I just read Positively Fifth Street and agree it is quite enjoyable. Other books in the "professional write/amateur poker player goes to Binion's for the World Series of Poker" genre include Anthony Holden's Big Deal and A. Alvarez's The Biggest Game in Town (both of which McManus freely admits inspiration from). I actually think Big Deal is slightly better written than Positively Fifth Street (McManus spends a bit too much time being gonzo in his writing), but of course McManus does better in his foray into no limit Hold 'Em.
Although McManus spends a bit less time than the others explaining how a poker player thinks, his glossary is actually better so you can follow along with phrases like "I got sucked out by the case nine on the river".
- adam
P.S. If you are instead a fan of the "gamble with your writing advance in Vegas" genre, 24/7 by Andres Martinez is pretty good.
OK, I admit the "Turing Test" reference was not a good example of AI. Also I did not state that AI was a "failure" and you are the one who implied that saying it was "academic" was an insult.
HOWEVER I stand by my statement that the vast majority of Microsoft people don't deal with AI issues and consider AI to be nothing they have to worry about. They are not writing spam filters or expert systems. They are dealing with pretty basic issues: you have an API you can call, you have an API you have to provide, fill in the gap.
My point was that a) Microsoft people don't spend much time at all trying to justify to themselves that the interview questions are effective, and within that, b) if they do spend any time, they don't use an analogy with AI. I thought the comparison to AI that Poundstone brought up was interesting, but I don't think it is generally correct (one of the reasons I thought it was interesting was because I had never heard it before).
I agree AI suffers from a "I don't know what it is -- but it sure as heck isn't anything that I've seen so far" problem among tech people. I think people have seen problems they thought could only be solved by simulating humans (such as playing chess) instead solved by brute force computation. So the excuse then is, "Yes I said if a computer could do that it would be AI, but now I changed my mind."
The Economist is supposed to be running a related article by Poundstone, and I even got email from someone from Reader's Digest who was fact-checking an extremely condensed version of the book. Both of those will probably come out closer to the book's official May 1 release date.
I googled this and also found it reported both ways. I personally had always thought she was his niece so I ran with that. Anyway I think people know the "famous female-relative-of-some-sort of Lord Byron who had something to do with early computers" that you are talking about.
This is actually mentioned in chapter 1 of Games for the Superintelligent, not as a puzzle for the reader, but as an alleged example of a student getting in trouble for thinking outside the box. The three answers he mentions, which I will obfuscate to keep them secret a bit longer, are the "standard" one, the "drop" one, and the "owner" one.
The article seems to focus on that, basically saying can we have only one data type (since all the rest are just speed optimizations), then some vague notion of allowing parallelism, plus mentions that software should be written in layers (which seems to have little to do with the language).
I had high hopes for this article, but there doesn't seem to be much there. I would think the main goal of the hundred-year programming language would be to reduce the number of *bugs* in programs. Solve that, and other details become much less important.
And as an aside, I think strong typing probably helps more than hurts with that. This guy was talking at a Python conference so presumably he likes the "list that is an element of a dictionary that has tuples as the key" etc you can do so easily in Python, but it does make it easy to pass the wrong variable in to a function (e.g. you have a list of lists and you forget to index into the outer list).
Hmmm, I read the outline and didn't see the chapter about "What if it's just a big ol' fad and we're all smoking crack?"
I like his intros to each paragraph summary. "Something happened! Clueless old media farts ignored it! Hip, happenin' bloggers circle-linked it to death! Hilarity and/or new paradigms ensued!"
I suppose if I actually felt it would make any difference, I would write a long piece about why this will not fundamentally change journalism, and submit it to Gillmor. But I suspect he is already too far down the path for it to have much of an effect.
You only need one piece of evidence: Brad Silverberg, former Microsoft Vice President,
works for a
wireless venture capital company.
This is from the "Investment Focus" page of their website: "We believe the potential exists to build large companies based on disruptive technologies and shifts in the value chain of usage and deployment in these sectors."
Conclusion: Wireless is more over-hyped than the dot-coms were.
HTML is good because they can see immediate results and show their website to a friend, and learn a lot from doing "View Source".
Lego Mindstorms is another one that is simplified but gives impressive results.
Beyond that I would look for something that let them write games with graphics easily. Don't have any great suggestions though. BASICA for the original IBM PC, which is where I wrote my first games, was pretty good, but I think with windowing systems, doing any graphics has gotten so complicated that there might not be any system where it is simple to do graphics. But definitely writing games is a good way to learn, even if it is just "guess what number I am thinking of" in Python.
- adam
Windows 2003 Server is due to release on April 24
on
Windows 2003 Going Gold
·
· Score: 2, Interesting
At least, this is true according to the giant banners with that date hung all over Microsoft's main campus, plus the digital sign near Building 26 that is counting down the days until it ships. It's for 2 other products besides Server, which I forget (Visual Studio and SQL, maybe?).
I am wondering if the PXE boot actually got some response from the network, but only a partial one in that it was told in the PXE response what file to load, but then the file did not exist on the server.
You should try it with a network sniffer running...or else try it with the network cable disconnected and see if it behaves the same way.
Also if it happens again, check if the boot sector really has changed. I also doubt that the PXE ROM would have an NT boot sector sitting somewhere that it could burn onto the hard drive (never mind actually doing so).
For end users, the GUI is great. But for servers (which is the matter at hand here), a command-line lets you easily administer a machine remotely and automate repetitive tasks. Of course you can still have a GUI for editing, etc. as Windows and Linux do. But Microsoft's neglect of the command-line has been one of the main factors that led to the rise of Linux.
It's nice Microsoft has decided not to use the GPL as the main focus of their anti-Linux activities. But if the new plan is to harp on "it's based on Unix and it uses a command line" then they won't do much better.
"I still believe Linux is an extension of the Unix paradigm. It's a command-line-focused approach that's not particularly designed to be user friendly. The Windows approach is very different."
Right! The Windows approach is much *worse* than the command line for high-end server administration. Hard to believe that this guy is in charge of Windows server marketing (or maybe not so hard to believe).
Of course someone who equates Zelazny and Piers Anthony might have strange tastes...but I would recommend Jack Vance. The "Demon Princes" books are some of the best SF out there, and "Planet of Adventure" isn't too bad either.
Are here, somebody actually tested things like does it matter if you debur the axles and stuff. I got to that from this site, which has some good general tips.
I just weighed in my son's car last night, now it is "impounded" until the trial on Saturday.
I think most people know about graphite, etc. so it may not give an advantage (but of course is needed!). One thing you should do is hold the car on its side so the wheel is resting on the end of the axle/nail. Then spin the wheel. It should spin pretty freely. Then do the same with the car flipped so the wheel is resting on the body of the car. If it doesn't spin freely in both those situations, then smooth it out, more graphite, etc.
One thing I did was put a bit of paint on the body right where the wheel touched it, then sprinkled graphite on the paint when it was still wet. We'll see if that helps. Excuse me, I meant to say MY SON did that.
Think about how Microsoft makes operating systems (no really). There is an extensive design phase involving meetings with users and potential users. Then there is coding and unit testing. Then the whole thing is integrated and tested exhaustively on as wide a hardwarde matrix as can be found. Documentation is written, and after the product ships, support is done through product support, building up the knowledge base, etc. Finally, developers are encouraged to write software for Windows through Microsoft's evangelism efforts.
So open source has the code/unit testing phase down, and now this attempts to solve the design part. But I think to truly compete with Windows, Linux has to handle the others. So, given sufficient machines, does wide-scale regression testing become shallow? Given enough authors, does documentation become shallow? Given enough online advocates, does support become shallow? Given enough technical users, does evangelism become shallow?
OK so some of those weren't the most elegantly stated, but hopefully you get the idea. It's the way Microsoft has done it, and though you might dispute the quality of some of the components as produced by Microsoft, I don't think you can dispute the advantages of all of them being there.
I actually worked on remote install of Windows 2000 when I was at Microsoft. Remote install meant you only booted setup off the network, then installed to a local hard drive. But it's not a huge step from there to remote boot, although Windows 2000 doesn't support that (don't know about XP and future products).
In terms of PXE hardware, you probably want a CardBus card, not a PCCard (which is what PCMCIA was renamed to). PCCard is 16 bit data path and cards are identified by a 64-character text string or something usly like that...PCCard is 32 bit data path and devices appear like PCI devices and are identified like PCI devices (I forget the details, but it's something like a 16-bit manufacturer ID and a 16-bit ID for that particular type of card).
Back in early 2000 or so, we had a PXE-compliant CardBus network adapter (not wireless, but that shouldn't matter to the software level) in our lab that would do remote install of Windows 2000. In fact we had to make zero changes to the code, it worked like any PXE-compliant PCI network card. So if you could find a PXE-compliant CardBus wireless network adapter, you should be able to do a remote install of Windows 2000/XP on it today. Of course this requires setting up a Windows server to hand out the images, etc. which there is undoubtedly a Microsoft white paper on somewhere.
This is a little article I wrote a while ago called "Can We Improve Computer-to-Human Bandwidth?" which I haven't done anything with...so I might as well post it here:
--------------- begin article --------------
I bet I can guess something about you: right now you are reading something on your computer screen. The text is displayed on a display set near eye level, probably in black text on a white background, or white text on a black background. You read all the text that is visible on your screen, then you press a key or click a mouse button to scroll down to see more text.
Was I right?
Since the early days of computing, fifty years ago, that is the way data has been transmitted from computers to people. The improvements have been quite modest, involving sharper displays, more readable fonts, better choice of foreground and background colors, and so on.
In the same time period, there have been many attempts to improve how data flows the other way, from people to computers. Different keyboards layouts have been designed. Voice recognition may be just around the corner. The mouse has changed how data is input, possibly not speeding it up for power users, but enabling a whole new class of users to communicate with a computer at all.
Data flow in the other direction has remained the same, an exact simulation of reading text on a printed page. Yet computers are much more powerful than a printed page. Is it time to take advantage of this? How could this be done?
Certainly the real limit on how fast people can read is how fast they can process the underlying information. But some part of a reader's brain is occupied with deciphering the text on the screen. For some dense texts that percentage will be trivial, but for many others it won't be, so the question becomes how much of that can be removed, getting people closer to their theoretical limit.
One change that already exists is to have computers read the text out loud. Unfortunately, while most people can speak much faster than they can type (or write), it is doubtful that most people can listen faster than they can read. One reason is that spoken language, with its elided sounds and lack of spelling, is less informationally dense than written language. Thus it is faster for a person to speak than to spell, but slower for he or she to listen than to read. While computer reading is a boon for people with certain disabilities, it does not speed up how fast data flows from computer to person.
A more radical idea would be to reconsider why the text stays still and the user's eyes move. Why not scroll the text so the eyes can stay still? Of course the computer would have to adjust the scroll rate for different users. Since your hands aren't doing much of anything when you are reading, so I could imagine reading text that was scrolling by with one hand on the mouse, with the left button slowing down the scroll rate and the right button speeding it up.
What about changing how the text itself is displayed? It's risky to get too far away from this because everyone has a lifetime of training in reading printed text in books. Still you can speculate. What if different parts of speech were color-coded on the fly, or displayed in different fronts, or in a slightly different location on the line? What if the computer compressed certain words as they appeared (such as compressing George W Bush to GWB - the reverse of a trick that writers use: typing frequently-used phrases in shorthand, then going back and replace them later, or letting Word's auto-correct feature do it for them). This may be disconcerting at first, but it may turn out that with practice, this can improve the transmission speed for people who need to quickly digest a lot of information coming at them from their computer.
Moving beyond text, consider the fact that a sign language translator can keep up with spoken language, and is also limited in speed by the need to move hands and arms around. One of the advantages of sign language is that location within space can be used to convey information; for example a room can be laid out visually and then movement within that room conveyed by changing where the signs are shown. Could computers use a similar trick on the screen to speed up how fast information is displayed? It could be a lot of work to learn how to interpret this, just as learning sign language is a lot of work, but the payoff could be worth it.
The main thing is to get out of the mindset that static text on a screen is necessarily the best way to present information. Once that assumption is shattered, interesting ideas should follow.
It looks like a Palm Pilot, but it's actually a
notepad. Even comes with fake stylus.
- adam
P.S. WARNING: This company is in Redmond, thus any money you spend there might wind up getting funnelled through a business tax and eventually pay for a roadway lane painting that might be used to get an employee to Microsoft -- so use your best judgement.
What you want is to take the best of the command line and the GUI together. OK the "best" of the GUI ain't a whole lot. But the idea is you start with a GUI-like screen, with a primary command-line at the bottom always waiting to accept a new command. Any commands that you type are automatically launched into their own little command windows (unless just launching a GUI app, I guess) -- but the primary command line still has the history of the commands available, standard shell stuff, etc. Then those little command windows are now stored as icons around the screen. If you open one, you can see the output of that command the last time it ran, and easily run it again if you want (with some standard key). Right-clicking on one of those little command windows gives you options, beyond re-running it, like editing the output in an editor, pasting the output into a new command (useful for using the output of something like "where" as an argument to your next command), mailing the output, feeding the output back into a new command (sort of like "take this previous command's output, feed it into a pipe, and connect the pipe as input to the next command I type in my primary command window"), etc.
So you wind up with a command line prompt, but it uses the GUI space to handle things that get a bit clunky in a regular command line prompt, like history, pipes, etc.
Needs work, but it's clear in my mind...that's the basic idea anyway.
Since it is "backed" by something and you have to create an account, it seems like there is a central repository (at e-gold's site).
The problem with.5% of 5000 3 cent transactions is that they cost 5000 times as much to record and store as a single.5% of 3 cents transaction. The credit cards work because most transactions are not for 3 cents but instead for enough that they can take a reasonable per-transaction cut. If every transaction was 3 cents the credit cards would lose lots of money (ignoring interest payments, their other source of income).
Digital money, that is a digital equivalent of cash, is impossible. It would have about the same value that real money would have if anyone could duplicate it at will -- that is, none at all.
What I mean is, having some magic string of bits that stands alone, with no link to a central server etc., and can be swapped around, will never happen. Because whatever those strings of bits are, they can be duplicated digitally.
So I think it is better to talk about digital checks or digital credit cards or whatever. So you pass around some blob, but to validate that blog and assign it a new owner, you have to go to some central repository.
Then you have to ask, what would happen to the checking system or the credit card system if they had a bunch of tiny transactions, checks for 5 cents and credit card payments of 3 cents? Well, either the central repositories would complain (because their percentage cut was too small per transaction) or the users would complain (because paying 14 cents for a virtual check for 3 cents makes about as much sense as paying 14 cents for a paper check and then writing it for 3 cents) or businesses getting paid would complain (for either of those two reasons).
Thus, the only way micropayments could work is without a central repository, using pure stand-alone digital money. And since stand-alone digital money is impossible, micropayments won't ever work.
Slashdot Mirror
Although McManus spends a bit less time than the others explaining how a poker player thinks, his glossary is actually better so you can follow along with phrases like "I got sucked out by the case nine on the river".
- adam
P.S. If you are instead a fan of the "gamble with your writing advance in Vegas" genre, 24/7 by Andres Martinez is pretty good.
Mostly I replied so your post would have a reply to it, which might inspire more people to read it.
- adam
- adam
HOWEVER I stand by my statement that the vast majority of Microsoft people don't deal with AI issues and consider AI to be nothing they have to worry about. They are not writing spam filters or expert systems. They are dealing with pretty basic issues: you have an API you can call, you have an API you have to provide, fill in the gap.
My point was that a) Microsoft people don't spend much time at all trying to justify to themselves that the interview questions are effective, and within that, b) if they do spend any time, they don't use an analogy with AI. I thought the comparison to AI that Poundstone brought up was interesting, but I don't think it is generally correct (one of the reasons I thought it was interesting was because I had never heard it before).
I agree AI suffers from a "I don't know what it is -- but it sure as heck isn't anything that I've seen so far" problem among tech people. I think people have seen problems they thought could only be solved by simulating humans (such as playing chess) instead solved by brute force computation. So the excuse then is, "Yes I said if a computer could do that it would be AI, but now I changed my mind."
- adam
- adam
- adam (author of the book review)
- adam
I had high hopes for this article, but there doesn't seem to be much there. I would think the main goal of the hundred-year programming language would be to reduce the number of *bugs* in programs. Solve that, and other details become much less important.
And as an aside, I think strong typing probably helps more than hurts with that. This guy was talking at a Python conference so presumably he likes the "list that is an element of a dictionary that has tuples as the key" etc you can do so easily in Python, but it does make it easy to pass the wrong variable in to a function (e.g. you have a list of lists and you forget to index into the outer list).
- adam
I like his intros to each paragraph summary. "Something happened! Clueless old media farts ignored it! Hip, happenin' bloggers circle-linked it to death! Hilarity and/or new paradigms ensued!"
I suppose if I actually felt it would make any difference, I would write a long piece about why this will not fundamentally change journalism, and submit it to Gillmor. But I suspect he is already too far down the path for it to have much of an effect.
- adam
This is from the "Investment Focus" page of their website: "We believe the potential exists to build large companies based on disruptive technologies and shifts in the value chain of usage and deployment in these sectors."
Conclusion: Wireless is more over-hyped than the dot-coms were.
- adam
Lego Mindstorms is another one that is simplified but gives impressive results.
Beyond that I would look for something that let them write games with graphics easily. Don't have any great suggestions though. BASICA for the original IBM PC, which is where I wrote my first games, was pretty good, but I think with windowing systems, doing any graphics has gotten so complicated that there might not be any system where it is simple to do graphics. But definitely writing games is a good way to learn, even if it is just "guess what number I am thinking of" in Python.
- adam
- adam
You should try it with a network sniffer running...or else try it with the network cable disconnected and see if it behaves the same way.
Also if it happens again, check if the boot sector really has changed. I also doubt that the PXE ROM would have an NT boot sector sitting somewhere that it could burn onto the hard drive (never mind actually doing so).
- adam
It's nice Microsoft has decided not to use the GPL as the main focus of their anti-Linux activities. But if the new plan is to harp on "it's based on Unix and it uses a command line" then they won't do much better.
- adam
Right! The Windows approach is much *worse* than the command line for high-end server administration. Hard to believe that this guy is in charge of Windows server marketing (or maybe not so hard to believe).
- adam
- adam
I just weighed in my son's car last night, now it is "impounded" until the trial on Saturday.
I think most people know about graphite, etc. so it may not give an advantage (but of course is needed!). One thing you should do is hold the car on its side so the wheel is resting on the end of the axle/nail. Then spin the wheel. It should spin pretty freely. Then do the same with the car flipped so the wheel is resting on the body of the car. If it doesn't spin freely in both those situations, then smooth it out, more graphite, etc.
One thing I did was put a bit of paint on the body right where the wheel touched it, then sprinkled graphite on the paint when it was still wet. We'll see if that helps. Excuse me, I meant to say MY SON did that.
- adam
So open source has the code/unit testing phase down, and now this attempts to solve the design part. But I think to truly compete with Windows, Linux has to handle the others. So, given sufficient machines, does wide-scale regression testing become shallow? Given enough authors, does documentation become shallow? Given enough online advocates, does support become shallow? Given enough technical users, does evangelism become shallow?
OK so some of those weren't the most elegantly stated, but hopefully you get the idea. It's the way Microsoft has done it, and though you might dispute the quality of some of the components as produced by Microsoft, I don't think you can dispute the advantages of all of them being there.
- adam
In terms of PXE hardware, you probably want a CardBus card, not a PCCard (which is what PCMCIA was renamed to). PCCard is 16 bit data path and cards are identified by a 64-character text string or something usly like that...PCCard is 32 bit data path and devices appear like PCI devices and are identified like PCI devices (I forget the details, but it's something like a 16-bit manufacturer ID and a 16-bit ID for that particular type of card).
Back in early 2000 or so, we had a PXE-compliant CardBus network adapter (not wireless, but that shouldn't matter to the software level) in our lab that would do remote install of Windows 2000. In fact we had to make zero changes to the code, it worked like any PXE-compliant PCI network card. So if you could find a PXE-compliant CardBus wireless network adapter, you should be able to do a remote install of Windows 2000/XP on it today. Of course this requires setting up a Windows server to hand out the images, etc. which there is undoubtedly a Microsoft white paper on somewhere.
- adam
This is a little article I wrote a while ago called "Can We Improve Computer-to-Human Bandwidth?" which I haven't done anything with...so I might as well post it here:
--------------- begin article --------------
I bet I can guess something about you: right now you are reading something on your computer screen. The text is displayed on a display set near eye level, probably in black text on a white background, or white text on a black background. You read all the text that is visible on your screen, then you press a key or click a mouse button to scroll down to see more text.
Was I right?
Since the early days of computing, fifty years ago, that is the way data has been transmitted from computers to people. The improvements have been quite modest, involving sharper displays, more readable fonts, better choice of foreground and background colors, and so on.
In the same time period, there have been many attempts to improve how data flows the other way, from people to computers. Different keyboards layouts have been designed. Voice recognition may be just around the corner. The mouse has changed how data is input, possibly not speeding it up for power users, but enabling a whole new class of users to communicate with a computer at all.
Data flow in the other direction has remained the same, an exact simulation of reading text on a printed page. Yet computers are much more powerful than a printed page. Is it time to take advantage of this? How could this be done?
Certainly the real limit on how fast people can read is how fast they can process the underlying information. But some part of a reader's brain is occupied with deciphering the text on the screen. For some dense texts that percentage will be trivial, but for many others it won't be, so the question becomes how much of that can be removed, getting people closer to their theoretical limit.
One change that already exists is to have computers read the text out loud. Unfortunately, while most people can speak much faster than they can type (or write), it is doubtful that most people can listen faster than they can read. One reason is that spoken language, with its elided sounds and lack of spelling, is less informationally dense than written language. Thus it is faster for a person to speak than to spell, but slower for he or she to listen than to read. While computer reading is a boon for people with certain disabilities, it does not speed up how fast data flows from computer to person.
A more radical idea would be to reconsider why the text stays still and the user's eyes move. Why not scroll the text so the eyes can stay still? Of course the computer would have to adjust the scroll rate for different users. Since your hands aren't doing much of anything when you are reading, so I could imagine reading text that was scrolling by with one hand on the mouse, with the left button slowing down the scroll rate and the right button speeding it up.
What about changing how the text itself is displayed? It's risky to get too far away from this because everyone has a lifetime of training in reading printed text in books. Still you can speculate. What if different parts of speech were color-coded on the fly, or displayed in different fronts, or in a slightly different location on the line? What if the computer compressed certain words as they appeared (such as compressing George W Bush to GWB - the reverse of a trick that writers use: typing frequently-used phrases in shorthand, then going back and replace them later, or letting Word's auto-correct feature do it for them). This may be disconcerting at first, but it may turn out that with practice, this can improve the transmission speed for people who need to quickly digest a lot of information coming at them from their computer.
Moving beyond text, consider the fact that a sign language translator can keep up with spoken language, and is also limited in speed by the need to move hands and arms around. One of the advantages of sign language is that location within space can be used to convey information; for example a room can be laid out visually and then movement within that room conveyed by changing where the signs are shown. Could computers use a similar trick on the screen to speed up how fast information is displayed? It could be a lot of work to learn how to interpret this, just as learning sign language is a lot of work, but the payoff could be worth it.
The main thing is to get out of the mindset that static text on a screen is necessarily the best way to present information. Once that assumption is shattered, interesting ideas should follow.
---------------- end article ---------------
- adam
- adam
P.S. WARNING: This company is in Redmond, thus any money you spend there might wind up getting funnelled through a business tax and eventually pay for a roadway lane painting that might be used to get an employee to Microsoft -- so use your best judgement.
So you wind up with a command line prompt, but it uses the GUI space to handle things that get a bit clunky in a regular command line prompt, like history, pipes, etc.
Needs work, but it's clear in my mind...that's the basic idea anyway.
- adam
The problem with .5% of 5000 3 cent transactions is that they cost 5000 times as much to record and store as a single .5% of 3 cents transaction. The credit cards work because most transactions are not for 3 cents but instead for enough that they can take a reasonable per-transaction cut. If every transaction was 3 cents the credit cards would lose lots of money (ignoring interest payments, their other source of income).
- adam
- adam
What I mean is, having some magic string of bits that stands alone, with no link to a central server etc., and can be swapped around, will never happen. Because whatever those strings of bits are, they can be duplicated digitally.
So I think it is better to talk about digital checks or digital credit cards or whatever. So you pass around some blob, but to validate that blog and assign it a new owner, you have to go to some central repository.
Then you have to ask, what would happen to the checking system or the credit card system if they had a bunch of tiny transactions, checks for 5 cents and credit card payments of 3 cents? Well, either the central repositories would complain (because their percentage cut was too small per transaction) or the users would complain (because paying 14 cents for a virtual check for 3 cents makes about as much sense as paying 14 cents for a paper check and then writing it for 3 cents) or businesses getting paid would complain (for either of those two reasons).
Thus, the only way micropayments could work is without a central repository, using pure stand-alone digital money. And since stand-alone digital money is impossible, micropayments won't ever work.
- adam