The most ridicule part being "must not be able to explain how". That doesn't even make sense for humans! If you ask artists, they'll tell you what their influences are, if you ask critics, they'll tell you why this particular piece of art was made this way and not in a completely different manner.
Fun fact: any program with yet unseen bugs that make their behavior totally unexplainable to their developers has passed the test. That gives you either an idea of the soundness of this crap, or a deep insight of what type of failure humankind is.
It depends of what you expect from an AI. If it is a perfect replica of a human mind, with which you can talk and share life as if it were human, then it will probably never be around. But that's also pretty useless, and most development in machine learning (ML) are in a more abstract level than trying to solve a very specific goal like this.
Now if you consider AI to be completely new intelligent species, that behave in an intelligent way (volontary fuzzy definition here), then it's probably already there. I mean, the ML programs that dictate the behaviour of you insurance policies so as to send you sport ads when you're a bit overweight, or holidays at the sea when you're close to a burn out, that raise the price of things predicted to induce a loss and lower the ones of things predicted have big return in order to influence your choices, etc, that, to me, sounds exactly like what you do with your pets when you decide they should eat that instead of this for some reason they could not handle with their inferior minds. Now, if you think of all the interconnected ML programs searching for new optima every second and exchanging information, you can view it as the new superior species of this planet.
A very short example: the vast majority of the human race wants to put an end to automated short-sighted finance, just like the vast majority of dogs wants to get free from their leashes. Bot never will until their recpective superior species allow them to. We talked a lot of the facebook experiment lately, the real question is how long has it been already done by the machine to fulfill goals we are not able to grasp? Maybe the singularity is already there since a few years, and just like for peak oil, we'll know it some time after. If we get to notice something more intelligent than us is governing our lives.
Seriously, the poster was presented yesterday at CVPR and ends up on/. today. There is nothing sensational about it. I'm getting sick of science turning into PR stunts all the time.
I have only one problem with fancy GUI that allow you to train a predicting model in 2 clicks: how confident can you be in your model, since all the parameters are masked and you have no knowledge about them? I still think it is dangerous to rely on a tool you don't understand and you can't control up to a satisfactory level, especially when it is to be used in prediction - something we expect to be highly reliable in many aspects due to old development of science like balistics.
I've written a ML library myself (also in Java, more lightweight than weka, but with no gui - although it comes with standalone binaries for some basic setups) and I can tell you there is no good default tunning that works well for every kind of situation. ML is seriously a young science that gets rapidly tricky even on very common problems, which is very different than field for which we have very accurate solvers that work most of the time (again balistics is probably a good example, at least because it is taught in school and sets the prototype of what we name science). I fear hidding the youth (and thus the imperfection) is only going to cause damage through misconception and false interpretation.
Asking if robots can be evil is about as futile as asking if a microwave can be happy.
That being said, there already are killer robots, with a pretty good track record in recent operations. But the evil lies in the humans who made them (from the top exec that launch the program to the small hand that does the job) and used them, not in the pile of steel and semiconductors.
caveat: Looking at your food, your microwave is probably sad, which explains their tendency to commit suicide.
JMLR is a fantastic journal, with high quality papers, high quality reviews, and completely open. The dream of many come true. I've always wondered why the idea did not spread to other fields.
If someone is willing to start a Journal of Computer Vision Research based on the same principles, count me in. I'll be happy to do editing/reviewing for such journal instead of well known IEEE/Elsevier/Springer journals.
Given that mir also means peace and considering what is happening in Ukraine, I sadly suppose you are right. Nonetheless I prefer puns when they don't carry a grim presage.
I am pretty much waiting for the other way around: when I, as a human, known that I am quite bad at handling the current situation and that the machine will do much better than I'll ever be capable of. In particular, I am thinking of traffic jams where you have to find maximum speed that maintains a smooth global flow, without falling into chaotic start/stop sequences like this: http://www.youtube.com/watch?v...
Most humans typically tend to drive too fast in such situations, leading to an average speed well below what machines could very simply achieve. So basically, I'll drive when it's easy, and let the machine do the complicated stuff, as usual.
My wife and I are in the same situation, and I never understood the selfishness argument. Why is it selfish? To whom? What harm does it bring and to what?
The more I have this discussion with family and friends, the more it turns out to be pure jealousy towards us better enjoying our life. Most of them didn't expect it is that hard to raise children, and especially the many things you have to give up due to the lack of time to do it.
" just about every SSL-encrypted internet communication over the last two years has been compromised."
No, it really hasn't.
It's accurate to say that just about every Open-SSL encrypted session for servers that were using NEW versions of OpenSSL (not all those ones out there still stuck on 0.9.8(whatever) that never had the bug) were potentially vulnerable to attack.
That's bad, but it's a universe away from "every SSL session is compromized!!!" because that's not really true.
They were vulnerable to attack, that is to say, the security was compromised. He didn't say they were hacked, stolen, eavesdropped, or surreptitiously recorded.
No, not if they didn't enabled the compromised feature.
Yeah, that's also my reaction. I regularly do between 50 and 60 hrs a week by working 11 - 12 hrs a day the whole week (and nothing the week end cause I would otherwise go insane), I've been doing occasonially 70 hrs (that is, add 10 hrs during the WE), and I think my max was around 80 hrs for some relly tough deadline near the end of my PhD. Right now, I finished a hard period, and I'll be calming things down to around 40 hrs a week in the few next month to regain some health. Seriously, 60 is hard, around 70 is just insane, and over that is ruining your health more quickly than anything else I've ever seen.
I've never met someone who was at work before me in the morning (8am) and still there when I quit (9pm) every day, and I'm "only" doing between 50 and 60. Of the friends that say they do big weeks, most of them try to call me before 8pm, so they're lying. So yeah, basically people count commute and lunch when they say over 60, and I am pretty sure not a lot of people have experienced a real 60 hrs week of work, without counting lunch, commute and pauses (which makes it around 14 hrs a day when you add these moments).
And anyone who has a kid and says he does over 50 is just lying...
All I'm asking for in Java 8 is the integration of vectorization instructions in the jvm. Please, do something for that >10x time factor compared to C++ with a compiler using correctly SSE/AVX instructions. I know most of the business doesn't care, but for the few who are still doing some computationally intensive processing (unrelated to databases btw), it is a game changer.
Researchers agree these terms because they have no other choice. Ok, seems nobody outside the academic gets the sense of publish or perish. Let me tell you why I continue to send my works to Elsevier (or the others) journals, whatever they are asking in the terms and conditions.
In my country (France), to get a research position you first have to get a "qualification" which involves a threshold on the number of journal papers you have. The higher the impact factor of the journal, the better it counts. Once you have this "qualification", you can try to get a position - the system is competition based, and most of the time it is based on the number of high impact factor journal papers you have. So yeah, basically, if you try to play the cowboy before you have the position, you'll never get one.
Now, I do have such position and I could put all my stuff on arxiv. But I also have PhD students, and they want to work in the academic. if I tell them to go the open access way, they'll never get the "qualification" and the position. Thus, we chase these "important" journals (read significant impact factor), and send the articles there. As long as articles in these journals is mandatory to get a position, we have no other choice than publishing there for the students.
To my mind, the solution lies not in the hands of the researchers, by is rather a political one. If the government dictates specific recommendations that positions should be awarded to people with open bibliography, the stupid behavior of Elsevier will die. As long as no political action is taken, it will continue as it was.
If you are really having a precision problem, even in double precision, then it means you are facing an ill-conditioned problem. And if you are facing an ill-conditioned problem, then there is nothing a technological tool can do for you. Try to reformulate the problem to avoid bad conditioning, and FP will be fine.
I'm actually managing an OS course for graduate students, and it's heavily based on linux (userspace and kernelspace). We do a few exercices (like writing a kernel module that computes averages), but nothing fancy. I've always been looking to propose them some projects related to kernel dev, but as I'm not a kernel hacker myself, I have clearly no idea of what seems reasonable.
So here's the deal: If you are involved on some subsystem of the linux kernel and you have something you want to get coded that can be a first experience with kernel dev, and that can be done under about 100 hours (the length of a typical project), you contact me. I'll do as much as possible as a first step filtering so that you won't get spamed. It's a win-win situation: I have great projects for my students, you get free work. For this year, it's a bit short, because projects are from September until January, but next year is ok.
It's funny to see people finally realize that the world we're headed to is very similar to that of East Germany, with the slight difference that you won't be assured to have a house, a job and food every day. Probably these points were not among the good things to retain from the Commies, whereas global surveillance was.
Exactly. My guess is that this publishing method is not the cause but the consequence of the current disfunction. Researcher are evaluated on prestige, exactly like businessmen are evaluated on money. That alone means that as a recognized researcher, you have absolutely no interest at seeing concurrent work get published, or perhaps only if they heavily cite your work. With such system, young researcher in small labs will never get any good publication (read in famous journals), whatever the quality of their work, except if they manage to attract a big name.
We have completely reproduced the capitalism system, albeit replacing money with citations. If you were "born" with a good initial capital because your advisors were rich, then you'll easily get more capital, creating dinasties of researchers. I'm not saying these dinasties are incompetent, I'm saying they are after some point barely correlated with competence, and that they mask really brilliant people, which in the end will lack proper funding and quit science.
You would have guess the world of science would have been more objective than that of business, but truth is humans are the same everywhere.
Slashdot Mirror
The most ridicule part being "must not be able to explain how". That doesn't even make sense for humans! If you ask artists, they'll tell you what their influences are, if you ask critics, they'll tell you why this particular piece of art was made this way and not in a completely different manner.
Fun fact: any program with yet unseen bugs that make their behavior totally unexplainable to their developers has passed the test. That gives you either an idea of the soundness of this crap, or a deep insight of what type of failure humankind is.
It depends of what you expect from an AI. If it is a perfect replica of a human mind, with which you can talk and share life as if it were human, then it will probably never be around. But that's also pretty useless, and most development in machine learning (ML) are in a more abstract level than trying to solve a very specific goal like this.
Now if you consider AI to be completely new intelligent species, that behave in an intelligent way (volontary fuzzy definition here), then it's probably already there. I mean, the ML programs that dictate the behaviour of you insurance policies so as to send you sport ads when you're a bit overweight, or holidays at the sea when you're close to a burn out, that raise the price of things predicted to induce a loss and lower the ones of things predicted have big return in order to influence your choices, etc, that, to me, sounds exactly like what you do with your pets when you decide they should eat that instead of this for some reason they could not handle with their inferior minds. Now, if you think of all the interconnected ML programs searching for new optima every second and exchanging information, you can view it as the new superior species of this planet.
A very short example: the vast majority of the human race wants to put an end to automated short-sighted finance, just like the vast majority of dogs wants to get free from their leashes. Bot never will until their recpective superior species allow them to. We talked a lot of the facebook experiment lately, the real question is how long has it been already done by the machine to fulfill goals we are not able to grasp? Maybe the singularity is already there since a few years, and just like for peak oil, we'll know it some time after. If we get to notice something more intelligent than us is governing our lives.
Seriously, the poster was presented yesterday at CVPR and ends up on /. today. There is nothing sensational about it. I'm getting sick of science turning into PR stunts all the time.
I have only one problem with fancy GUI that allow you to train a predicting model in 2 clicks: how confident can you be in your model, since all the parameters are masked and you have no knowledge about them? I still think it is dangerous to rely on a tool you don't understand and you can't control up to a satisfactory level, especially when it is to be used in prediction - something we expect to be highly reliable in many aspects due to old development of science like balistics.
I've written a ML library myself (also in Java, more lightweight than weka, but with no gui - although it comes with standalone binaries for some basic setups) and I can tell you there is no good default tunning that works well for every kind of situation. ML is seriously a young science that gets rapidly tricky even on very common problems, which is very different than field for which we have very accurate solvers that work most of the time (again balistics is probably a good example, at least because it is taught in school and sets the prototype of what we name science). I fear hidding the youth (and thus the imperfection) is only going to cause damage through misconception and false interpretation.
Asking if robots can be evil is about as futile as asking if a microwave can be happy.
That being said, there already are killer robots, with a pretty good track record in recent operations. But the evil lies in the humans who made them (from the top exec that launch the program to the small hand that does the job) and used them, not in the pile of steel and semiconductors.
caveat: Looking at your food, your microwave is probably sad, which explains their tendency to commit suicide.
JMLR is a fantastic journal, with high quality papers, high quality reviews, and completely open. The dream of many come true. I've always wondered why the idea did not spread to other fields.
If someone is willing to start a Journal of Computer Vision Research based on the same principles, count me in. I'll be happy to do editing/reviewing for such journal instead of well known IEEE/Elsevier/Springer journals.
Also mir is dead. NASA confirms it.
Given that mir also means peace and considering what is happening in Ukraine, I sadly suppose you are right. Nonetheless I prefer puns when they don't carry a grim presage.
I am pretty much waiting for the other way around: when I, as a human, known that I am quite bad at handling the current situation and that the machine will do much better than I'll ever be capable of. In particular, I am thinking of traffic jams where you have to find maximum speed that maintains a smooth global flow, without falling into chaotic start/stop sequences like this: http://www.youtube.com/watch?v...
Most humans typically tend to drive too fast in such situations, leading to an average speed well below what machines could very simply achieve. So basically, I'll drive when it's easy, and let the machine do the complicated stuff, as usual.
Funny, I often say that I have enough students to make having children unnecessary.
My wife and I are in the same situation, and I never understood the selfishness argument. Why is it selfish? To whom? What harm does it bring and to what?
The more I have this discussion with family and friends, the more it turns out to be pure jealousy towards us better enjoying our life. Most of them didn't expect it is that hard to raise children, and especially the many things you have to give up due to the lack of time to do it.
" just about every SSL-encrypted internet communication over the last two years has been compromised."
No, it really hasn't.
It's accurate to say that just about every Open-SSL encrypted session for servers that were using NEW versions of OpenSSL (not all those ones out there still stuck on 0.9.8(whatever) that never had the bug) were potentially vulnerable to attack.
That's bad, but it's a universe away from "every SSL session is compromized!!!" because that's not really true.
They were vulnerable to attack, that is to say, the security was compromised. He didn't say they were hacked, stolen, eavesdropped, or surreptitiously recorded.
No, not if they didn't enabled the compromised feature.
Stop the bullshit. You're not changing your mind, you're trying to gain a lot very quickly by gambling.
If you don't understand the implications of what you're doing, please go to the casino instead of messing up our global economy.
This. The internet and the books of Richard Dawkins are endless sources of arguments.
On the kickstarter page:
Power: 750 Watts continuous
Speed: 18mph without pedaling
Range: 12 miles with standard battery, 20 miles with extended battery
Battery: 240Wh / 400Wh
Well, it seems with the extended battery you can get about 32 minutes or 9.6 miles at 18mph, which is only half the range...
Yeah, that's also my reaction. I regularly do between 50 and 60 hrs a week by working 11 - 12 hrs a day the whole week (and nothing the week end cause I would otherwise go insane), I've been doing occasonially 70 hrs (that is, add 10 hrs during the WE), and I think my max was around 80 hrs for some relly tough deadline near the end of my PhD. Right now, I finished a hard period, and I'll be calming things down to around 40 hrs a week in the few next month to regain some health. Seriously, 60 is hard, around 70 is just insane, and over that is ruining your health more quickly than anything else I've ever seen.
I've never met someone who was at work before me in the morning (8am) and still there when I quit (9pm) every day, and I'm "only" doing between 50 and 60. Of the friends that say they do big weeks, most of them try to call me before 8pm, so they're lying. So yeah, basically people count commute and lunch when they say over 60, and I am pretty sure not a lot of people have experienced a real 60 hrs week of work, without counting lunch, commute and pauses (which makes it around 14 hrs a day when you add these moments).
And anyone who has a kid and says he does over 50 is just lying...
A cat is a cat, no more, no less!
Design a clean API and stick to well known coding standards.
This is /., no need to read more than the title.
All I'm asking for in Java 8 is the integration of vectorization instructions in the jvm. Please, do something for that >10x time factor compared to C++ with a compiler using correctly SSE/AVX instructions. I know most of the business doesn't care, but for the few who are still doing some computationally intensive processing (unrelated to databases btw), it is a game changer.
Researchers agree these terms because they have no other choice. Ok, seems nobody outside the academic gets the sense of publish or perish.
Let me tell you why I continue to send my works to Elsevier (or the others) journals, whatever they are asking in the terms and conditions.
In my country (France), to get a research position you first have to get a "qualification" which involves a threshold on the number of journal papers you have. The higher the impact factor of the journal, the better it counts. Once you have this "qualification", you can try to get a position - the system is competition based, and most of the time it is based on the number of high impact factor journal papers you have. So yeah, basically, if you try to play the cowboy before you have the position, you'll never get one.
Now, I do have such position and I could put all my stuff on arxiv. But I also have PhD students, and they want to work in the academic. if I tell them to go the open access way, they'll never get the "qualification" and the position. Thus, we chase these "important" journals (read significant impact factor), and send the articles there. As long as articles in these journals is mandatory to get a position, we have no other choice than publishing there for the students.
To my mind, the solution lies not in the hands of the researchers, by is rather a political one. If the government dictates specific recommendations that positions should be awarded to people with open bibliography, the stupid behavior of Elsevier will die. As long as no political action is taken, it will continue as it was.
Seems other users have a bigger cache than yours...
If you are really having a precision problem, even in double precision, then it means you are facing an ill-conditioned problem. And if you are facing an ill-conditioned problem, then there is nothing a technological tool can do for you. Try to reformulate the problem to avoid bad conditioning, and FP will be fine.
I'm actually managing an OS course for graduate students, and it's heavily based on linux (userspace and kernelspace). We do a few exercices (like writing a kernel module that computes averages), but nothing fancy. I've always been looking to propose them some projects related to kernel dev, but as I'm not a kernel hacker myself, I have clearly no idea of what seems reasonable.
So here's the deal: If you are involved on some subsystem of the linux kernel and you have something you want to get coded that can be a first experience with kernel dev, and that can be done under about 100 hours (the length of a typical project), you contact me. I'll do as much as possible as a first step filtering so that you won't get spamed. It's a win-win situation: I have great projects for my students, you get free work. For this year, it's a bit short, because projects are from September until January, but next year is ok.
I don't need to go back and look at some photos, I was born there.
It's funny to see people finally realize that the world we're headed to is very similar to that of East Germany, with the slight difference that you won't be assured to have a house, a job and food every day. Probably these points were not among the good things to retain from the Commies, whereas global surveillance was.
Exactly. My guess is that this publishing method is not the cause but the consequence of the current disfunction. Researcher are evaluated on prestige, exactly like businessmen are evaluated on money. That alone means that as a recognized researcher, you have absolutely no interest at seeing concurrent work get published, or perhaps only if they heavily cite your work. With such system, young researcher in small labs will never get any good publication (read in famous journals), whatever the quality of their work, except if they manage to attract a big name.
We have completely reproduced the capitalism system, albeit replacing money with citations. If you were "born" with a good initial capital because your advisors were rich, then you'll easily get more capital, creating dinasties of researchers. I'm not saying these dinasties are incompetent, I'm saying they are after some point barely correlated with competence, and that they mask really brilliant people, which in the end will lack proper funding and quit science.
You would have guess the world of science would have been more objective than that of business, but truth is humans are the same everywhere.