Apples and oranges. Linux doesn't revolutionize the desktop, as automobiles revolutionized transportation. Linux's big selling point is that it's cheaper than Windows. If it can't interoperate or be used without more training or something of this nature, then the price advantage disappears.
Besides, the automobile took some time before it caught on everywhere--horses were still used for some purposes in WWI, and I'm sure the army wasn't the only one.
CUSTOMER: Well, we're having trouble making our new Linux boxes talk to our large installed base of Windows boxes.
MICROSOFT: You don't say.
C: No, really. We'd really like it if you were to make Windows boxes easier to talk to by publishing your heretofore closed standard.
M: So let me get this straight. You want us to make it easier for our competitors to replace us? If we do this, then our market share goes down because the barrier to switching lowers. But if we don't, you'll keep buying Windows because it's cheaper than doing a complete rollover. Let me consider that.
I'm not sure about the number of jurors, but perhaps more important is that the jurors only have to decide if the preponderance of evidence points to violation (that is, are you 50% sure the defendant is guilty), rather than beyond a reasonable doubt, as in criminal cases.
I think you misread the parent. I think the author says that this is a standard that corporate world does not necessarily adhere to, and thus it is in everyone's best interest to keep collection at the IRS.
My question is, why give the debt collectors all that information? All they need is a name, phone number, address, and the amount owed. No need to give them your entire tax history. So data privacy concerns are sort of bogus.
Yeah, it kinda does mean that. SCO owns the rights, not the management. If we all buy a share of stock (13.1M shares outstanding, but linuxcounter tells us there are ~18M users), then WE own the company and thus the rights to Unix that SCO has.
Not that I'm suggesting this is a good course of action, but there you go.
Yes, the initial set is random. And if you read the article, you'd see that each successive generation of this particular formulation contains 80% completely random programs. This is not uncommon. In addition, the 20% that come from the previous generation are randomly mutated. There's no analysis of the program to determine which pieces of the program worked and which didn't. Even in instances of "sexual" GA, the mating between two programs takes a randomly selected piece of one and joins it with a randomly selected piece of the other.
Take your mastermind analogy. It's like playing 100 games of mastermind at the same time (with the same goal). Make a move with random plays. Then take the 20 games that got the best scores last round. Replay the same move with a slight change (maybe swap positions of two pegs or change the color of a single peg) for those 20 boards, and act completely randomly in the other 80. Then you'll get GAs.
Sure, it's a refinement, but it's essentially random and pretty inefficient. Mastermind shouldn't be played with GAs; not all problems lend themselves to solution through genetic algorithms.
A paragraph about reinforcement learning from Reinforcement Learning: An Introduction, an excellent book on the subject by Rich Sutton and Andrew Barto (much of which is online):
Reinforcement learning is learning what to do---how to map situations to actions---so as to maximize a numerical reward signal. The learner is not told which actions to take, as in most forms of machine learning, but instead must discover which actions yield the most reward by trying them. In the most interesting and challenging cases, actions may affect not only the immediate reward, but also the next situation and, through that, all subsequent rewards. These two characteristics---trial-and-error search and delayed reward---are the two most important distinguishing features of reinforcement learning.
Well, I don't know about giving up on the theory of evolution entirely, but even accepting evolution, that statement is badly out of whack. Do the authors think that 1)early humans appeared on a savannah and were able to take a single step before falling, 2) had children who were slightly better,... and 3) later humans were thus able to walk? The analogy is so rough as to be wholly inaccurate.
This is one of the falling-down points of GAs: GAs create a bunch of random programs until one of them does roughly the right thing, instead of a single program that adapts itself based on the results it gets. Most skills aren't evolved, they're learned. GAs have a place, but not as big a place as they seem to get.
I guess this means that RH is going to focus almost entirely on B2B sales. I would imagine that the sales of the box sets barely offset the cost of marketing, etc. If they focus on business customers only, it probably simplifies their service commitments. I wonder if they'll sell service contracts and docs directly on the website, or if they'll just make the ISOs available and leave it at that.
Matt
Re:Just wait for the game with this feature...
on
Mutating Animations
·
· Score: 1
There are a lot of gamers -- if all the experience is combined (and transmitted to one place), you can probably get 20 player-years in an hour or two.
But don't use GAs. Wrong tool for the job here. Maybe look into reinforcement learning.
That may be. NNs aren't really my area, but my handy-dandy Russell and Norvig says
Marvin Minsky and Dean Edmonds built the first neural network computer in 1951.... Minksy was later to prove influential theorems showing the limitations of neural network research.
This may be the theorem that single-layer feed-forward NNs can only represent linearly separable functions (e.g. AND and OR, but not XOR).
Well that's an interesting point of view. I guess voice recognition software and translation software don't exist, to name a couple of prevalent (if imperfect) examples. Or were they around in the 40s?
AI has come a long way. For instance, expert systems have become useful tools in many fields. Machine learning techniques have been used to improve elevator control algorithms. In fact, just the other day, there was a story on/. in which people were arguing that AI that takes video as input and uses that to track cars, people, etc. is not far away.
There's an interesting quote at the end of a
Wired article in which Marvin Minsky (one of the early AI giants) argues that AI hasn't improved since the 70s (though I think he's defining AI as a very narrow problem):
"As soon as we solve a problem," said [Professor Martha] Pollack, "instead of looking at the solution as AI, we come to view it as just another computer system."
(Prof. Pollack is a professor of AI at the University of Michigan. FULL DISCLOSURE: I (sometimes) work for Prof. Pollack).
I think you're running into that trap. AI has come a long way. Sure, some problems have had relatively little progress (e.g. an AI that is a "total person" or at least one that passes the Turing test), but researchers in various areas--planning, learning, etc. have made great strides.
Some good points about visiting the correct website. But I think you underestimate the difficulty of shipping internationally.
First, you must be able to ship there. OK, maybe that's no problem, though the unit cost'll be pretty high and insurance may be necessary (and also pretty expensive) because of package loss problems mentioned in other threads.
Second, you must know what duties and taxes to pay on the shipment. And you'll have to keep up as tariffs change. This must be done for each country that you want to ship to.
Third, warranties: If you're a retailer like Amazon, you may not have much say in removing warranties. Additionally, different countries have different laws about how much warranty is inherent (as do different states in the U.S., and for that matter, probably different states, provinces, federal districts, etc. in other countries). So you'll need a legal framework (for each country/region/province/whatever, mind you) to handle the finer points of warranty law; this must also be kept up to date with the most recent laws.
I'm sure this list is not exhaustive.
The volume is likely to be much lower for overseas shipments than domestic shipments, especially for small companies, meaning that the overhead in two and three can't really be rolled into the unit shipping costs without pricing yourself right out of reasonability.
It's one thing if you've got a presence in the country. But if you don't, it's probably not that reasonable to try to ship to that country.
Way to jump rails there, chief. Your examples of secret evidence against Afghanistan and Iraq are qualitatively different than this situation as those were not court cases.
In this case, the evidence is not being disclosed to the public because SCO calls it a trade secret. IANAL, but I'm fairly certain that evidence is often withheld from the public in trade secret lawsuits, because releasing the evidence does more damage than one company using the trade secret.
That doesn't mean that I don't think SCO's going about this in an underhanded way, it's just a comment that your post has little to do with reality.
First, point by point: 4) As he's mentioned before and I thought was clear from context, he meant physical ports. That is, if someone's not paying, they don't get service.
6) Setting up a cache is kind of pointless if it's optional. Set up a transparent cache--there's plenty of caches-in-a-box out there that do the job nicely and work. If you make a cache optional, people won't use it, and you're wasting upstream bandwidth.
7)/8) He doesn't say anything about blocking 25. He says scan incoming mail for viruses. This is reasonable and implemented in many ISPs, in particular many universities.
9) I'm not sure why he does that. He mentions elsewhere that spam can come in on these ports, and someone else has suggested the better solution of blocking the ports at the border but allowing inter-unit sharing.
13) Have you not been paying attention at all to wireless? It's wide open by default in most APs, and most users don't know enough to make it right. Allowing WAPs willy-nilly breaks a hole wide open in your network.
On a holistic note, way to assume the worst and use "clever" and "biting" remarks to "refute" the original poster's arguments. He makes reasonable suggestions for the most part, and you jump on his case. Way to contribute to society.
Amusing link. I especially like this quote: "Over the past year, a movement dubbed "Open Source" has gained in popularity throughout Europe and is now rearing its head in the United States."
Yeah, before '02, I had never even heard of Open Source.
But they do make an interesting point, even if they may use FUD about OSS to make it: When government gives preference to OSS simply because it is open source, its citizens lose. But neither should the commericial software get preference. Fair competition and a case-by-case judgment of the best tool for the job should be used.
You misunderstood my post. My point is, that type of broad statement is silly. The broad statement that one should not invent/use a technology that/could/ be used for some nefarious purpose solely for that reason is wrong. I tried to point this out by using the same sentence structure to make an argument that most of the Slashdot crowd would agree is silly.
My point was that either Schneier had more context to his quote, or that he was just plain wrong, and that that argument should not be used to decide whether or not DRM is supported. Linus points out some valid uses for the technology, and those should not be precluded just because there are invalid uses.
Well, then I'm another exception. I'm working on a Ph.D. in computer science, and it's in large part because of my teachers in (public) high school (and, of course, my parents). By and large, the teachers I had for AP classes encouraged me to grow and learn (cheesy, I know). My aptitude for learning and abstraction and self-motivation don't hurt, but these don't suddenly appear one day.
My parents tended towards letting me make my own decisions by high school, but by that time they had guided me toward the right track. Anecdote: When I was in elementary school (maybe 2nd grade), my penmanship was awful (remember that at that age, penmanship is an important part of school). My mother made me practice daily until I improved.
To make a long story short, parental supervision is necessary for child development; just how much supervision is at the discretion of the parents.
Second definition of "hacker" from dictionary.com: "2. One who uses programming skills to gain illegal access to a computer network or file."
First definition of "doctor" from dictionary.com: "5. A practitioner of folk medicine or folk magic."
Just because a word has a common usage doesn't mean that it is a precisely correct definition of the word. I think that most people will agree that practitioners of folk yadda yadda shouldn't hand out business cards that proclaim them doctors. Programmers can call themselves engineers amongst themselves, but shouldn't advertise themselves as engineers without more justification than the fact they can program.
Software engineers do exist. There should be an accrediting board that identifies those people. Someone in another thread mentioned the ACM; that seems like a good candidate, and perhaps others could be put forward.
I definitely agree with this breakdown. Just as not everyone who works on hardware is an electrical engineer (e.g. technicians), not everyone who works on software is an engineer.
I have a degree in electrical engineering, and I've seen the curriculum of undergraduate computer scientists at my alma mater. I would say that CS degrees create the potential for software engineering just as EE degrees create the potential for electrical engineering--the courses provide the framework but do not an engineer make. There's a reason that PE licenses require some work experience as well as passing a test--experience in the field, following correct processes, etc, is a necessity to create an Engineer (sic).
Just as a note, there is some engineering going on in research institutions as well. I'm currently in the Computer Science graduate program at a research university, and some of the applications research (applying new ideas to particular problems) involves engineering as much as science.
The two trolls look remarkably similar in facial structure, etc, which I found interesting, but the LOTR troll was more vicious. I would actually say that the harry potter troll seemed smarter than the one in LOTR. Still, considerably less lethal-seeming, even if both trolls came out with the same death toll in the end.
I seriously doubt Dumbledore ever threw down (and down and down...) with an evil being of shadow and fire and came out on the not-dead side. Nuff said.
I don't know the prices/loads involved, but I would guess that they win on this one with building their own. It might be another story if the only choice is a multi-million dollar supercomputer, and definitely if they couldn't keep it busy. But this is a quarter-million dollar machine with plenty of MIPS that I'm sure will be used--Los Alamos astrophysicists probably can come up with enough simulations to fill the queue.
So the question is, I suppose, what is Intel's/IBM's/whoever's cost per cycle per second compared with the long-term costs associated with this machine--node replacement, maintenance personnel, etc? Again, although I don't know the numbers, a guess would put the Beowulf cluster cheaper, as the cycle sellers are looking to compete with more expensive supercomputers.
You're kidding, right? Providing default settings is a software practice you may be unfamiliar with, but it has provided great results in one or two projects.
Applying this practice to, say, Windows, would result in everyone's PC looking the same straight from the factory; the people who can care can mess with the config files (in Windows? ha!) or applets or whatever. Just because those 95% never touch anything doesn't mean it's a loss for them.
It actually ends up being a win, because those users don't have to depend on all software developers being able to accurately copy other programs' GUIs to provide a similar experience; it just will out of the box.
Slashdot Mirror
Apples and oranges. Linux doesn't revolutionize the desktop, as automobiles revolutionized transportation. Linux's big selling point is that it's cheaper than Windows. If it can't interoperate or be used without more training or something of this nature, then the price advantage disappears.
Besides, the automobile took some time before it caught on everywhere--horses were still used for some purposes in WWI, and I'm sure the army wasn't the only one.
Matt
Yes, I can see that meeting now:
CUSTOMER: Well, we're having trouble making our new Linux boxes talk to our large installed base of Windows boxes.
MICROSOFT: You don't say.
C: No, really. We'd really like it if you were to make Windows boxes easier to talk to by publishing your heretofore closed standard.
M: So let me get this straight. You want us to make it easier for our competitors to replace us? If we do this, then our market share goes down because the barrier to switching lowers. But if we don't, you'll keep buying Windows because it's cheaper than doing a complete rollover. Let me consider that.
C: Thanks, we'd really appreciate it!
Matt
Oops, much too long without refreshing :)
I'm not sure about the number of jurors, but perhaps more important is that the jurors only have to decide if the preponderance of evidence points to violation (that is, are you 50% sure the defendant is guilty), rather than beyond a reasonable doubt, as in criminal cases.
I think you misread the parent. I think the author says that this is a standard that corporate world does not necessarily adhere to, and thus it is in everyone's best interest to keep collection at the IRS.
My question is, why give the debt collectors all that information? All they need is a name, phone number, address, and the amount owed. No need to give them your entire tax history. So data privacy concerns are sort of bogus.
Matt
Yeah, it kinda does mean that. SCO owns the rights, not the management. If we all buy a share of stock (13.1M shares outstanding, but linuxcounter tells us there are ~18M users), then WE own the company and thus the rights to Unix that SCO has.
Not that I'm suggesting this is a good course of action, but there you go.
Matt
Yes, the initial set is random. And if you read the article, you'd see that each successive generation of this particular formulation contains 80% completely random programs. This is not uncommon. In addition, the 20% that come from the previous generation are randomly mutated. There's no analysis of the program to determine which pieces of the program worked and which didn't. Even in instances of "sexual" GA, the mating between two programs takes a randomly selected piece of one and joins it with a randomly selected piece of the other.
Take your mastermind analogy. It's like playing 100 games of mastermind at the same time (with the same goal). Make a move with random plays. Then take the 20 games that got the best scores last round. Replay the same move with a slight change (maybe swap positions of two pegs or change the color of a single peg) for those 20 boards, and act completely randomly in the other 80. Then you'll get GAs.
Sure, it's a refinement, but it's essentially random and pretty inefficient. Mastermind shouldn't be played with GAs; not all problems lend themselves to solution through genetic algorithms.
Matt
A paragraph about reinforcement learning from Reinforcement Learning: An Introduction, an excellent book on the subject by Rich Sutton and Andrew Barto (much of which is online):
Well, I don't know about giving up on the theory of evolution entirely, but even accepting evolution, that statement is badly out of whack. Do the authors think that 1)early humans appeared on a savannah and were able to take a single step before falling, 2) had children who were slightly better, ... and 3) later humans were thus able to walk? The analogy is so rough as to be wholly inaccurate.
This is one of the falling-down points of GAs: GAs create a bunch of random programs until one of them does roughly the right thing, instead of a single program that adapts itself based on the results it gets. Most skills aren't evolved, they're learned. GAs have a place, but not as big a place as they seem to get.
MattI guess this means that RH is going to focus almost entirely on B2B sales. I would imagine that the sales of the box sets barely offset the cost of marketing, etc. If they focus on business customers only, it probably simplifies their service commitments. I wonder if they'll sell service contracts and docs directly on the website, or if they'll just make the ISOs available and leave it at that.
Matt
There are a lot of gamers -- if all the experience is combined (and transmitted to one place), you can probably get 20 player-years in an hour or two.
But don't use GAs. Wrong tool for the job here. Maybe look into reinforcement learning.
Matt
This may be the theorem that single-layer feed-forward NNs can only represent linearly separable functions (e.g. AND and OR, but not XOR).
Matt
Well that's an interesting point of view. I guess voice recognition software and translation software don't exist, to name a couple of prevalent (if imperfect) examples. Or were they around in the 40s?
AI has come a long way. For instance, expert systems have become useful tools in many fields. Machine learning techniques have been used to improve elevator control algorithms. In fact, just the other day, there was a story on /. in which people were arguing that AI that takes video as input and uses that to track cars, people, etc. is not far away.
There's an interesting quote at the end of a Wired article in which Marvin Minsky (one of the early AI giants) argues that AI hasn't improved since the 70s (though I think he's defining AI as a very narrow problem):
(Prof. Pollack is a professor of AI at the University of Michigan. FULL DISCLOSURE: I (sometimes) work for Prof. Pollack).
I think you're running into that trap. AI has come a long way. Sure, some problems have had relatively little progress (e.g. an AI that is a "total person" or at least one that passes the Turing test), but researchers in various areas--planning, learning, etc. have made great strides.
Matt
Oops, got beat by someone who know more.
Some good points about visiting the correct website. But I think you underestimate the difficulty of shipping internationally.
First, you must be able to ship there. OK, maybe that's no problem, though the unit cost'll be pretty high and insurance may be necessary (and also pretty expensive) because of package loss problems mentioned in other threads.
Second, you must know what duties and taxes to pay on the shipment. And you'll have to keep up as tariffs change. This must be done for each country that you want to ship to.
Third, warranties: If you're a retailer like Amazon, you may not have much say in removing warranties. Additionally, different countries have different laws about how much warranty is inherent (as do different states in the U.S., and for that matter, probably different states, provinces, federal districts, etc. in other countries). So you'll need a legal framework (for each country/region/province/whatever, mind you) to handle the finer points of warranty law; this must also be kept up to date with the most recent laws.
I'm sure this list is not exhaustive.
The volume is likely to be much lower for overseas shipments than domestic shipments, especially for small companies, meaning that the overhead in two and three can't really be rolled into the unit shipping costs without pricing yourself right out of reasonability.
It's one thing if you've got a presence in the country. But if you don't, it's probably not that reasonable to try to ship to that country.
Matt
Way to jump rails there, chief. Your examples of secret evidence against Afghanistan and Iraq are qualitatively different than this situation as those were not court cases.
In this case, the evidence is not being disclosed to the public because SCO calls it a trade secret. IANAL, but I'm fairly certain that evidence is often withheld from the public in trade secret lawsuits, because releasing the evidence does more damage than one company using the trade secret.
That doesn't mean that I don't think SCO's going about this in an underhanded way, it's just a comment that your post has little to do with reality.
Matt
First, point by point:
4) As he's mentioned before and I thought was clear from context, he meant physical ports. That is, if someone's not paying, they don't get service.
6) Setting up a cache is kind of pointless if it's optional. Set up a transparent cache--there's plenty of caches-in-a-box out there that do the job nicely and work. If you make a cache optional, people won't use it, and you're wasting upstream bandwidth.
7)/8) He doesn't say anything about blocking 25. He says scan incoming mail for viruses. This is reasonable and implemented in many ISPs, in particular many universities.
9) I'm not sure why he does that. He mentions elsewhere that spam can come in on these ports, and someone else has suggested the better solution of blocking the ports at the border but allowing inter-unit sharing.
13) Have you not been paying attention at all to wireless? It's wide open by default in most APs, and most users don't know enough to make it right. Allowing WAPs willy-nilly breaks a hole wide open in your network.
On a holistic note, way to assume the worst and use "clever" and "biting" remarks to "refute" the original poster's arguments. He makes reasonable suggestions for the most part, and you jump on his case. Way to contribute to society.
Matt
Amusing link. I especially like this quote: "Over the past year, a movement dubbed "Open Source" has gained in popularity throughout Europe and is now rearing its head in the United States."
Yeah, before '02, I had never even heard of Open Source.
But they do make an interesting point, even if they may use FUD about OSS to make it: When government gives preference to OSS simply because it is open source, its citizens lose. But neither should the commericial software get preference. Fair competition and a case-by-case judgment of the best tool for the job should be used.
You misunderstood my post. My point is, that type of broad statement is silly. The broad statement that one should not invent/use a technology that /could/ be used for some nefarious purpose solely for that reason is wrong. I tried to point this out by using the same sentence structure to make an argument that most of the Slashdot crowd would agree is silly.
My point was that either Schneier had more context to his quote, or that he was just plain wrong, and that that argument should not be used to decide whether or not DRM is supported. Linus points out some valid uses for the technology, and those should not be precluded just because there are invalid uses.
Matt
Is it then poor civil hygiene to install technologies that could someday be used to circumvent laws?
Either that statement has more context or it is silly.
Matt
Well, then I'm another exception. I'm working on a Ph.D. in computer science, and it's in large part because of my teachers in (public) high school (and, of course, my parents). By and large, the teachers I had for AP classes encouraged me to grow and learn (cheesy, I know). My aptitude for learning and abstraction and self-motivation don't hurt, but these don't suddenly appear one day.
My parents tended towards letting me make my own decisions by high school, but by that time they had guided me toward the right track. Anecdote: When I was in elementary school (maybe 2nd grade), my penmanship was awful (remember that at that age, penmanship is an important part of school). My mother made me practice daily until I improved.
To make a long story short, parental supervision is necessary for child development; just how much supervision is at the discretion of the parents.
Matt
Second definition of "hacker" from dictionary.com: "2. One who uses programming skills to gain illegal access to a computer network or file."
First definition of "doctor" from dictionary.com: "5. A practitioner of folk medicine or folk magic."
Just because a word has a common usage doesn't mean that it is a precisely correct definition of the word. I think that most people will agree that practitioners of folk yadda yadda shouldn't hand out business cards that proclaim them doctors. Programmers can call themselves engineers amongst themselves, but shouldn't advertise themselves as engineers without more justification than the fact they can program.
Software engineers do exist. There should be an accrediting board that identifies those people. Someone in another thread mentioned the ACM; that seems like a good candidate, and perhaps others could be put forward.
Matt
I definitely agree with this breakdown. Just as not everyone who works on hardware is an electrical engineer (e.g. technicians), not everyone who works on software is an engineer.
I have a degree in electrical engineering, and I've seen the curriculum of undergraduate computer scientists at my alma mater. I would say that CS degrees create the potential for software engineering just as EE degrees create the potential for electrical engineering--the courses provide the framework but do not an engineer make. There's a reason that PE licenses require some work experience as well as passing a test--experience in the field, following correct processes, etc, is a necessity to create an Engineer (sic).
Just as a note, there is some engineering going on in research institutions as well. I'm currently in the Computer Science graduate program at a research university, and some of the applications research (applying new ideas to particular problems) involves engineering as much as science.
Matt
The two trolls look remarkably similar in facial structure, etc, which I found interesting, but the LOTR troll was more vicious. I would actually say that the harry potter troll seemed smarter than the one in LOTR. Still, considerably less lethal-seeming, even if both trolls came out with the same death toll in the end.
I seriously doubt Dumbledore ever threw down (and down and down...) with an evil being of shadow and fire and came out on the not-dead side. Nuff said.
I don't know the prices/loads involved, but I would guess that they win on this one with building their own. It might be another story if the only choice is a multi-million dollar supercomputer, and definitely if they couldn't keep it busy. But this is a quarter-million dollar machine with plenty of MIPS that I'm sure will be used--Los Alamos astrophysicists probably can come up with enough simulations to fill the queue.
So the question is, I suppose, what is Intel's/IBM's/whoever's cost per cycle per second compared with the long-term costs associated with this machine--node replacement, maintenance personnel, etc? Again, although I don't know the numbers, a guess would put the Beowulf cluster cheaper, as the cycle sellers are looking to compete with more expensive supercomputers.
Matt
You're kidding, right? Providing default settings is a software practice you may be unfamiliar with, but it has provided great results in one or two projects.
Applying this practice to, say, Windows, would result in everyone's PC looking the same straight from the factory; the people who can care can mess with the config files (in Windows? ha!) or applets or whatever. Just because those 95% never touch anything doesn't mean it's a loss for them.
It actually ends up being a win, because those users don't have to depend on all software developers being able to accurately copy other programs' GUIs to provide a similar experience; it just will out of the box.
Matt