A million dollars? With that they might be able to buy a tiny house near Google. But then they'd be broke and have to keep working. And somehow, in the mindset of Silicon Valley, that seems normal and rational.
Of course, they could always retire and move someplace sane.
Maybe the Brazilians are dominating Orkut because lots of Americans, like me, have declined all their Orkut invitations. Friendster swept through my circle of friends a while back. Lots of people joined, and then we discovered that there's not actually anything to do there. Once the novelty wore off, we stopped logging in. Then several people (from a different circle of friends) invited me to Orkut, and I thought "why bother?" and declined the invitations.
Once the Brazilians figure out how lame and useless these social networking things are, their numbers will drop.
Provost, J (1999). Naive-Bayes vs. Rule-Learning in Classification of Email. The University of Texas at Austin, Artificial Intelligence Lab. Technical Report AI-TR-99-284
I should mention that I don't think I'm the first to use Naive-Bayes on email. I think some folks from Microsoft did it in 1998, and there may be others too.
I'm not sure what the $20 million is for, since (at least in AI) peer-review is done for free anyway, as a service to the community. The big journals charge money while getting editing, review, and often even typsetting for free from their editorial boards or authors.
Since peer-review is the main service provided by the big journals, it was only a matter of time before the reviewers organized themselves. The tenure issue is a bit of a problem, since untenured faculty will want to publish in the best established journals. However, that should work itself out over time, as tenured researchers choose to publish in the new free journals. Eventually the new journals will be well enough established for young researchers to feel comfortable publishing in them.
There absolutely *is* and archive. All broadcast material is archived on tape
Who cares? It's inaccessible to the viewers. Nobody links to a specific airing of a TV story and expects it to be the same the next time they see it.
I can understand why CNN thinks this is no big deal. CNN was (and is) primarily a TV news station. On TV news, there is no archive or changelog for writethroughs: the copy gets rewritten, and the reporter or anchor reads it on the air. The only way you notice the changes is if you happened to see a previous version of the story earlier in the day.
CNN obviously sees the web as a translation of their TV news business, rather than as a translation of a print-news wire service business, so to them it seems fine! To them the web is a transient medium, like TV, not a fixed medium like print.
Of course, at first glance this seems fine, until linking of stories factors into the equation.
Of course, there are technological solutions to this, but getting CNN to adopt them could be a challenge, because it means converting them from a TV mindset to a print mindset.
The comment about HAL totally misses the point. Of course we won't have voice interfaces to do stuff like highlighting text and scrolling. The idea behind a computer like HAL is that you could treat it like another person.
When you have someone working for you, you don't stand over his shoulder, verbally telling him how to scroll and what to type. You say, "write me a report on X." Then when the report comes, you read it while the employee is working on something else. The report (as well as email, etc) is a visual interface to the employee, which allows you to use your working memory for thinking, rather than talking.
The comment in the article about speech "not carrying the load" of vision still assumes that the human is doing most of the work, and the computer is just a glorified pencil and paper. The idea behind HAL was that humans told him what to do, not how to do it. For that, you want speech.
There's been lots of work on auto-classifying email.
I did my semester project in Machine Learning on this in 1999.
It's a fairly simple study, but it seems like a Naive Bayesian
classifier using word counts as features does a pretty
decent job of classifying email, and does really well on
spam.
Actually, despite what the Intel website says, if you read Moore's paper he clearly predicts that component density will double every 12 months, not every 18-24 months.
The complexity for minimum component costs has in- creased at a rate of roughly a factor of two per year (see graph on next page). (p.2, second column).
The graph on p.3 clearly shows an annual doubling of components (transistors). The current 18-24 month
rate is an updated version of the law, to take into
account the fact that the rate has slowed since the 60's. Of course, this means that this "Law" isn't, in fact, much of a law at all.
The articles only talk about Open Source in terms of companies trying to make money from it. But education, specifically university CS departments, are both huge users and huge resources for the open source community, and will help keep it afloat in hard times.
Not having to buy licenses for much or all of the software on their un*x workstations saves departments huge amounts of money. Moreover, they can build workstations from commodity components. This allows them to provide more machines for students, and simultaneously exposes huge numbers of CS undergrads and grad students to free software.
Also, the dot-com bubble bursting caused CS graduate school enrollments to swell enormously. Grad schools have traditionally been places where much free software is born, as student researchers put their work out there for everyone to see.
The problem is that only a few schools really do research in user interfaces and similar areas that will advance free software in the mainstream. But in a lot of less visible areas: like the core-OS, distributed computing, networking, scientific computing, high-performance graphics, AI and robotics, free software will continue to progress and improve through universities. In the process the universities will continue to graduate students who are used to working with free software, and who will wonder why they should buy licenses for software when so much is available for free.
If you want to be challenged and have fun, you can't be afraid to take risks, and you can't worry too much about money.
11 years ago I was in a similar position. I was a senior in CS, and really bored with the curriculum, though I was a good programmer. In last semester I took a software engineering class where we spent all our time writing specs, and none writing code, and a theory of programming languages class, with no programming. What a drag.
What did I do? I found a job outside of school that was fun.
First, I spent a few months with a startup, ultimately with nothing to show for it but 1000 worthless shares of stock and a helluva fun time. Then I got a job as one of two programmers in a university research lab, working on what was ultimately a huge project (> 150,000 lines of C). It didn't pay nearly as well as an industry job would have, but it was great! I had nearly complete freedom in design and implementation, and though I was working in a very small niche market, I got to build a program that was used by researchers all over the world. After several years I went to work for a small company, working on a project in the same niche market (psychological experiment software).
Eventually I got bored with that, and 3 years ago I came to grad school to get my PhD in C.S. I'm making a quarter of what I would have been making if I'd stayed in my job, but I'm working on cutting edge new research in AI, and it's a blast!
Grad school in CS, at least in a doctoral program, has much less programming than you might think, and the programming you do is your research. You actually spend a lot of time reading the latest research (from recent conference and journal articles, not textbooks), writing up your own research, and preparing and giving talks on your or others' research.
Be warned, though: If you are thinking about a PhD in C.S., make sure that you do it for the experience of doing it not for the degree. It is too long a road to always be looking that far ahead.
Whatever you choose make sure you're having a blast!
If they were truly random, we'd have a different PI each time we calculate it. Remember that the so-called "random number generators" are in fact pseudorandom. Like the digits of PI, they are deterministic, and with the same starting point in the sequence, you always get the same set of numbers.
Pseudorandom numbers are often used in place of true random numbers, because usually what is needed is a set of numbers with certain properties common to random numnbers, e.g. uniform distribution. Note that for cryptography, pseudorandomness is often not sufficient, and truly random numbers are needed. These are usually generated by sensing the physical world in some way, where, we assume that the combination of chaotic processes and quantum effects makes the incoming values truly unpredictable.
A NEW STUDY by the British Economic and Social Research Council (ESRC), has determined that intelligent children with good hand-eye coordination and strong powers of concentration are more likely to play video games. "It seems, surprisingly, that these children have a greater level of success playing video games, and thus enjoy them more than the average child," said Dr. Inieda Kluu, the principal investigator in the study. Researchers in the study were also suprised to note that more intelligent children were more likely to attend college and eventually have a high-paying job.
"This was totally news to us," said Dr. Klu, "we had always assumed that all children were equally likely to attend college and get good paying jobs. Isn't that what we've been taught since childhood, after all?"
Among other controversial results of the study, they found that children who played video games regularly were more likely to live in a house with a computer than those who did not, were more likely to have parents with above-average incomes, and were more likely to have parents of above-average intelligence.
"This temporal-reverse causation from children to their parents is the most astounding aspect of the study," Dr. Klu said. "Who would ever have imagined that children's video game playing could cause increased affluence and intelligence in their parents?"
Investigators from the study are considering consulting with renowned Cambridge physicist Dr. Stephen Hawking on the possibility of a temporal worm-hole created by video game playing. "We think it's something like what happened at the end of the movie A.I.," Dr. Klu explained, "where they were able to clone that woman, and extract her memories from the fabric of space-time. We think a similar mechanism is at work with video games, only backward: increasing the intelligence of the parents through a reverse space-time-DNA wormhole."
But before the university can sell the seeds to farmers, it must get clearance from holders of as many as 34 patents, said Dr. Ana Sittenfeld, an associate professor there.
One possible solution to this is the creation of a rights clearinghouse much like BMI or ASCAP for music publishing rights, combined with a compulsory licensing scheme with set royalty rates.
This is basically the solution that music publishers and music licensees came up with years ago (or that the government came up with for them) to solve a similar problem with music licenses. Similar schemes are being proposed for AIDS drugs and other medicines and for online music.
Of course this doesn't solve the myriad other problems associated with GM foods and restrictive patents, but it's a start.
Why they then call it a System 324 and not a 336 is a
conundrum.
...
The way RLX has managed to tuck 24 servers into a 3U enclosure is to stick them in
vertically.
Not that much of a conundrum. A 3U enclosure containing 24 servers.
For an example of code as expression, it's worth looking at ACL2. From the page: ACL2 is both a programming language in which you can model computer systems and a tool to help you prove properties of those models.
I took a class on ACL2 last year, and it's actually much more than the quote above indicates. It is a fully executable subset of Common LISP, which has been augmented to be a sound logic in which general theorems can be stated and proven. In other words, you can write a program in it, and then prove theorems about that program. It can also be used to model hardware systems, and then prove theorems about them.
I think ACL2 is a perfect example of the ambiguity of the distinction between code and speech. Mathematical theorems and their proofs would generally be considered to be expression, regardless of what formal language is used to write the proof. But if you choose ACL2 as your proof language, then a theorem could itself be, or contain, executable LISP programs. How is one to distinguish, then, between a "circumvention device" and a proof or theorem about it, if they're both expressed in the same language?
Take a look at Neural network objects by Johannes Steffens. It's a C++ class library that supports both supervised and unsupervised learning networks including ordinary backpropagation (or MLP, multi-layer perceptron) nets, as well as Kohonen feature maps (KFM, aka SOM) and learning vector quantization.
It also supports several growing network architectures developed by Bernd Fritzke et al. There's also a Java demo of growing networks, with code available. It's fun to change the input probability distribution and watch the nets adapt on the fly!
The patent seems to be an implementation of an idea that I first saw in this Jan 1997 Hotwired article. Note that the article predates the filing of the patent. Since the article doesn't describe a specific implementation, I have no idea how it relates to their filing WRT "prior art". It certainly seems like they didn't think of the idea.
I dunno, it seems like their implementation is pretty obvious to anybody who's read the article. Any patent experts care to educate on how this works?
Interestingly, I read the hotwired article when it came out, and now, nearly 3 years later, I've been thinking of doing some x-mas break hacking to throw together a generic implementation of this very thing! It would have been GPL'ed. Anyone know how different the implementation would have to be to avoid hassles?
This is a perfect example of the "chilling effect" of software patents. As a grad student, I definitely don't have the resources to defend myself against a legal attack from Intel!
Slashdot Mirror
"In the future the rate of correctly predicted new friends could be even higher." ...or it could be lower.
Hah ha! That's good. I especially like "from a motorboat.."
. html
Try this for the real lyrics:
http://www.skynyrd.com/lyrics/91-97/box/disc2/d2c
A million dollars? With that they might be able to buy a tiny house near Google. But then they'd be broke and have to keep working. And somehow, in the mindset of Silicon Valley, that seems normal and rational.
Of course, they could always retire and move someplace sane.
Once the Brazilians figure out how lame and useless these social networking things are, their numbers will drop.
Here is the reference:
Provost, J (1999). Naive-Bayes vs. Rule-Learning in Classification of Email. The University of Texas at Austin, Artificial Intelligence Lab. Technical Report AI-TR-99-284
I should mention that I don't think I'm the first to use Naive-Bayes on email. I think some folks from Microsoft did it in 1998, and there may be others too.
This is already happening in Artificial Intelligence. The Journal of AI Research (JAIR), and The Journal of Machine Learning Research (JMLR) are peer-reviewed journals published on the web for free.
I'm not sure what the $20 million is for, since (at least in AI) peer-review is done for free anyway, as a service to the community. The big journals charge money while getting editing, review, and often even typsetting for free from their editorial boards or authors.
Since peer-review is the main service provided by the big journals, it was only a matter of time before the reviewers organized themselves. The tenure issue is a bit of a problem, since untenured faculty will want to publish in the best established journals. However, that should work itself out over time, as tenured researchers choose to publish in the new free journals. Eventually the new journals will be well enough established for young researchers to feel comfortable publishing in them.
There absolutely *is* and archive. All broadcast material is archived on tape Who cares? It's inaccessible to the viewers. Nobody links to a specific airing of a TV story and expects it to be the same the next time they see it.
I can understand why CNN thinks this is no big deal. CNN was (and is) primarily a TV news station. On TV news, there is no archive or changelog for writethroughs: the copy gets rewritten, and the reporter or anchor reads it on the air. The only way you notice the changes is if you happened to see a previous version of the story earlier in the day.
CNN obviously sees the web as a translation of their TV news business, rather than as a translation of a print-news wire service business, so to them it seems fine! To them the web is a transient medium, like TV, not a fixed medium like print.
Of course, at first glance this seems fine, until linking of stories factors into the equation.
Of course, there are technological solutions to this, but getting CNN to adopt them could be a challenge, because it means converting them from a TV mindset to a print mindset.
When you have someone working for you, you don't stand over his shoulder, verbally telling him how to scroll and what to type. You say, "write me a report on X." Then when the report comes, you read it while the employee is working on something else. The report (as well as email, etc) is a visual interface to the employee, which allows you to use your working memory for thinking, rather than talking.
The comment in the article about speech "not carrying the load" of vision still assumes that the human is doing most of the work, and the computer is just a glorified pencil and paper. The idea behind HAL was that humans told him what to do, not how to do it. For that, you want speech.
There's been lots of work on auto-classifying email. I did my semester project in Machine Learning on this in 1999. It's a fairly simple study, but it seems like a Naive Bayesian classifier using word counts as features does a pretty decent job of classifying email, and does really well on spam.
The paper is here here.
J.
The complexity for minimum component costs has in- creased at a rate of roughly a factor of two per year (see graph on next page). (p.2, second column).
The graph on p.3 clearly shows an annual doubling of components (transistors). The current 18-24 month rate is an updated version of the law, to take into account the fact that the rate has slowed since the 60's. Of course, this means that this "Law" isn't, in fact, much of a law at all.
Not having to buy licenses for much or all of the software on their un*x workstations saves departments huge amounts of money. Moreover, they can build workstations from commodity components. This allows them to provide more machines for students, and simultaneously exposes huge numbers of CS undergrads and grad students to free software.
Also, the dot-com bubble bursting caused CS graduate school enrollments to swell enormously. Grad schools have traditionally been places where much free software is born, as student researchers put their work out there for everyone to see.
The problem is that only a few schools really do research in user interfaces and similar areas that will advance free software in the mainstream. But in a lot of less visible areas: like the core-OS, distributed computing, networking, scientific computing, high-performance graphics, AI and robotics, free software will continue to progress and improve through universities. In the process the universities will continue to graduate students who are used to working with free software, and who will wonder why they should buy licenses for software when so much is available for free.
11 years ago I was in a similar position. I was a senior in CS, and really bored with the curriculum, though I was a good programmer. In last semester I took a software engineering class where we spent all our time writing specs, and none writing code, and a theory of programming languages class, with no programming. What a drag.
What did I do? I found a job outside of school that was fun.
First, I spent a few months with a startup, ultimately with nothing to show for it but 1000 worthless shares of stock and a helluva fun time. Then I got a job as one of two programmers in a university research lab, working on what was ultimately a huge project (> 150,000 lines of C). It didn't pay nearly as well as an industry job would have, but it was great! I had nearly complete freedom in design and implementation, and though I was working in a very small niche market, I got to build a program that was used by researchers all over the world. After several years I went to work for a small company, working on a project in the same niche market (psychological experiment software).
Eventually I got bored with that, and 3 years ago I came to grad school to get my PhD in C.S. I'm making a quarter of what I would have been making if I'd stayed in my job, but I'm working on cutting edge new research in AI, and it's a blast!
Grad school in CS, at least in a doctoral program, has much less programming than you might think, and the programming you do is your research. You actually spend a lot of time reading the latest research (from recent conference and journal articles, not textbooks), writing up your own research, and preparing and giving talks on your or others' research.
Be warned, though: If you are thinking about a PhD in C.S., make sure that you do it for the experience of doing it not for the degree. It is too long a road to always be looking that far ahead.
Whatever you choose make sure you're having a blast!
Pseudorandom numbers are often used in place of true random numbers, because usually what is needed is a set of numbers with certain properties common to random numnbers, e.g. uniform distribution. Note that for cryptography, pseudorandomness is often not sufficient, and truly random numbers are needed. These are usually generated by sensing the physical world in some way, where, we assume that the combination of chaotic processes and quantum effects makes the incoming values truly unpredictable.
Among other controversial results of the study, they found that children who played video games regularly were more likely to live in a house with a computer than those who did not, were more likely to have parents with above-average incomes, and were more likely to have parents of above-average intelligence.
"This temporal-reverse causation from children to their parents is the most astounding aspect of the study," Dr. Klu said. "Who would ever have imagined that children's video game playing could cause increased affluence and intelligence in their parents?"
Investigators from the study are considering consulting with renowned Cambridge physicist Dr. Stephen Hawking on the possibility of a temporal worm-hole created by video game playing. "We think it's something like what happened at the end of the movie A.I.," Dr. Klu explained, "where they were able to clone that woman, and extract her memories from the fabric of space-time. We think a similar mechanism is at work with video games, only backward: increasing the intelligence of the parents through a reverse space-time-DNA wormhole."
Dr. Hawking was unavailable for comment.
One possible solution to this is the creation of a rights clearinghouse much like BMI or ASCAP for music publishing rights, combined with a compulsory licensing scheme with set royalty rates.
This is basically the solution that music publishers and music licensees came up with years ago (or that the government came up with for them) to solve a similar problem with music licenses. Similar schemes are being proposed for AIDS drugs and other medicines and for online music.
Of course this doesn't solve the myriad other problems associated with GM foods and restrictive patents, but it's a start.
...
The way RLX has managed to tuck 24 servers into a 3U enclosure is to stick them in vertically.
Not that much of a conundrum. A 3U enclosure containing 24 servers.
J.
I took a class on ACL2 last year, and it's actually much more than the quote above indicates. It is a fully executable subset of Common LISP, which has been augmented to be a sound logic in which general theorems can be stated and proven. In other words, you can write a program in it, and then prove theorems about that program. It can also be used to model hardware systems, and then prove theorems about them.
I think ACL2 is a perfect example of the ambiguity of the distinction between code and speech. Mathematical theorems and their proofs would generally be considered to be expression, regardless of what formal language is used to write the proof. But if you choose ACL2 as your proof language, then a theorem could itself be, or contain, executable LISP programs. How is one to distinguish, then, between a "circumvention device" and a proof or theorem about it, if they're both expressed in the same language?
It also supports several growing network architectures developed by Bernd Fritzke et al. There's also a Java demo of growing networks, with code available. It's fun to change the input probability distribution and watch the nets adapt on the fly!
J.
I dunno, it seems like their implementation is pretty obvious to anybody who's read the article. Any patent experts care to educate on how this works?
Interestingly, I read the hotwired article when it came out, and now, nearly 3 years later, I've been thinking of doing some x-mas break hacking to throw together a generic implementation of this very thing! It would have been GPL'ed. Anyone know how different the implementation would have to be to avoid hassles?
This is a perfect example of the "chilling effect" of software patents. As a grad student, I definitely don't have the resources to defend myself against a legal attack from Intel!
J.
J.