The lump of platinum is precise at around 5e-8, the watt balance has reached 3.6e-8 at that point, but they expect to do better than that. The Nist, while holding the record right now, does not seem to have chosen the best path to reach better precisions than that. 1e-9 is expected (hoped?;-) in some of the french experiments in a handful of years.
The Reason you use gas is it's easier to turn on and off the Coal/Nuclear.
Gas is only marginally easier than coal - it's the preheat time for the water in the loop that's the real killer AIUI.
Gas turbines (no water) are fast to start and stop, and they're very good when you need power *now*. They're used in France a lot for peaks. Civilian nuclear takes 2-3 days to change power levels significantly. I suspect that, compared to submarines, size matters.
There are so many things that can trip you up and give a completely wrong answer - and unlike programming (where your program doesn't compile or run correctly) you have no automatic way of discovering mistakes.
Errr, you probably don't realize it, but that sentence proves you're not qualified to talk about programming. Defining and verifying "run correctly" can be as hard as validating experimental results, simply because a run of a program is an experiment.
One interesting thing is that some of the most successful physical theories, successful in an "experimental validation" sense of the word, are pretty much out there where it comes to how things are really happening. Quantum electrodynamics are an impressive case. If you have the time, read "QED: The Strange Theory of Light and Matter", by R. Feynman. Not only it's very interesting and well written, it's also pretty illuminating on that subject. A good quarter of the book is about driving home the point that it is a mathematical theory that works impressively well but at the same time makes no intuitive sense, some like Einstein would say no sense at all ("God does not play dice" is the common paraphrase), where it comes to understanding the world.
You can also extend question 4 into "How much of that warming trend is due to human-produced CO2, how much is due to human-produced anything else, and how much is natural variation?".
His point three is not about whether the current concentrations of CO2 are human-produced (as you say, the isotopic ratios seem conclusive), but how much of the measured warming is due to CO2 concentrations. "We can't think of anything else" is not very good as an answer and, according to him (I have no idea), predictive models of temperature-vs-CO2 concentration seem to be lacking.
The state-of-the-art is crap, but it's still the state-of-the-art. In any case, that's what is currently best, for a value of best meaning "gets the best score in most MT evaluations when used competently". Moses is a bunch of code implementing fun statistical algorithms though, not a full translation system. The quality of the system you get depends on the quality of the training you do, and pretty much how you setup the system together. The guy was asking for the algorithms though:-)
There's no reason to think the video is not real. I'm pretty sure it originally comes from the TC-Star european project (sorry, original web site is dead). Making it fit on a iPhone and work reasonably well with the comparatively puny CPU and disk space is an impressive feat though.
OG.
State-of-the-art is Moses for decoding with Giza++ for word-alignment for training. The MT domain has an egyptian naming tradition for some reason (Moses is the open-source successor to Pharaoh).
OG.
IMHO, Eve Online is much more commerce oriented than Second Life. The makers, CCP, support a pseudo mechanism for real to virtual currency excahnge. A player can use real money to buy a game card, then sell the game code to another player for virutal money. So, there is market established exchange rate.
But that only goes from real to virtual. There is no CCP-supported mechanism for virtual to real conversion afaict, which would change things, starting with this little thing called taxes.
Bitsort / bucketsort algorithms run in O(n log(log(n))) time on data with a roughly uniform distribution, so O(n*log(n)) isn't a hard lower limit.
n*log(n) is a hard limit when you don't add preconditions like a roughly uniform distribution. It's even easy to see why. There are n! possible orders for n elements and you need to find out which it is to sort them. Any test operation you do splits the sets of possibles orders m-way for a fixed (non-dependant on n) m. So k tests will sort between m**k orders at max.
So we want the smallest k such as m**k >= n!. Which means k=log(n!)/log(m) ~= (n.log(n)-n)/log(m) (Stirling's approximation for n!).
Since m is fixed, in O notation that gives you O(n.log(n)).
What you *can* do, is to be faster on an interesting subset of the n! cases (partially ordered lists for instance) or to add constraints on the values so that they actually give you, more or less directly, information on the initial order (uniform distribution is such a case). That's very much a case of "knowing what your input is going to look like".
And your oh-so-magical supervising kernel knows which registers of the network card control the addresses in ram that can be accessed how, exactly? Iommus are a recent evolution in the x86 world and even then I'm not entirely sure you can have one per hardware device with fast switching.
The idea is to only enable the unsafe stuff when it is truly necessary (and probably, to have code reviews where the default answer to any unsafe code is to reject it unless it can be justified). If you need to peek and poke device registers, you can do it, but by default, your code is safe and immune to buffer overflows. On a percentage basis, very little of a kernel is actually talking to bare metal; most of it is talking about talking to the bare metal.
A quick measurement shows that driver and other bare metal code is around 80% of the linux kernel code. Very little indeed.
Depends on what their system needs. If it's just localization of output, it's reasonably easy. But if it's linguistic resources for language analysis, what's available for english is orders of magnitude more than for any other language (WordNet, VerbNet, FrameNet, XTag...).
Yeah if you do that enough times, maybe it will become true!
You just explained in your #1 bullet point exactly how it is capable of infecting your code. You include it in your project and the whole thing becomes GPL if you distribute the binary even once. Even if you take it out and close it later, you've been forced to expose your source code to the whole world.
No it doesn't become GPL. Changing the license on your code is not one of the remedies a court can decide. The court can only decide on injunction against distributing the code, money and in extreme cases jail.
The only way the license can change on your code is if you decide it, possibly through a settlement.
Take the phenomenon of black males in America (as opposed to most African nations) who carry a sexual fetish for paler, light haired women.
Among dark-skinned races, lighter skin is seen as beautiful. I don't know why but I guess it's the counterpart to light-skinned races' fixation on sun tans.
I don't know about black people, but the history of beauty standards in western europe show that the attraction is about "I don't have to work for a living". In medieval times, people worked in the fields and as a result has a tan. The standards of beauty were as a result "light skin is better". Now people work in offices and don't see the sun, so the standard is the other way around.
The GPL and the LGPL are probably Stallman's biggest contributions, even more important than gcc and all the other unix tools reimplementations. It was the first license with the "share alike" property which allows a lot of free software programmers not to feel potentially ripped off when they release code.
The truth of what happens isn't important, lots of BSD code gets in-kind contributions from corporations and lots of GPL code gets ripped off anyway when there's nobody willing to sue. The perception is what's important, specifically the perception of who is writing the code.
The problem at least for Qt is that it existed before STL, and seriously before boost. And severely breaking compatibility between versions of a library is not nice.
It seems you're right though, the Wikipedia page for IBM ViaVoice needs citations, and the page for its Windows 3.1 and OS/2 predecessor VoiceType, which is the one I saw demonstrated, has not even been created yet.
Ah, I see. You judge the state of speech recognition by the state of dictation products. That's a little... limited, you know?
Slashdot Mirror
You missed that one Watt Balance.
The lump of platinum is precise at around 5e-8, the watt balance has reached 3.6e-8 at that point, but they expect to do better than that. The Nist, while holding the record right now, does not seem to have chosen the best path to reach better precisions than that. 1e-9 is expected (hoped? ;-) in some of the french experiments in a handful of years.
OG.
No we don't. This is not the crawler you're looking for.
OG.
Gas is only marginally easier than coal - it's the preheat time for the water in the loop that's the real killer AIUI.
Gas turbines (no water) are fast to start and stop, and they're very good when you need power *now*. They're used in France a lot for peaks. Civilian nuclear takes 2-3 days to change power levels significantly. I suspect that, compared to submarines, size matters.
OG.
Then some would say an experiment just has to be good enough to publish... :-)
OG.
There are so many things that can trip you up and give a completely wrong answer - and unlike programming (where your program doesn't compile or run correctly) you have no automatic way of discovering mistakes.
Errr, you probably don't realize it, but that sentence proves you're not qualified to talk about programming. Defining and verifying "run correctly" can be as hard as validating experimental results, simply because a run of a program is an experiment.
OG.
One interesting thing is that some of the most successful physical theories, successful in an "experimental validation" sense of the word, are pretty much out there where it comes to how things are really happening. Quantum electrodynamics are an impressive case. If you have the time, read "QED: The Strange Theory of Light and Matter", by R. Feynman. Not only it's very interesting and well written, it's also pretty illuminating on that subject. A good quarter of the book is about driving home the point that it is a mathematical theory that works impressively well but at the same time makes no intuitive sense, some like Einstein would say no sense at all ("God does not play dice" is the common paraphrase), where it comes to understanding the world.
OG.
You can also extend question 4 into "How much of that warming trend is due to human-produced CO2, how much is due to human-produced anything else, and how much is natural variation?".
OG.
His point three is not about whether the current concentrations of CO2 are human-produced (as you say, the isotopic ratios seem conclusive), but how much of the measured warming is due to CO2 concentrations. "We can't think of anything else" is not very good as an answer and, according to him (I have no idea), predictive models of temperature-vs-CO2 concentration seem to be lacking.
OG.
The state-of-the-art is crap, but it's still the state-of-the-art. In any case, that's what is currently best, for a value of best meaning "gets the best score in most MT evaluations when used competently". Moses is a bunch of code implementing fun statistical algorithms though, not a full translation system. The quality of the system you get depends on the quality of the training you do, and pretty much how you setup the system together. The guy was asking for the algorithms though :-)
OG.
There's no reason to think the video is not real. I'm pretty sure it originally comes from the TC-Star european project (sorry, original web site is dead). Making it fit on a iPhone and work reasonably well with the comparatively puny CPU and disk space is an impressive feat though. OG.
State-of-the-art is Moses for decoding with Giza++ for word-alignment for training. The MT domain has an egyptian naming tradition for some reason (Moses is the open-source successor to Pharaoh). OG.
IMHO, Eve Online is much more commerce oriented than Second Life. The makers, CCP, support a pseudo mechanism for real to virtual currency excahnge. A player can use real money to buy a game card, then sell the game code to another player for virutal money. So, there is market established exchange rate.
But that only goes from real to virtual. There is no CCP-supported mechanism for virtual to real conversion afaict, which would change things, starting with this little thing called taxes.
OG.
They use document.write so that the page can't show up until the ad image is finished loading.
OG.
What next, a warning about how Windows 7 requires 16 GB of storage...
Aw crap! I hope Windows 7 is 64-bit, because a 32-bit OS can't address that much!
Yes it can. That's PAE for you. We had 5 16G 32bits linux machines at a time (now we're buying 64G 64bits ones).
OG.
Bitsort / bucketsort algorithms run in O(n log(log(n))) time on data with a roughly uniform distribution, so O(n*log(n)) isn't a hard lower limit.
n*log(n) is a hard limit when you don't add preconditions like a roughly uniform distribution. It's even easy to see why. There are n! possible orders for n elements and you need to find out which it is to sort them. Any test operation you do splits the sets of possibles orders m-way for a fixed (non-dependant on n) m. So k tests will sort between m**k orders at max.
So we want the smallest k such as m**k >= n!. Which means k=log(n!)/log(m) ~= (n.log(n)-n)/log(m) (Stirling's approximation for n!).
Since m is fixed, in O notation that gives you O(n.log(n)).
What you *can* do, is to be faster on an interesting subset of the n! cases (partially ordered lists for instance) or to add constraints on the values so that they actually give you, more or less directly, information on the initial order (uniform distribution is such a case). That's very much a case of "knowing what your input is going to look like".
OG.
And your oh-so-magical supervising kernel knows which registers of the network card control the addresses in ram that can be accessed how, exactly? Iommus are a recent evolution in the x86 world and even then I'm not entirely sure you can have one per hardware device with fast switching.
OG.
The idea is to only enable the unsafe stuff when it is truly necessary (and probably, to have code reviews where the default answer to any unsafe code is to reject it unless it can be justified). If you need to peek and poke device registers, you can do it, but by default, your code is safe and immune to buffer overflows. On a percentage basis, very little of a kernel is actually talking to bare metal; most of it is talking about talking to the bare metal.
A quick measurement shows that driver and other bare metal code is around 80% of the linux kernel code. Very little indeed.
OG.
Depends on what their system needs. If it's just localization of output, it's reasonably easy. But if it's linguistic resources for language analysis, what's available for english is orders of magnitude more than for any other language (WordNet, VerbNet, FrameNet, XTag...).
OG.
Let's repeat.
Yeah if you do that enough times, maybe it will become true!
You just explained in your #1 bullet point exactly how it is capable of infecting your code. You include it in your project and the whole thing becomes GPL if you distribute the binary even once. Even if you take it out and close it later, you've been forced to expose your source code to the whole world.
No it doesn't become GPL. Changing the license on your code is not one of the remedies a court can decide. The court can only decide on injunction against distributing the code, money and in extreme cases jail.
The only way the license can change on your code is if you decide it, possibly through a settlement.
OG.
Take the phenomenon of black males in America (as opposed to most African nations) who carry a sexual fetish for paler, light haired women.
Among dark-skinned races, lighter skin is seen as beautiful. I don't know why but I guess it's the counterpart to light-skinned races' fixation on sun tans.
I don't know about black people, but the history of beauty standards in western europe show that the attraction is about "I don't have to work for a living". In medieval times, people worked in the fields and as a result has a tan. The standards of beauty were as a result "light skin is better". Now people work in offices and don't see the sun, so the standard is the other way around.
OG.
You can't fragment files in iso9660. That means adding data to a file can often require rewriting the whole file somewhere else.
OG.
The GPL and the LGPL are probably Stallman's biggest contributions, even more important than gcc and all the other unix tools reimplementations. It was the first license with the "share alike" property which allows a lot of free software programmers not to feel potentially ripped off when they release code.
The truth of what happens isn't important, lots of BSD code gets in-kind contributions from corporations and lots of GPL code gets ripped off anyway when there's nobody willing to sue. The perception is what's important, specifically the perception of who is writing the code.
OG.
Maybe the NSA has proven that P=NP.
OG.
The problem at least for Qt is that it existed before STL, and seriously before boost. And severely breaking compatibility between versions of a library is not nice.
OG.
It seems you're right though, the Wikipedia page for IBM ViaVoice needs citations, and the page for its Windows 3.1 and OS/2 predecessor VoiceType, which is the one I saw demonstrated, has not even been created yet.
Ah, I see. You judge the state of speech recognition by the state of dictation products. That's a little... limited, you know?
OG.