The Dark Ages refer to a period of European history, but other regions have all had their dark ages too. The fall of the Mayan empire, various political catastrophes that befell China, the conquering of golden age Arabia by the Mongols, etc.
Civilization has been more of a two steps forward, two steps back enterprise until fairly recently (maybe).
Actually, statistically, that's correct. There have been a few occasions, due to strikes or disasters, when communities have lost access to hospitals. Death rates in those communities have gone down.
If you're really sick then a hospital can save your life. If you're not, hospitals are dangerous. And most people who go to the hospital don't need to be there.
I'm not sure which way you envision the skew. The distribution of car ages is a type of life distribution, which is commonly modelled by exponential or Weibull distributions. This type of distribution results whenever you have new items (people, cars, widgets) that are created (nothing is less than 0 years old) and fail, either at a constant rate (a few types of plants and animals) or a rate that accelerates with age (pretty much everything else, including cars). Characteristically, the median is lower than the mean. That means that the majority of cars are younger than the mean age. Your statement that it would take at least the mean age (11 years) to replace half the cars is false: 50% replacement time is the median age, which is less.
In terms of actual replacement time, there's no reason why the mean or median age of cars has to be as high as it is. My ten year old car is pretty much just like the latest model, except for the lack of Bluetooth (which I added myself). In the past, when new models were worth upgrading to, the mean/median age was lower. When electric vehicles have some real desirable advantages over gas the upgrade cycle will probably shorten. It will go even faster as infrastructure is switched over and gas becomes more expensive. Even faster if high carbon taxes are implemented.
No, it's not. Google is neck deep in deep learning. They use it for everything. The key is half of the name: learning. You don't program a deep learning system, or construct a database for it, you teach it.
People are going to keep moving the goal posts on "intelligence" until skynet exterminates them, but a system that can learn (in some cases all by itself) to do the things we take for granted today was absolutely considered AI, and pretty miraculous AI, twenty years ago. It's certainly not human level yet, but many of the properties of these systems check a lot of boxes on reasonable definitions of intelligence that don't involve magic, and their capacity is growing very fast.
Your reply, along with many others here, suggests that you're quite likely in the group of people who don't really know what modern AI actually is.
I'm surprised it's so close, actually. 160 GWh (about 5% of total production) isn't that much capacity to add to supply a total conversion of US vehicles to electric.
The Alberta government has used oil revenue to fund diversification, mostly into high tech, for at least fifty years. An Alberta informatics research scholarship funded part of my PhD (in medical imaging). The tories made a hash of that over the last five years or so, but it was a fairly strong program before that.
Alberta probably should give more PR to industries other than oil. They do exist.
The real AI (for definitions of AI that include systems that learn) is hidden, or more subtle. Google image search, and probably Google's regular search, are now sophisticated deep learning networks. Google probably also has a lot of targeted ad and business decision type stuff that uses deep learning.
Alpha Go was a public demonstration, like an Indy race. Siri is a toy for the public to feed Apple data they'll use to build something more sophisticated.
If you say "whether" as a single word to me, I'm going to assume you meant "weather?" too. If you use the word in a sentence I'll judge which one you meant by context. Siri does that too.
The Siri system is voice recognition coupled with natural language processing. It's also a learning system, not a programmed one.
I don't think that's true. Copyright restrictions apply to copying and giving copyrighted material to other people. I can do whatever I want, so long as I don't give it to someone else. If I dynamically link my program to a GPL library the only GPLed bits of code in my program are interface code, which is fair use. I can then distribute both my program and the GPL library, with the restriction that I provide all the source necessary to modify and rebuild the GPLed library.
The GPL (and LGPL) cover distribution. If I write a program that dynamically links to a GPL library, it includes only necessary code only from the interface. If I then distribute that binary, without the actual library, I haven't distributed any GPLed code except the necessary interface. As others have pointed out, if use of the interface is fair use, then I've complied with the GPL.
The biggest difference between the GPL and the LGPL is the extension of restrictions via copyright to the interface.
I think you could do it even more simply than that. Just link against gfoo.so. No actual GPL code is included in your binary except bits of the interface, which is now fair use. The end user has to supply their own gfoo.so.
The reason it wouldn't work before is that linking against gfoo.so (or ngfoo.so) would include some stuff derived from gfoo.h, or a clone of it. If that interface code is copyrightable, the GPL applies.
To me that was always an overstep by the GPL. If GPL code is compiled into your binary, absolutely the whole thing should be open source. If it's dynamically linked, it shouldn't. The user can still swap versions of the library if they so choose, which is the stated intention of that part of the GPL.
Slashdot Mirror
It's hard to think of a boat that's more fuel efficient than a sailboat.
If using Python between batches is a problem for you, make your batches bigger. Or use a C optimizer. Or use Cython for the training loop.
It does a lot better than me at about 90% of the languages it knows. Maybe 95%. And it's faster than anybody.
Also, AI != better than human intelligence.
The Dark Ages refer to a period of European history, but other regions have all had their dark ages too. The fall of the Mayan empire, various political catastrophes that befell China, the conquering of golden age Arabia by the Mongols, etc.
Civilization has been more of a two steps forward, two steps back enterprise until fairly recently (maybe).
"Historians use the term "Dark Ages" for times about which we have not written history. "
You're incorrect: https://en.wikipedia.org/wiki/...
Ever used Google translate?
If you're driving a jeep in Jurassic Park.
No, he drives like that on the way HOME from the pub.
Not if the software you're using the password for isn't dumb. Use a phrase. Long, easy to remember, and very hard to guess.
Actually, statistically, that's correct. There have been a few occasions, due to strikes or disasters, when communities have lost access to hospitals. Death rates in those communities have gone down.
If you're really sick then a hospital can save your life. If you're not, hospitals are dangerous. And most people who go to the hospital don't need to be there.
Slashdot editors are all crackpots at heart, and crackpots like to Capitalize random Words because it Seems Biblical or Something.
I'm not sure which way you envision the skew. The distribution of car ages is a type of life distribution, which is commonly modelled by exponential or Weibull distributions. This type of distribution results whenever you have new items (people, cars, widgets) that are created (nothing is less than 0 years old) and fail, either at a constant rate (a few types of plants and animals) or a rate that accelerates with age (pretty much everything else, including cars). Characteristically, the median is lower than the mean. That means that the majority of cars are younger than the mean age. Your statement that it would take at least the mean age (11 years) to replace half the cars is false: 50% replacement time is the median age, which is less.
In terms of actual replacement time, there's no reason why the mean or median age of cars has to be as high as it is. My ten year old car is pretty much just like the latest model, except for the lack of Bluetooth (which I added myself). In the past, when new models were worth upgrading to, the mean/median age was lower. When electric vehicles have some real desirable advantages over gas the upgrade cycle will probably shorten. It will go even faster as infrastructure is switched over and gas becomes more expensive. Even faster if high carbon taxes are implemented.
No, it's not. Google is neck deep in deep learning. They use it for everything. The key is half of the name: learning. You don't program a deep learning system, or construct a database for it, you teach it.
People are going to keep moving the goal posts on "intelligence" until skynet exterminates them, but a system that can learn (in some cases all by itself) to do the things we take for granted today was absolutely considered AI, and pretty miraculous AI, twenty years ago. It's certainly not human level yet, but many of the properties of these systems check a lot of boxes on reasonable definitions of intelligence that don't involve magic, and their capacity is growing very fast.
Your reply, along with many others here, suggests that you're quite likely in the group of people who don't really know what modern AI actually is.
I'm surprised it's so close, actually. 160 GWh (about 5% of total production) isn't that much capacity to add to supply a total conversion of US vehicles to electric.
The Alberta government has used oil revenue to fund diversification, mostly into high tech, for at least fifty years. An Alberta informatics research scholarship funded part of my PhD (in medical imaging). The tories made a hash of that over the last five years or so, but it was a fairly strong program before that.
Alberta probably should give more PR to industries other than oil. They do exist.
What would be the advantage for an oil company to push findings that oil is going out of style?
Someone who assumes the age distribution of motor vehicles is symmetric might want to be a little more cautious accusing others of being bad at math.
The real AI (for definitions of AI that include systems that learn) is hidden, or more subtle. Google image search, and probably Google's regular search, are now sophisticated deep learning networks. Google probably also has a lot of targeted ad and business decision type stuff that uses deep learning.
Alpha Go was a public demonstration, like an Indy race. Siri is a toy for the public to feed Apple data they'll use to build something more sophisticated.
Sure she can.
If you say "whether" as a single word to me, I'm going to assume you meant "weather?" too. If you use the word in a sentence I'll judge which one you meant by context. Siri does that too.
The Siri system is voice recognition coupled with natural language processing. It's also a learning system, not a programmed one.
"Infringement happens when linking occurs"
I don't think that's true. Copyright restrictions apply to copying and giving copyrighted material to other people. I can do whatever I want, so long as I don't give it to someone else. If I dynamically link my program to a GPL library the only GPLed bits of code in my program are interface code, which is fair use. I can then distribute both my program and the GPL library, with the restriction that I provide all the source necessary to modify and rebuild the GPLed library.
The GPL (and LGPL) cover distribution. If I write a program that dynamically links to a GPL library, it includes only necessary code only from the interface. If I then distribute that binary, without the actual library, I haven't distributed any GPLed code except the necessary interface. As others have pointed out, if use of the interface is fair use, then I've complied with the GPL.
The biggest difference between the GPL and the LGPL is the extension of restrictions via copyright to the interface.
I think you could do it even more simply than that. Just link against gfoo.so. No actual GPL code is included in your binary except bits of the interface, which is now fair use. The end user has to supply their own gfoo.so.
The reason it wouldn't work before is that linking against gfoo.so (or ngfoo.so) would include some stuff derived from gfoo.h, or a clone of it. If that interface code is copyrightable, the GPL applies.
To me that was always an overstep by the GPL. If GPL code is compiled into your binary, absolutely the whole thing should be open source. If it's dynamically linked, it shouldn't. The user can still swap versions of the library if they so choose, which is the stated intention of that part of the GPL.
Yes. He did. Scientifically.
Nah. He posted the same kind of thing non-anonymously above. He's a jack of all trades Slashdot troll.
Don't forget that the study doesn't appear to have been properly blinded. Improper blinding can easily cause effects as big as they observed.
Nice catch on the multiple comparisons.