Really current IO is pretty clunky for this kind of proposition. To have to refocus your eyes or have a free hand in order to interface with a computer greatly reduces the utility of always-present devices.
Recently, speech recognition and synthesis have been approaching 'adequate' - HERE is a niche for these technologies.
I was under the impression that EV6 was only a 100MHz clock, but using the UDMA trick (of utilizing both the rising and falling edges of the signals to send data) to double the data bandwidth. Does anyone know what this is called, and what the most accurate way to discuss speeds of this kind of deal are?
Firefly, CDNOW, and Amazon have all tried to make 'intelligent' recommendations by demographic means (by which I mean they say "who else liked what you like, and what else did they like?"). These all generally don't work very well, and the more they're used the worse they're going to get.
I think they're using some kind of neural networks to boil down a large ammount of data into a small number of recommendations. Neural nets are notorious for being sneaky - if there's a way of coming up with a clever yet still useles solution, it's a good chance that a neural net will find it. (A real person wouldn't think of it because it's useless (there ARE real people who are not detered by this fact)) Here they learned that they can 'cheat' by recommending the most popular music - if everybody else likes X and I like Y, It's better just to recommend X than using its limited capacity to discern what Y' is.
This is compounded with the problem that if people aren't familliar with a recommended CD or book they're likely to just skip it and leave it unrated rather than shelling out $15+ to see if CDNOW's AI is any good, so massively popular items are the only things it even has a significant data set for. Suddenly it's not worth my bother - If I want to hear massively popular music, I'll turn on the radio or emptyV or whatever, rather than ask some stupid program what IT thinks I'll like.
To its credit Firefly USED to be able to understand that I wasn't so big on stuff that was too full of guitars (though it forgot this as more data and more noise was added).
It's a difficult problem - I don't necessarily want its recommendations to be similar to what I already said I like, just to be more stuff I like. In fact, if it's too similar I won't buy it unless I'm a collector or reviewer. (How many SQL books does one person need? I may like Bjork, but do I need all her singles and japanese import versions with one extra track? maybe not..)
It'll be a while until somebody gets this problem solved.
Interesting. I tried firefly fairly early as well (before they became BigNote and all that passport crap), but I don't recall their recommendations being very good - but my recollection is kind of fuzzy - I didn't spend much time on the site - just played with the music recommender (I was already spending too much time on IRC to enter another online community).
Have you used anything since then that's produced good recommendations?
From the story: ``Sony has been promoting various types of hard disks with several partners, and it won't start investments until it becomes clear which technology is the best,'' said Masashi Kubota, an analyst at ING Barings. This seems to suggest to me that this is fairly early on, and it will be some time before we see these. Oh well, I have enough storage to last me quite some time. I find denser media more exciting than cheaper media.
Slashdot Mirror
Really current IO is pretty clunky for this kind of proposition. To have to refocus your eyes or have a free hand in order to interface with a computer greatly reduces the utility of always-present devices.
Recently, speech recognition and synthesis have been approaching 'adequate' - HERE is a niche for these technologies.
I was under the impression that EV6 was only a 100MHz clock, but using the UDMA trick (of utilizing both the rising and falling edges of the signals to send data) to double the data bandwidth. Does anyone know what this is called, and what the most accurate way to discuss speeds of this kind of deal are?
Firefly, CDNOW, and Amazon have all tried to make 'intelligent' recommendations by demographic means (by which I mean they say "who else liked what you like, and what else did they like?"). These all generally don't work very well, and the more they're used the worse they're going to get.
I think they're using some kind of neural networks to boil down a large ammount of data into a small number of recommendations. Neural nets are notorious for being sneaky - if there's a way of coming up with a clever yet still useles solution, it's a good chance that a neural net will find it. (A real person wouldn't think of it because it's useless (there ARE real people who are not detered by this fact)) Here they learned that they can 'cheat' by recommending the most popular music - if everybody else likes X and I like Y, It's better just to recommend X than using its limited capacity to discern what Y' is.
This is compounded with the problem that if people aren't familliar with a recommended CD or book they're likely to just skip it and leave it unrated rather than shelling out $15+ to see if CDNOW's AI is any good, so massively popular items are the only things it even has a significant data set for. Suddenly it's not worth my bother - If I want to hear massively popular music, I'll turn on the radio or emptyV or whatever, rather than ask some stupid program what IT thinks I'll like.
To its credit Firefly USED to be able to understand that I wasn't so big on stuff that was too full of guitars (though it forgot this as more data and more noise was added).
It's a difficult problem - I don't necessarily want its recommendations to be similar to what I already said I like, just to be more stuff I like. In fact, if it's too similar I won't buy it unless I'm a collector or reviewer. (How many SQL books does one person need? I may like Bjork, but do I need all her singles and japanese import versions with one extra track? maybe not..)
It'll be a while until somebody gets this problem solved.
-me
Interesting. I tried firefly fairly early as well (before they became BigNote and all that passport crap), but I don't recall their recommendations being very good - but my recollection is kind of fuzzy - I didn't spend much time on the site - just played with the music recommender (I was already spending too much time on IRC to enter another online community).
Have you used anything since then that's produced good recommendations?
-me
From the story: ``Sony has been promoting various types of hard disks with several partners, and it won't start investments until it becomes clear which technology is the best,'' said Masashi Kubota, an analyst at ING Barings. This seems to suggest to me that this is fairly early on, and it will be some time before we see these. Oh well, I have enough storage to last me quite some time. I find denser media more exciting than cheaper media.