"...use recommendations, search, a grid, or any other means you care to name to figure out which of those thousands and thousands and thousands of hours of programming you want to watch..."
Those poor channel-hoppers, who can't watch a programme for more than 10 minutes without wondering what else might be on, will now have all the material from the past to choose from aswell. Lucky them!
I was so impressed by this idea that I wrote something quickly to do exactly what you suggested. I've set it running, and I'll let you know the answer just as soon as it has finished....;^)
Interesting point. I guess the training data is quite likely to contain examples of this kind, especially in the early stages of the game, because it is possible in chess to "transpose" from one opening to another.
The spam filter chess program is interesting in this respect, as you say: it will take into account the game history - effectively being influenced by its opponent's psychology, to some degree. I'm not aware of any other chess program that does this (disclaimer: I don't play chess against computers very often!). But should a good player be thrown by this kind of situation, or should s/he purely consider the best move from this position?
Agreed. There is no way the spam filter is maintaining any kind of useful internal representation of the board state - you can bet your life that it will suggest moves from a given position differently depending on the sequence of moves that led there.
There is an interesting article here (PDF) that describes how northen cod in southern Labrador and eastern Newfoundland fisheries suffered as a result of maturing at ever-earlier ages/smaller sizes (a trend caused by the way the fisheries were managed).
And don't forget the poor bulldog (scroll down to the Health section), which is now such a genetic mess that they have to rely human intervention in order to survive (the pups are usually delivered by c-section due to their large heads).
Okay, fair enough, but weather prediction models don't keep coming out with things like "the whole world will be consumed by hurricanes and lightning" - that's the kind of crazy nonsense I'm talking about.
If a simulation of the universe's evolution predicted that it would collapse after (say) 10 years, you'd have to say the underlying model was wrong. If it predicted a universe which after about 13*10^9 years had the broad characteristics of the current observed universe (avg. density, size of density fluctuations etc.), you'd say you were along the right lines.
Okay, for example: the weather is predicted by running simulations on a computer, and although the detailed predictions are notorious for their unreliability (especially where I live in Britain), the fact that they do at least predict something "plausible looking" indicates that the underlying model is along the right lines. If the model was way off, the prediction would be crazy nonsense.
I think this the point that Black Parrot was making in the GP post - simulations like this provide you with a "sanity check" for your theory... please correct me if I misunderstood.
Not wanting to veer too far OT, but having read another story on/. I was inspired to reply to your point in pictorial form. I just found this picture of a bunch of open source developers - they seem quite happy to me, certainly not starving;^)
I'm quite probably talking rubbish but... is hacking the GPU the best possible solution to the problem (i.e. fastest sort possible for the amount of money spent on the hardware)? Or is this just someone having some fun (which is fair enough)?
Unless, I suppose, you have data stored on some bizzare medium that can only be read by old hardware. As far as I know (which is not all that far, admittedly) only a Spectrum +3 can read the old +3 floppy disks. If memory serves they were 3 inch rather than 3.5.
Actually there is a loophole: A particle can travel faster than light within the framework of SR, provided its (equivalent of) "rest mass" is imaginary (not negative as stated somewhere earlier). Such particles would still have real energy/momentum vectors, but could never be decelerated below light speed - it would require an infinite amount of work. The reasons why FTL particles (tachyons) are normally disregarded in SR are:
1. It would allow signals to be sent back in time, which some people argue leads to paradoxes.
2. Such particles have never been observed. This is the real important one!
Slashdot Mirror
"...use recommendations, search, a grid, or any other means you care to name to figure out which of those thousands and thousands and thousands of hours of programming you want to watch..."
Those poor channel-hoppers, who can't watch a programme for more than 10 minutes without wondering what else might be on, will now have all the material from the past to choose from aswell. Lucky them!
Perhaps a slashdot poll would be a better idea. No-one could possibly doubt the validity of the results from that...
In the meantime - put me down for Google.
I was so impressed by this idea that I wrote something quickly to do exactly what you suggested. I've set it running, and I'll let you know the answer just as soon as it has finished.... ;^)
Interesting point. I guess the training data is quite likely to contain examples of this kind, especially in the early stages of the game, because it is possible in chess to "transpose" from one opening to another.
The spam filter chess program is interesting in this respect, as you say: it will take into account the game history - effectively being influenced by its opponent's psychology, to some degree. I'm not aware of any other chess program that does this (disclaimer: I don't play chess against computers very often!). But should a good player be thrown by this kind of situation, or should s/he purely consider the best move from this position?
Agreed. There is no way the spam filter is maintaining any kind of useful internal representation of the board state - you can bet your life that it will suggest moves from a given position differently depending on the sequence of moves that led there.
There is an interesting article here (PDF) that describes how northen cod in southern Labrador and eastern Newfoundland fisheries suffered as a result of maturing at ever-earlier ages/smaller sizes (a trend caused by the way the fisheries were managed).
And don't forget the poor bulldog (scroll down to the Health section), which is now such a genetic mess that they have to rely human intervention in order to survive (the pups are usually delivered by c-section due to their large heads).
My mate Ad says it makes perfect sense.
Okay, fair enough, but weather prediction models don't keep coming out with things like "the whole world will be consumed by hurricanes and lightning" - that's the kind of crazy nonsense I'm talking about.
If a simulation of the universe's evolution predicted that it would collapse after (say) 10 years, you'd have to say the underlying model was wrong. If it predicted a universe which after about 13*10^9 years had the broad characteristics of the current observed universe (avg. density, size of density fluctuations etc.), you'd say you were along the right lines.
Okay, for example: the weather is predicted by running simulations on a computer, and although the detailed predictions are notorious for their unreliability (especially where I live in Britain), the fact that they do at least predict something "plausible looking" indicates that the underlying model is along the right lines. If the model was way off, the prediction would be crazy nonsense.
I think this the point that Black Parrot was making in the GP post - simulations like this provide you with a "sanity check" for your theory... please correct me if I misunderstood.
Not wanting to veer too far OT, but having read another story on /. I was inspired to reply to your point in pictorial form. I just found this picture of a bunch of open source developers - they seem quite happy to me, certainly not starving ;^)
Probably not a good idea in light of this article (sorry you can't read the entire story without a subscription but you get the idea).
I'm quite probably talking rubbish but... is hacking the GPU the best possible solution to the problem (i.e. fastest sort possible for the amount of money spent on the hardware)? Or is this just someone having some fun (which is fair enough)?
Someone please enlighten me!
Or at least use the preview button... ;^)
Unless, I suppose, you have data stored on some bizzare medium that can only be read by old hardware. As far as I know (which is not all that far, admittedly) only a Spectrum +3 can read the old +3 floppy disks. If memory serves they were 3 inch rather than 3.5.
Actually there is a loophole: A particle can travel faster than light within the framework of SR, provided its (equivalent of) "rest mass" is imaginary (not negative as stated somewhere earlier). Such particles would still have real energy/momentum vectors, but could never be decelerated below light speed - it would require an infinite amount of work. The reasons why FTL particles (tachyons) are normally disregarded in SR are: 1. It would allow signals to be sent back in time, which some people argue leads to paradoxes. 2. Such particles have never been observed. This is the real important one!