's a jar.
Oh, I see you've played baggy-jarry before.
I believe that the Netherlands was originally called Holland by many outsiders due to the confusion that originally came from the Dutch language itself. As Hollands is one of the old terms used to describe the Dutch language. To make matters worse, things that were considered "Dutch", at least in old Dutch, were called Hollandse. These two issues most likely contributed to the confusion about Holland vs Netherlands.
Another issue is that the Netherlands, about 200 years ago or so, consisted of, at least in part (I don't recall if it was the entirety or not), of Belgium, and Belgium then was known as South Netherlands.
It would appear that there is a strong push to try and make Netherlands to be the official term and uniform term for the Netherlands and it's language, as the language is now known as "Nederlands" in Dutch, and the land is refered to as "Nederland."
I would definitely agree with that.
I remember one of my teachers telling us about a project he worked on where they needed to keep track of the people who were working on it. They came onto the bright idea of just giving everyone a cell phone. Simple, as the employees would be happy to have a cool new phone to use, and they could easily keep track of them.
There was, I believe, a service that you could call that would allow you to get GPS like service before GPS became popular. This service was also used by emergency service to determine where you were as well.
Although I could just be off my rocker.
Having multiple booleans in a single word definitely makes sense to keep the true == 1 convension. I suspect, however, that in a case like the voyager probe they would likely have had multiple copies of the same memory to prevent radiation problems.
However, as Dog-Cow pointed out, true is defined as !false, which means that any value, including -1, would be classified as true. So, in at least the cases where one were trying to go for software that could counteract possible hardware problems of that likeness it would be a safer idea to go with -1 instead of 1.
Would probably be even more expensive given the number of features implemented in Windows. :/
The voyager probe code was most likely entirely purpose written, which is much easier to manage than something like windows which is designed for third party programs and tries to allow for general purpose computing.
Although, there are a few things that could be done to improve some reliability, even in old c++. Although I'm sure there might be difficulties with platforms other than x86. Like using -1 instead of 1 as "true". Thereby having all 1's set instead of a single bit which, in theory, could change.
But in reality, there's probably more issue with programmer's bugs than hardware issues.
One major question that I have had is why has the standard library not been expanded over the years. Code reuse is one of the best ways to reduce working bugs in pretty much all code. Although that's probably a result of many patent and IP issues.