This is apparently all because HTTP has a referer field (unless the user turns it off in the browser), so clickthroughs on ads have the url you were on when you clicked. FB has lots of urls with user id's in them, which lead to pages with the user's public information, friends, etc. Researchers have already crawled most of these urls without much trouble, but the definition of "giving away private information" seems to have changed a bit under the influence of lawyers.
Nowadays it looks like FB puts ad clickthroughs through a redirect that hides the referer. I suppose the WSJ will discover next that clickthroughs reveal the user's IP address and blame FB for it.
They spent most of their money building nuclear weapons instead of investing in useful public utilities, like food. Even their native energy sources like coal are sold to China for foreign exchange.
FWIW, I had lunch with the Netflix search guy a while ago and told him the same thing. For me it's Korean. I get recommended all kinds of unrelated stuff.
I think a lot of the metadata comes from Hollywood's view of the world, so "foreign" it is.
Come to think of it, I even had to return one Korean film because it had no Korean soundtrack -- it was a Hong Kong release dubbed in Chinese, with English subtitles. I had some serious doubts about my Korean skills listening to that one.
There's one Korean film I rented a while ago that has a lot of Japanese: "The President's Last Bang". Good film.
That's about what I expected. I've worked with a similar entity-extraction tool, and it's largely a process of twiddling the rules until it gets most of what you think is in there. The regex's, etc., used to find the entities are so large that the engine can easily bog down, and it's a complete pain to have minor changes balloon into 2x or worse performance hits.
One shortcoming is the lack of interactivity on large datasets. Most web searchers iterate through a few queries until they get what they want, but in this case the iterations could take days.
Is Naver technically a better search engine? In other words, does it do better with segmentation and stemming than Google? I'm curious because I speak some Korean but I don't have enough experience to notice problems like these. I assume you're seeing search results that miss or confuse certain forms of a word, or fail to find stem words embedded in particles or verb endings.
There are only a few companies that make Korean stemmers, and I'm familiar with two of them. Each uses a large staff to hard-code stemming and segmentation cases for hundreds of thousands of words. The dictionaries that result run into hundreds of megabytes, but they usually do a decent job, or as good a job as anybody else.
Google, IIRC, was using one of these products to do its Korean indexing, so I'm surprised that there are quality differences.
Unfortunately, these efforts will be useless, as unimpeachably scientific studies have shown. Cameras do not eliminate politics; they merely push military regimes into other areas. Democracy is doomed, I tell you.
There should be more than enough immigrants available to satisfy food demand. The management company, Soylent, Inc., always has plenty of people on hand for harvesting and processing.
He seems to be popularizing a lot of research, but it's the type of thing many of us do naturally. For me, chasing down hidden indicators is pretty much par for the course. I live in San Francisco, and the city is so routinely out of whack, corrupt, negligent, etc., that I practically maintain my own data warehouse of performance data.
His "surprising facts" were not that surprising. The fact that many crack dealers live with their mothers? That's a no-brainer. Mom doesn't have a criminal record, so mom is eligible for public housing, and junior waltzes in without a lease. Free housing, plus a layer of deniability for eviction court. Public housing lessees are something like 65% female.
Likewise, in a town with little effective prosecution, how do you get rid of drug dealers? Get them parking tickets. Then when the parking people don't do their jobs you have to go after meter revenues, file sunshine requests on ticket totals, even go out and count cars at expired meters (around 80%).
How do you get a map of all gunfire incidents in San Francisco? Not from the cops (until recently) or the DA. The Health department publishes one, though.
It's all a game of people protecting their turf, and the public tricking them into revealing their information. These days, anyone can crunch data, so the task is mainly to think of useful measures to apply to available sources. The book wasn't bad, but it didn't teach me much.
In 1922 diodes were vacuum tubes with two wires in them, one of them heated and capable of emitting electrons to pass a current. What he seems to have discovered is an inverse photoelectric effect, where electrons emit light when striking a metal target.
Although, come to think of it, there were Germanium diodes at the time, so it could have been something more semiconductor-related, but the article isn't clear.
The efficiency of an LED die is limited by how many photons actually get out of the material and into the open air (if they don't, they reflect internally and turn into heat). Many of the recent improvements in LED efficiency have come about through better light transmission, which requires a careful transition from the high-index-of-refraction die to low-index-of-refraction air.
Security cameras are a tool. They're actually a less oppressive tool than people think, because recorded information can often be subpoenaed or requested through freedom-of-information laws.
It's not the tool that determines the use. In a free society, tools like this can have a net positive effect, so long as the public has equal access to the information generated.
From what I've seen the mechanisms are even simpler -- they filter out events that weren't heard by more than one sensor. There are patents online for some of these systems, but they seem to use simple heuristics.
Nobody seems to hold a significant patent on these techniques, so a number of companies make acoustic and radar, etc. gunshot location systems. The interesting thing about the military ones is that instead of pointing a camera at the shooting, they point a howitzer. Much more effective, IMHO.
Slashdot Mirror
This is apparently all because HTTP has a referer field (unless the user turns it off in the browser), so clickthroughs on ads have the url you were on when you clicked. FB has lots of urls with user id's in them, which lead to pages with the user's public information, friends, etc. Researchers have already crawled most of these urls without much trouble, but the definition of "giving away private information" seems to have changed a bit under the influence of lawyers.
Nowadays it looks like FB puts ad clickthroughs through a redirect that hides the referer. I suppose the WSJ will discover next that clickthroughs reveal the user's IP address and blame FB for it.
3) Farmville would be happy to make accounts portable. Game makers aren't happy about being dependent on Facebook either.
This should be able to serve over 2000 popunder ads per second.
They spent most of their money building nuclear weapons instead of investing in useful public utilities, like food. Even their native energy sources like coal are sold to China for foreign exchange.
The product placements.
Nokia and Budweiser? What a depressing future.
No way. My Macbook has the MAC printed on the label inside the battery compartment. Obviously it can't be changed.
Where is Mrs. Z?
FWIW, I had lunch with the Netflix search guy a while ago and told him the same thing. For me it's Korean. I get recommended all kinds of unrelated stuff.
I think a lot of the metadata comes from Hollywood's view of the world, so "foreign" it is.
Come to think of it, I even had to return one Korean film because it had no Korean soundtrack -- it was a Hong Kong release dubbed in Chinese, with English subtitles. I had some serious doubts about my Korean skills listening to that one.
There's one Korean film I rented a while ago that has a lot of Japanese: "The President's Last Bang". Good film.
That's about what I expected. I've worked with a similar entity-extraction tool, and it's largely a process of twiddling the rules until it gets most of what you think is in there. The regex's, etc., used to find the entities are so large that the engine can easily bog down, and it's a complete pain to have minor changes balloon into 2x or worse performance hits.
One shortcoming is the lack of interactivity on large datasets. Most web searchers iterate through a few queries until they get what they want, but in this case the iterations could take days.
What's weird is that they actually presented this at an internet conference. Is it calculated media whoring, or just mathematical illiteracy?
But the USB cables are $50,000.
Is Naver technically a better search engine? In other words, does it do better with segmentation and stemming than Google? I'm curious because I speak some Korean but I don't have enough experience to notice problems like these. I assume you're seeing search results that miss or confuse certain forms of a word, or fail to find stem words embedded in particles or verb endings.
There are only a few companies that make Korean stemmers, and I'm familiar with two of them. Each uses a large staff to hard-code stemming and segmentation cases for hundreds of thousands of words. The dictionaries that result run into hundreds of megabytes, but they usually do a decent job, or as good a job as anybody else.
Google, IIRC, was using one of these products to do its Korean indexing, so I'm surprised that there are quality differences.
Unfortunately, these efforts will be useless, as unimpeachably scientific studies have shown. Cameras do not eliminate politics; they merely push military regimes into other areas. Democracy is doomed, I tell you.
There should be more than enough immigrants available to satisfy food demand. The management company, Soylent, Inc., always has plenty of people on hand for harvesting and processing.
This is an old story -- hunter/gatherers displaced by farmers and traders. The hunters always complain, until the farmers introduce them to beer.
The calls are coming from inside the house!
It's obvious he'd be a great ambassador to the third world -- you know, developing countries.
Unfortunately we have a large contingent of people who read Orwell the way fundamentalists read the Bible.
He seems to be popularizing a lot of research, but it's the type of thing many of us do naturally. For me, chasing down hidden indicators is pretty much par for the course. I live in San Francisco, and the city is so routinely out of whack, corrupt, negligent, etc., that I practically maintain my own data warehouse of performance data.
His "surprising facts" were not that surprising. The fact that many crack dealers live with their mothers? That's a no-brainer. Mom doesn't have a criminal record, so mom is eligible for public housing, and junior waltzes in without a lease. Free housing, plus a layer of deniability for eviction court. Public housing lessees are something like 65% female.
Likewise, in a town with little effective prosecution, how do you get rid of drug dealers? Get them parking tickets. Then when the parking people don't do their jobs you have to go after meter revenues, file sunshine requests on ticket totals, even go out and count cars at expired meters (around 80%).
How do you get a map of all gunfire incidents in San Francisco? Not from the cops (until recently) or the DA. The Health department publishes one, though.
It's all a game of people protecting their turf, and the public tricking them into revealing their information. These days, anyone can crunch data, so the task is mainly to think of useful measures to apply to available sources. The book wasn't bad, but it didn't teach me much.
I may have seen a .sig about that somewhere.
Finally, a use for overripe kimchi. The smell would be no worse than stir-frying.
In 1922 diodes were vacuum tubes with two wires in them, one of them heated and capable of emitting electrons to pass a current. What he seems to have discovered is an inverse photoelectric effect, where electrons emit light when striking a metal target.
Although, come to think of it, there were Germanium diodes at the time, so it could have been something more semiconductor-related, but the article isn't clear.
The efficiency of an LED die is limited by how many photons actually get out of the material and into the open air (if they don't, they reflect internally and turn into heat). Many of the recent improvements in LED efficiency have come about through better light transmission, which requires a careful transition from the high-index-of-refraction die to low-index-of-refraction air.
Security cameras are a tool. They're actually a less oppressive tool than people think, because recorded information can often be subpoenaed or requested through freedom-of-information laws.
It's not the tool that determines the use. In a free society, tools like this can have a net positive effect, so long as the public has equal access to the information generated.
From what I've seen the mechanisms are even simpler -- they filter out events that weren't heard by more than one sensor. There are patents online for some of these systems, but they seem to use simple heuristics.
Nobody seems to hold a significant patent on these techniques, so a number of companies make acoustic and radar, etc. gunshot location systems. The interesting thing about the military ones is that instead of pointing a camera at the shooting, they point a howitzer. Much more effective, IMHO.