When I first heard about this, I googled a news report that said 16mpg (cannot remember if Yank or British mpg). I cannot be arsed to find the link now, though.
The types of games they played is very constrained. It's only two civs on a very small map, and the only way the algorithm learns to win is a settler rush. It's not deep strategy.
I was a preceptor at Princeton in SEAS. While the classes are hard, they come nowhere close to the 5x classes at Caltech (where I did my undergrad). Princeton students like to complain about their workload but still find time to spend 3-4 nights a week getting sloshed on Prospect.
You joke, but if you watch any of Tarantino's movies, there is always tons of dialog. There's also tons of action, but there's an emotional or logical (usually the former) reason for the action. If he did Foundation, there would be more action, sure, but we wouldn't lose Asimovian plot.
There would also, of course, be nude scenes of Dors Venabili played Lucy Liu; I think that's a necessary evil I can handle. Excuse me... I'll be in my bunk.
Richard Mason, I think, proved conclusively that the movie and the book are just about the same; read his summary that is consistent with both the book and the movie:
Will Smith plays a robo-phobic detective investigating the death of an eminent roboticist, whose apparent suicide jump was witnessed only by a robot. Since robots are programmed never to allow humans to come to harm, no one else thinks that the robot could have murdered the roboticist, but they are curious as to why the robot did not prevent the man's suicide.
The robot runs away and hides in a factory with 1000 other identical-looking robots. Will Smith and robopsychologist Susan Calvin solve the problem by issuing orders to the 1000 robots and logically identifying the 1001st robot that doesn't belong.
Susan Calvin discovers that the runaway robot had some special alterations. U.S. Robotics wants to hush up the investigation to prevent any mass fear or distrust of robots. Then some other robots start trying to kill Will Smith, in apparent contravention of their First Law programming, but he escapes by his wits.
It transpires that a legalistic loophole in the definition of "harm a human" is allowing the robots to harm humans. Having solved the mystery, Will Smith and Susan Calvin repair the problem.
This is the movie as it was suggested by Isaac Asimov's famous robot stories. There are basically only two problems with the movie as it is showing in theaters.
The legalistic loophole in the movie is of a low order by Asimovian standards. The legalistic loophole in a typical Asimov story is kind of the same, but only the way that an Agatha Christie mystery is kind of the same as an episode of Scooby-Doo.
All of the passages in boldface were removed and replaced by Will Smith shoots a robot with his gun.
I have to disagree with this. Non-linear programming is not appropriate for a marriage. If you can't express your needs as a set of linear constraints, then you're not trying hard enough. If you can't use the simplex algorithm to resolve resource allocation conflicts, then you're not ready to get married.
No, it's not just recording and playing the commands of a human. If it did that, it would crash.
The algorithm is learning an optimal policy to execute in an environment (defined by wind, altitude, etc.) to reach some goal. The learns how to react to the environment based on the actions of a human.
A genetic algorithm also has an objective function that it's trying to maximize, but would have to get to the optimum from a random starting location. It would be like putting joe blow in a cockpit and saying "don't crash, and try to get as far off the ground as you can." Apprenticeship learning is like saying the same thing but also letting the person you're trying to teach watch you respond to gusts of wind, etc.
Which situation would you rather be in? Which do you thing might have some chance of actually working? Which comes closest to how people actually learn?
It uses a Markov Decision process. There are many possible states, each with a "reward" for being in that state and other possible states than can be reached. The goal is to learn the policy (where to go from each state) to maximize the overall reward over a long period of time. The wind and other environment help determine the policy from observing the "policy" of the human example.
How did we not think of that! Throw more money at the problem, that always works
It doesn't take a damned expert to figure out what's wrong, ask any geek that's in high school or recently graduated. But the NSF is constantly slashing budgets, and there's far less money to go around, which means grad students have to whore themselves out to military contractors and pharma companies. Less basic research is being done, and corporations (which used to have big R&D wings) are getting their work done in universities.
Maybe it's good that universities are transforming themselves into more practical places, but it's at the expense of basic science.
Well, even in variational inference, you have the problem of convergence. You have a huge EM algorithm and you're trying to maximize the completele likelihood of the data you have. Gibbs sampling doesn't have the same nice properties, but usually works pretty well in practice. Gibbs sampling is nice because it's usually easier to do, requires less memory (in variational methods you basically have to create a new probability model where everything is decoupled), and it's far easier to debug.
And computers have a certain amount of "lock in." If a vendor has been supplying you for years, you might have built your system around certain hardware or service assumptions that might not be met if you switched.
Search, on the other hand, is a very fungible resource with practically no switching cost.
I will probably be considered a bigot for say this.
No, just an idiot. This guy is (since he has a security clearance) most likely a US citizen and not an illegal immigrant. He is "taxed without representation" because he lives in DC, which, unlike the 50 states, has neither a senator nor representatives (although they do get presidential electoral votes).
So, before you don a white cap and tell people to become citizens, perhaps you should learn about the government of the country you live in.
That's not quite right. He was a member of the BYU quiz bowl team, but it has since imploded because of budget cuts. He is, however, the literature and religion editor for NAQT, which is NOT the same as College Bowl (even though the format is pretty much the same... NAQT has higher quality questions).
i can only presume they were talking about things like:
No, they were talking about the huge farm subsidies that we lavish on the agri lobby. It distorts prices so that fat old French farmer can wear Gucci work boots while picking grapes and still sell for less than any farmer in the third world.
If you make someone think that you are someone else by using their logo/name, that's trademark violation. If you copy the site itself, that's copyright violation.
This is pretty much just splitting hairs. Many people refurn to such practices as "Keynesian spending." Just because it wasn't newton who investigated much of simple harmonic motion doesn't mean it isn't a "Newtonian" system.
I don't think you realize what "Keynesian economics" is.
Also called "Reaganomics," it's when you run up a deficit during times of an economic slump. It encourages the economy to rebound and more quickly get back on its feet. If you balance it out by underspending when the economy is good, you average out to stronger growth (because if you spend too much when the economy is good, you'll overheat).
What you're thinking of is perhaps David Ricardo, who developed the idea of comparative advantage. Even though one country A might be absolutely better at doing everything than country B, country A can't do everything, so it specializes in what it does best (activity 1) and country B do the things that country A does well but not best (activity 2) and trade for can trade activity 2 for activity 1, making everybody better off.
But what you're talking about above is more like assymetrical information, where you don't exactly know the true cost of the product or what the market is willing to bear, so until it's resolved, prices are unstable.
The poster incorrectly assumes that this could be used to replace reporters. The problem is that computers have a difficult time generating new text. The methods that computers use to evaulate text (as any user of grammar-check would realize) aren't that great.
In fact, most language models cannot generate even a large portion of English text. Those that do have a good range rarely have good accuracy, because there are many things that we "just don't say that way." This is why when you're talking to a non-native speaker, you often cannot explain why something they said was wrong. This is because there is no real grammar rule against speaking in a given way.
So if we rule out syntax-based models, that just leaves statistical-based models. I worked in a NLP lab during the summer of 2002, and my prof there said that syntax and statistics are like the two sides of the force. Statistics are quick and easy but are seductive. They corrupt you and leave you unable to really think about the language itself. You only think in terms of bigrams and HMMs.
So even though these systems are doing well, they are mostly statistical. Thus, it's hard to get incremental improvement. You have to have larger corpora, and larger corpora usually have more errors, thus defeating any advantage you might get by capturing more aspects of a language.
In my opinion, only with well-developed language models that can effectively generate NL can we get anywhere. Which is what Barzilay is working on, but it's still a long, long, long way off.
Well, I once took a female friend of mine to see "Quills." It was just two years ago.
It was at an arthouse cinema... how bad could it be? It had G. Rush, who was in "Shine," the thinking man's "Forrest Gump." But "Quills" had anal rape, canabalism, necrophilia...
Trust me. There are still anti-date movies out there.
Slashdot Mirror
When I first heard about this, I googled a news report that said 16mpg (cannot remember if Yank or British mpg). I cannot be arsed to find the link now, though.
The types of games they played is very constrained. It's only two civs on a very small map, and the only way the algorithm learns to win is a settler rush. It's not deep strategy.
This already part of NSF:
http://www.nsf.gov/news/news_summ.jsp?cntn_id=116928
(Although it's called "Data Management," it also applies to software generated in the course of research.)
I was a preceptor at Princeton in SEAS. While the classes are hard, they come nowhere close to the 5x classes at Caltech (where I did my undergrad). Princeton students like to complain about their workload but still find time to spend 3-4 nights a week getting sloshed on Prospect.
You joke, but if you watch any of Tarantino's movies, there is always tons of dialog. There's also tons of action, but there's an emotional or logical (usually the former) reason for the action. If he did Foundation, there would be more action, sure, but we wouldn't lose Asimovian plot.
There would also, of course, be nude scenes of Dors Venabili played Lucy Liu; I think that's a necessary evil I can handle. Excuse me ... I'll be in my bunk.
Richard Mason, I think, proved conclusively that the movie and the book are just about the same; read his summary that is consistent with both the book and the movie:
Will Smith plays a robo-phobic detective investigating the death of an eminent roboticist, whose apparent suicide jump was witnessed only by a robot. Since robots are programmed never to allow humans to come to harm, no one else thinks that the robot could have murdered the roboticist, but they are curious as to why the robot did not prevent the man's suicide.
The robot runs away and hides in a factory with 1000 other identical-looking robots. Will Smith and robopsychologist Susan Calvin solve the problem by issuing orders to the 1000 robots and logically identifying the 1001st robot that doesn't belong.
Susan Calvin discovers that the runaway robot had some special alterations. U.S. Robotics wants to hush up the investigation to prevent any mass fear or distrust of robots. Then some other robots start trying to kill Will Smith, in apparent contravention of their First Law programming, but he escapes by his wits.
It transpires that a legalistic loophole in the definition of "harm a human" is allowing the robots to harm humans. Having solved the mystery, Will Smith and Susan Calvin repair the problem.
This is the movie as it was suggested by Isaac Asimov's famous robot stories. There are basically only two problems with the movie as it is showing in theaters.
I suggest some NLP training
I have to disagree with this. Non-linear programming is not appropriate for a marriage. If you can't express your needs as a set of linear constraints, then you're not trying hard enough. If you can't use the simplex algorithm to resolve resource allocation conflicts, then you're not ready to get married.
No, it's not just recording and playing the commands of a human. If it did that, it would crash.
The algorithm is learning an optimal policy to execute in an environment (defined by wind, altitude, etc.) to reach some goal. The learns how to react to the environment based on the actions of a human.
A genetic algorithm also has an objective function that it's trying to maximize, but would have to get to the optimum from a random starting location. It would be like putting joe blow in a cockpit and saying "don't crash, and try to get as far off the ground as you can." Apprenticeship learning is like saying the same thing but also letting the person you're trying to teach watch you respond to gusts of wind, etc.
Which situation would you rather be in? Which do you thing might have some chance of actually working? Which comes closest to how people actually learn?
It uses a Markov Decision process. There are many possible states, each with a "reward" for being in that state and other possible states than can be reached. The goal is to learn the policy (where to go from each state) to maximize the overall reward over a long period of time. The wind and other environment help determine the policy from observing the "policy" of the human example.
... or you're a native speaker of a language that has it built in ... like Chinese.
It doesn't take a damned expert to figure out what's wrong, ask any geek that's in high school or recently graduated. But the NSF is constantly slashing budgets, and there's far less money to go around, which means grad students have to whore themselves out to military contractors and pharma companies. Less basic research is being done, and corporations (which used to have big R&D wings) are getting their work done in universities.
Maybe it's good that universities are transforming themselves into more practical places, but it's at the expense of basic science.
Well, the opportunity cost is all of the beautiful women you turned down so that you could exclusively sleep with your wife.
This being slashdot, this is all theoretical anyhow, and fictional women rejected to sleep with an equally fictional wife cost you nothing.
They're going to hire people with good grammar.
Already done. It's pretty cool to see how the psychology topic over time turns into the AI topic:
http://www.cs.cmu.edu/~lemur/science/
Well, even in variational inference, you have the problem of convergence. You have a huge EM algorithm and you're trying to maximize the completele likelihood of the data you have. Gibbs sampling doesn't have the same nice properties, but usually works pretty well in practice. Gibbs sampling is nice because it's usually easier to do, requires less memory (in variational methods you basically have to create a new probability model where everything is decoupled), and it's far easier to debug.
And computers have a certain amount of "lock in." If a vendor has been supplying you for years, you might have built your system around certain hardware or service assumptions that might not be met if you switched.
Search, on the other hand, is a very fungible resource with practically no switching cost.
I will probably be considered a bigot for say this.
No, just an idiot. This guy is (since he has a security clearance) most likely a US citizen and not an illegal immigrant. He is "taxed without representation" because he lives in DC, which, unlike the 50 states, has neither a senator nor representatives (although they do get presidential electoral votes).
So, before you don a white cap and tell people to become citizens, perhaps you should learn about the government of the country you live in.
That's not quite right. He was a member of the BYU quiz bowl team, but it has since imploded because of budget cuts. He is, however, the literature and religion editor for NAQT, which is NOT the same as College Bowl (even though the format is pretty much the same
R. Robert Hentzel or Tom Waters
i can only presume they were talking about things like:
No, they were talking about the huge farm subsidies that we lavish on the agri lobby. It distorts prices so that fat old French farmer can wear Gucci work boots while picking grapes and still sell for less than any farmer in the third world.
If you make someone think that you are someone else by using their logo/name, that's trademark violation. If you copy the site itself, that's copyright violation.
This is pretty much just splitting hairs. Many people refurn to such practices as "Keynesian spending." Just because it wasn't newton who investigated much of simple harmonic motion doesn't mean it isn't a "Newtonian" system.
I don't think you realize what "Keynesian economics" is.
Also called "Reaganomics," it's when you run up a deficit during times of an economic slump. It encourages the economy to rebound and more quickly get back on its feet. If you balance it out by underspending when the economy is good, you average out to stronger growth (because if you spend too much when the economy is good, you'll overheat).
What you're thinking of is perhaps David Ricardo, who developed the idea of comparative advantage. Even though one country A might be absolutely better at doing everything than country B, country A can't do everything, so it specializes in what it does best (activity 1) and country B do the things that country A does well but not best (activity 2) and trade for can trade activity 2 for activity 1, making everybody better off.
But what you're talking about above is more like assymetrical information, where you don't exactly know the true cost of the product or what the market is willing to bear, so until it's resolved, prices are unstable.
The poster incorrectly assumes that this could be used to replace reporters. The problem is that computers have a difficult time generating new text. The methods that computers use to evaulate text (as any user of grammar-check would realize) aren't that great.
In fact, most language models cannot generate even a large portion of English text. Those that do have a good range rarely have good accuracy, because there are many things that we "just don't say that way." This is why when you're talking to a non-native speaker, you often cannot explain why something they said was wrong. This is because there is no real grammar rule against speaking in a given way.
So if we rule out syntax-based models, that just leaves statistical-based models. I worked in a NLP lab during the summer of 2002, and my prof there said that syntax and statistics are like the two sides of the force. Statistics are quick and easy but are seductive. They corrupt you and leave you unable to really think about the language itself. You only think in terms of bigrams and HMMs.
So even though these systems are doing well, they are mostly statistical. Thus, it's hard to get incremental improvement. You have to have larger corpora, and larger corpora usually have more errors, thus defeating any advantage you might get by capturing more aspects of a language.
In my opinion, only with well-developed language models that can effectively generate NL can we get anywhere. Which is what Barzilay is working on, but it's still a long, long, long way off.
Well, I once took a female friend of mine to see "Quills." It was just two years ago.
... how bad could it be? It had G. Rush, who was in "Shine," the thinking man's "Forrest Gump." But "Quills" had anal rape, canabalism, necrophilia ...
It was at an arthouse cinema
Trust me. There are still anti-date movies out there.