"The US constitution forbids ex-post-facto laws" - the generally accepted interpretation of the prohibition on ex-post facto laws is that Congress may not make something illegal after-the-fact; this does not, however, prevent them from retroactively making it legal.
If theory says that black holes beyond 10 solar masses cannot form, how do they explain the conjectured supermassive black holes at the center of our and other galaxies?
"That said, I wonder if they took reversions into account in their analysis." - yes, they did, in a negative way. If you have an otherwise high trust metric and revert someone's edit back to the old version, then your retention rate as a percentage of characters divided by total edits goes down - in other words, you become less trust worthy after doing anti-vandal patrolling.
This is a troll, and is patently false, but I'll bite nonetheless: the Foundation's privacy policy (which governs the use of checkuser) strictly limits the conditions under which "personally identifiable data collected in the server logs, or through records in the database via the CheckUser feature" may be released. The release of aggregated, anonomyized data, such as I did above, is perfectly acceptable under the privacy policy and is a common practice in web traffic analysis
Doing such a study requires checkuser access, which is something only a few people on Wikipedia have. Fortunately, I am one of them. I just sampled ten users out of the new user log. I am assuming a 1:1 mapping between IP and user (that is, that a user made no anonymous edits except with the IP he used to register his account). The number of anonymous edits prior to registeration for each user was:
A - 0 B - 0 C - 0 D - 2 E - 0 F - 0 G - 0 H - 0 I - 0 J - 0
In short: most of the people registering accounts had made no edits prior to registering. It's common knowledge on Wikipedia that something like half of all accounts registered never make any edits at all, so this makes sense.
The graph on page 31 is the retention rate of characters contributed by an editor relative to the total edits by that editor to the article ("The dependent variable is the retention rate, R, of contributions, measured as the percent of characters retained per contribution by each contributor.")
This metric makes sense if the wiki is new, and most of the edits are adding new content. The metric is virtually meaningless if the wiki is established, and most of the edits by a group of people are vandalism or reverts - people fixing the article will have a lower score by virtue of the fact that they are making the same edit (more or less) over and over again.
Normally, you'd expect that the more edits a user makes, the more trustworthy he is. If he were vandalizing, he wouldn't make more than a few before being blocked. If he's making hundreds, he should be considered more trust worthy (and have a higher retention rate) than if he's new. The results here show the exact opposite for anonymous users. In short, the methodology is flawed and the results are wrong.
For any game that cannot solved by enumeration, you need to be able to evaluate different decision paths and select a move. In most cases, this computation needs to be done in a reasonable amount of time (like the time it takes a chess player to think of a move). Path evaluation and state comparison is an AI application to game theory.
The article says that the copiers can "understand" the documents, so it's clearly more than a simple search-and-replace. This implies context recognition, which as I previously mentioned, is an extremely difficult problem.
I suppose the product either (A) lives up to the hype, or (B) it does not.
So, which sounds more likely: Either Xerox jumped light years ahead of the field with this product (in which case, A), or they put out a shoddy product that won't live up to the hype (in which case, B). Frankly, I think the latter is a heck of a lot more likely.
On the contrary, I'm do computer engineering research for a living. And don't get me wrong - I think this is a perfectly valid area to research. But a redacting copier is 3 (or more) decades from being a viable product - the technology just isn't there yet. Wildly exaggerated claims leading to disappointment have plagued the AI field for decades, and putting out products like this only contributes to that.
On second thought, I'd like to amend my above statement - OCR is the second easiest application in AI, after game theory. A number of games have been completely solved (Connect 4), effectively solved (like Checkers, announced recently), or are very well done (Chess). Granted, they are not complex games (Find me an AI that can play Twilight Imperium well) but they are not trivial either.
You have missed my point. I don't deny that OCR makes life easier for people who have to digitize documents. The point I am making is that OCR is, as far as AI applications go, the easiest problem there is. And, even with such an easy problem, the best applications out there deliver substantially less than reliable performance. (If you think 99% is OK, then imagine that for a 100,000 word novel, at 4 characters per word, that's 2,000 words that need fixing).
Now, with this copier, you are talking about a *substantially* harder problem, which has far less tolerance for errors. (Meaning that you want absolutely no false negatives) The chances that this copier works as advertised, or anywhere close to it, is basically nil. It was a waste of money for Xerox to develop it (because anyone even moderately knowledgeable about AI should have been able to tell them this) and it's a waste of money for anyone who buys it.
AI is a disaster through-and-through. It never works well. Ever.
Consider hand-writing recognition, autonomous robotics, and game theory, just to name a few of the narrowest, most-well defined (read:easiest) AI applications. AI works well in none of these - at best, it's so-so (like the 95-98% success rates in OCR).
Now what you have here, with the automatic redacting copier, is that the copier needs to understand the document its reading, and determine which parts to redact. Contextual understanding is *HARD* - it's the same class of problem as automated translation - only harder in this case.
This copier idea is a huge flop. I don't know why they waste money on it. Anyone who relies on this copier to redact documents is a fool, because it is bound to make all kinds of mistakes (both type 1 - missing things it should have picked up, and type 2 - redacting things it shouldn't).
I'd just like to say that the guy responsible for this - Sheldon Rampton (author of "Toxic Sludge Is Good for You!: Lies, Damn Lies and the Public Relations Industry" and "Bananna Republicans: How the Right Wing is Turning America Into a One-Party State") is awesome. His books are very informative (my mother made his first book required reading for her environmental science class). After meeting him in person at Wikimania, I think he's awesome.
Try nominating something on Wikipedia's Featured Article Candidates. The rules there (which, admittedly, I wrote) are expressly designed to avoid person-person conflicts and to improve articles.
Explanations from a hardcore Wikipedian
on
Has Wikipedia Peaked?
·
· Score: 2, Interesting
For about 3 years, I've been manually keeping track of some statistics that are important to me here.
Some of Dragonfly's (the person who created those graphs) observations are fairly easy to explain, others require some knowledge of the site. Since I've been at Wikipedia (starting back in mid 2003) new article production has gone through 3 phases: (1) First it was super-linear. That is, each month, we produced slightly more new article than the previous month. Many people predicted that this would ultimately become exponential, and (2) eventually exponential growth is what we got. However, since last August, (3) that has mostly flattened out, to a relatively constant 40k-60k new article per month. I think the answer why is pretty obvious - all of the low-hanging fruit is long gone. When I started editing, there were lots of red links (links to articles that don't exist) that any non-expert might be able to churn out in 2 minutes. Many of the new articles I create nowadays are highly esoteric, some of which I created after seeing them mentioned in journal papers I was peer reviewing. (Examples: Gustafson's law, Antigenic escape).
As far as new account registration, that's a bit more complex to explain. First and most obviously, Wikipedia is not new anymore. We're not going to see the kind of new-user account registrations that we used to. But there's another, more complicated factor at work. For about 9 months (March to December 2006), there existed a technique to vandalize Wikipedia with impunity. You register lots of accounts, and then use each one to vandlize exactly once, log out, log back in with another accout and vandalize, etc. Mediawiki did not block your IP unless you attempted to register from a blocked account, so by editing with each one exactly once you avoided ever having your IP blocked. The only effective way to combat this was to have checkuser access (which I have, but I'm one of only about 10 people on the English Wikipedia who do) I filed a bug report, which was fixed in December 2006. I suspect that a lot of the drop-off in user account registration has to do with this bug being fixed. Registering 100 throw-away accounts was no longer effective, so people did not do it, therefore - I suspect- account registration went down.
"Maybe drilling hole into people's heads is about as useful as calling a witch doctor to perform magical healing rituals. I am personally NOT impressed by trepanation, unless it's a competent modern surgeon who performs it as the first step of a REAL medical procedure."
Then you would be wrong. Removing part of the skull (a craniotomy) is the only way to treat a subdural hematoma - an *EXTREMELY* dangerous condition where blood starts collecting in the brain, putting pressure on the brain tissue there. See this and this. (If memory serves, in Star Trek IV, they stop a doctor just before he's about to perform a craniotomy on Checkov to treat a subdural hematoma)
So the dependency tree looks like this: Samba --OpenChange -- KDE
KDE operates on top of OpenChange, which operates on top of KDE. From the write up, OpenChange is "heavily dependent on Samba code for the underlying protocol support". So they mix in outside (Samba) code which changed license, and thus are forced to change the license. Fair enough - they want the free lunch, they have to use the license specified by the guy who wrote the code.
What doesn't make any sense is that OpenChange cannot support KDE now. Of course they can. As long as they don't share any code, they can be licensed independently.
Slashdot Mirror
"The US constitution forbids ex-post-facto laws" - the generally accepted interpretation of the prohibition on ex-post facto laws is that Congress may not make something illegal after-the-fact; this does not, however, prevent them from retroactively making it legal.
Pages on meta are not policy; Foundation policy pages and statements are. My link is the correct one.
If theory says that black holes beyond 10 solar masses cannot form, how do they explain the conjectured supermassive black holes at the center of our and other galaxies?
Actually, I have one planned (tentatively) for the 25th.
"That said, I wonder if they took reversions into account in their analysis." - yes, they did, in a negative way. If you have an otherwise high trust metric and revert someone's edit back to the old version, then your retention rate as a percentage of characters divided by total edits goes down - in other words, you become less trust worthy after doing anti-vandal patrolling.
This is a troll, and is patently false, but I'll bite nonetheless: the Foundation's privacy policy (which governs the use of checkuser) strictly limits the conditions under which "personally identifiable data collected in the server logs, or through records in the database via the CheckUser feature" may be released. The release of aggregated, anonomyized data, such as I did above, is perfectly acceptable under the privacy policy and is a common practice in web traffic analysis
Doing such a study requires checkuser access, which is something only a few people on Wikipedia have. Fortunately, I am one of them. I just sampled ten users out of the new user log. I am assuming a 1:1 mapping between IP and user (that is, that a user made no anonymous edits except with the IP he used to register his account). The number of anonymous edits prior to registeration for each user was:
A - 0
B - 0
C - 0
D - 2
E - 0
F - 0
G - 0
H - 0
I - 0
J - 0
In short: most of the people registering accounts had made no edits prior to registering. It's common knowledge on Wikipedia that something like half of all accounts registered never make any edits at all, so this makes sense.
Go write me another hurricane featured article, Titoxd :)
The graph on page 31 is the retention rate of characters contributed by an editor relative to the total edits by that editor to the article ("The dependent variable is the retention rate, R, of contributions, measured as the percent of characters retained per contribution by each contributor.")
This metric makes sense if the wiki is new, and most of the edits are adding new content. The metric is virtually meaningless if the wiki is established, and most of the edits by a group of people are vandalism or reverts - people fixing the article will have a lower score by virtue of the fact that they are making the same edit (more or less) over and over again.
Normally, you'd expect that the more edits a user makes, the more trustworthy he is. If he were vandalizing, he wouldn't make more than a few before being blocked. If he's making hundreds, he should be considered more trust worthy (and have a higher retention rate) than if he's new. The results here show the exact opposite for anonymous users. In short, the methodology is flawed and the results are wrong.
For any game that cannot solved by enumeration, you need to be able to evaluate different decision paths and select a move. In most cases, this computation needs to be done in a reasonable amount of time (like the time it takes a chess player to think of a move). Path evaluation and state comparison is an AI application to game theory.
The article says that the copiers can "understand" the documents, so it's clearly more than a simple search-and-replace. This implies context recognition, which as I previously mentioned, is an extremely difficult problem.
I suppose the product either (A) lives up to the hype, or (B) it does not.
So, which sounds more likely: Either Xerox jumped light years ahead of the field with this product (in which case, A), or they put out a shoddy product that won't live up to the hype (in which case, B). Frankly, I think the latter is a heck of a lot more likely.
On the contrary, I'm do computer engineering research for a living. And don't get me wrong - I think this is a perfectly valid area to research. But a redacting copier is 3 (or more) decades from being a viable product - the technology just isn't there yet. Wildly exaggerated claims leading to disappointment have plagued the AI field for decades, and putting out products like this only contributes to that.
See Moravec's paradox
On second thought, I'd like to amend my above statement - OCR is the second easiest application in AI, after game theory. A number of games have been completely solved (Connect 4), effectively solved (like Checkers, announced recently), or are very well done (Chess). Granted, they are not complex games (Find me an AI that can play Twilight Imperium well) but they are not trivial either.
You have missed my point. I don't deny that OCR makes life easier for people who have to digitize documents. The point I am making is that OCR is, as far as AI applications go, the easiest problem there is. And, even with such an easy problem, the best applications out there deliver substantially less than reliable performance. (If you think 99% is OK, then imagine that for a 100,000 word novel, at 4 characters per word, that's 2,000 words that need fixing).
Now, with this copier, you are talking about a *substantially* harder problem, which has far less tolerance for errors. (Meaning that you want absolutely no false negatives) The chances that this copier works as advertised, or anywhere close to it, is basically nil. It was a waste of money for Xerox to develop it (because anyone even moderately knowledgeable about AI should have been able to tell them this) and it's a waste of money for anyone who buys it.
AI is a disaster through-and-through. It never works well. Ever.
Consider hand-writing recognition, autonomous robotics, and game theory, just to name a few of the narrowest, most-well defined (read:easiest) AI applications. AI works well in none of these - at best, it's so-so (like the 95-98% success rates in OCR).
Now what you have here, with the automatic redacting copier, is that the copier needs to understand the document its reading, and determine which parts to redact. Contextual understanding is *HARD* - it's the same class of problem as automated translation - only harder in this case.
This copier idea is a huge flop. I don't know why they waste money on it. Anyone who relies on this copier to redact documents is a fool, because it is bound to make all kinds of mistakes (both type 1 - missing things it should have picked up, and type 2 - redacting things it shouldn't).
I'd just like to say that the guy responsible for this - Sheldon Rampton (author of "Toxic Sludge Is Good for You!: Lies, Damn Lies and the Public Relations Industry" and "Bananna Republicans: How the Right Wing is Turning America Into a One-Party State") is awesome. His books are very informative (my mother made his first book required reading for her environmental science class). After meeting him in person at Wikimania, I think he's awesome.
I don't know which guidelines you're having issues with, but you never know until you try.
I think there's more to the drop in user registration than most people are aware. See my comment here
Whoops - that link should be Featured article candidates
Try nominating something on Wikipedia's Featured Article Candidates. The rules there (which, admittedly, I wrote) are expressly designed to avoid person-person conflicts and to improve articles.
I meant to link to it, but apparently I forgot: here is the bug report I mentioned
For about 3 years, I've been manually keeping track of some statistics that are important to me here.
Some of Dragonfly's (the person who created those graphs) observations are fairly easy to explain, others require some knowledge of the site. Since I've been at Wikipedia (starting back in mid 2003) new article production has gone through 3 phases: (1) First it was super-linear. That is, each month, we produced slightly more new article than the previous month. Many people predicted that this would ultimately become exponential, and (2) eventually exponential growth is what we got. However, since last August, (3) that has mostly flattened out, to a relatively constant 40k-60k new article per month. I think the answer why is pretty obvious - all of the low-hanging fruit is long gone. When I started editing, there were lots of red links (links to articles that don't exist) that any non-expert might be able to churn out in 2 minutes. Many of the new articles I create nowadays are highly esoteric, some of which I created after seeing them mentioned in journal papers I was peer reviewing. (Examples: Gustafson's law, Antigenic escape).
As far as new account registration, that's a bit more complex to explain. First and most obviously, Wikipedia is not new anymore. We're not going to see the kind of new-user account registrations that we used to. But there's another, more complicated factor at work. For about 9 months (March to December 2006), there existed a technique to vandalize Wikipedia with impunity. You register lots of accounts, and then use each one to vandlize exactly once, log out, log back in with another accout and vandalize, etc. Mediawiki did not block your IP unless you attempted to register from a blocked account, so by editing with each one exactly once you avoided ever having your IP blocked. The only effective way to combat this was to have checkuser access (which I have, but I'm one of only about 10 people on the English Wikipedia who do) I filed a bug report, which was fixed in December 2006. I suspect that a lot of the drop-off in user account registration has to do with this bug being fixed. Registering 100 throw-away accounts was no longer effective, so people did not do it, therefore - I suspect- account registration went down.
"Maybe drilling hole into people's heads is about as useful as calling a witch doctor to perform magical healing rituals. I am personally NOT impressed by trepanation, unless it's a competent modern surgeon who performs it as the first step of a REAL medical procedure."
Then you would be wrong. Removing part of the skull (a craniotomy) is the only way to treat a subdural hematoma - an *EXTREMELY* dangerous condition where blood starts collecting in the brain, putting pressure on the brain tissue there. See this and this. (If memory serves, in Star Trek IV, they stop a doctor just before he's about to perform a craniotomy on Checkov to treat a subdural hematoma)
So the dependency tree looks like this:
Samba --OpenChange -- KDE
KDE operates on top of OpenChange, which operates on top of KDE. From the write up, OpenChange is "heavily dependent on Samba code for the underlying protocol support". So they mix in outside (Samba) code which changed license, and thus are forced to change the license. Fair enough - they want the free lunch, they have to use the license specified by the guy who wrote the code.
What doesn't make any sense is that OpenChange cannot support KDE now. Of course they can. As long as they don't share any code, they can be licensed independently.