I still think you're not getting the nature of Block's claim correct...
Now, if you think the huge(infinite) lookup table machine thinks, then you're denying a premise of Block's, but you're not offering a challenge to the validity [ in the technical sense of "if the premises are true, then the conclusion must also be true" ] of his reasoning.
I'm not denying that if the premises are true, then the conclusion is true. I'm denying that the premises are true.
If I claimed that normal bricks can't fly, and you said "Imagine a brick that could fly. That's a brick, and it flies, so your claim is clearly false." I wouldn't deny that your statement is logically correct, but I would deny that such a brick exists.
Block say "Imagine a Blockhead that passes the Turing test." and that's what I object to.
In practice, you can't build one. That's easy to prove.
I claim that are infinitely many possible theoretical conversations so a theoretical Blockhead would need to be infinite to contain all possible conversations. Hence my statement that you can't even build one in theory. I'm a little unclear on why the theoretical question space must be limited to what is "reasonable" while the theoretical answer space is not so limited, but ok, you don't accept that the number of theoretical conversations is infinite. Fine.
I still claim that a Blockhead that contains a set of answers large enough to pass the Turing test would no longer qualify as "clearly not thinking". This is because a large enough state tree is indistinguishable from a general purpose computer. In other words, a Blockhead big enough to need a terabyte of ram just to store it's position in the tree isn't really a Blockhead anymore.
The point he was trying to make was that the building blocks used to build up the language must be as simple as possible.
I agree that that's his point, but I don't agree that it's true.
RISC was a great practical example of why eliminating useless instructions was a good idea. We could eliminate numbers, but if every program we write uses numbers, then it's better to have them as intrinsics. Even if only 90% of the programs we write use them, it's still better to have to them built in then to recreate them (or include the numbers library) every time we use them. Any concept and data type should be examined to determine how often it's used, and if including it costs more than we save by not including it on average. The "cost" here is the cost in thinking about them, not code space or compile time, and that should include learning time as well time spent thinking about how to use them in a particular program. Since numbers are taught in grade school, the thinking time for them is very small.
It doesn't matter, for that kind of argument, whether the described system is one that could actually be built...
Ah, but it does matter. Approach it from the other side; The contrapositive of the Turing test says that a machine that doesn't think, can't pass. Block says "imagine a device that doesn't think, but passes the Turing test anyway." if such a device could exist, it would certainly invalidate the Turing test. But if the device can't exist, it proves nothing.
My 1000 coin flips example was a simple one, designed to show that Blockhead can't exist in the real world. I can extend the example to force the device to be arbitrarily large, but if we're talking theory, why stop at an arbitrary value? 1.) Roll a die and write down that number. 2.) Unless you rolled a 6, go to step 1. 3.) ask Blockhead what number is 1 larger than that. To answer, Blockhead needs to be infinite.
Now I claim that you can't even theoretically build an infinitely large device, but if you want to continue then I'm going to point out that an infinitely large Blockhead would be Turing complete. This puts it outside the realm of conventional state trees, so it's no longer an easy matter to dismiss it's ability to think. I.e. just because conventional state trees don't think, it doesn't follow that unconventional ones don't either.
in order to accurately estimate the true damages, a little flashback to microeconomics class is in order. ..
The price they can charge for a song is directly related to the supply. Dilute the supply, and the optimum selling price goes down. It's not likely that you'll hear the RIAA complain that piriacy forces them to lower their prices, but it's a real consequence of the economics. If you want to do a real analysis of the record companies loss, this lower price should be part of calculation.
To accept the blockhead as a counter-example you have to accept both that you can build one, and that if you built one, it doesn't actually think.
Consider this: Flip a coin 1000 times, then ask Blockhead how many Heads there were. To answer this question, Blockhead needs a tree of at least 2^1000 states. So if Blockhead could exist, it would be larger than the universe.
If you release Blockhead from constraints of physical reality, then it's relatively simple to give it memory. (they "pointer" to the location in the tree is stateful) and from there, we have a general purpose computing device. If theoretical Blockhead is a general purpose computing device, then you to answer the question, you must already know the answer. Block claims that Blockhead obviously doesn't think, but Saying "computers can't think because obviously computers can't think" isn't very convincing.
The Matrin County School Board has a new way of post a student's grades online for a parent to check.
This sounds like a great idea to me. When I was in school, most of my fellow students wasted huge amounts of time worrying about concealing their grades and attendance records from their parents. I on the other hand learned very quickly to completely disregard my parents opinions on the matter. By making it harder for students to snow their parents, they will be forced to learn a proper measure of disrespect for authority.
Further discussion is bound to appear on sci.crypt.random-numbers before too long. First, the best paper I've been able to find is this one; http://www.via.com.tw/en/viac3/via_c3_padlock_ev al uation.pdf
This HRNG is based on sampling a high speed oscillator with a low speed oscillators XORed together. The output is run through a "von Neumann Corrector Register" before finally being output. (The corrector can be bipassed.) Both oscillators and Von Neumann's method of correcting for bias have a bad history in HRNG, so this particular HRNG doesn't give me that warm glowy feeling of having been designed by someone who knew what they were doing.
The output of this generator is biased, even in the "corrected" mode. This isn't damning by itself, but it shows that as usual, the raw bits are not independant, so the Von Neumann corrector doesn't, and it's IMO overall a waste. Not a big waste, it has some good properties, but there are much better things that could have been built out of the realestate.
The interdependance of the bits is typical of an oscillator design, and we can expect all the usual failures. In particular, if they ever do a shrink of this chip, they need to redesign the oscillator portions.
Although I'm not particular impressed with this design, and the output needs to be processed before being used, it's still way better than nothing. With a raw output of over 3 megabytes a second, what it lacks in unbiasedness can be more than made up with oversampling. In the processed mode, it's only 1/8 the speed so it wouldn't be an acceptable replacement for statistical programs even if the output were acceptably unbiased. Despite the claims of the paper of.85-.99 bits of entropy per bit, Clearly the amount of entropy in the raw mode is far less than.5, probably closer to.3. (If it were as high as claimed, then the corrector wouldn't be as slow as it is.) The bad entropy estimate is really a failure of the paper though, not the HRNG. Many HRNG designs produce less than.1 e-bits/per bit so.3 isn't bad.
Look at last month's Netcraft survey and scroll down to the "Totals for Active Servers Across All Domains" graph. Note that IIS increased until about 2002-04, when it leveled off. Compare that to Apache which continues it's rocky climb into 2003. The numbers suggest that neither one is actually displacing the other yet. I.e. few, if any, are replacing IIS with Apache.
Yes, Apache has always had the lead, and it's lead continues to widen, but not at the expense of IIS.
Cornell University Professor David Pimental clearly shows how creating Ethanol (gas for tractors, refining, delivery, nitrogen fertilizer, etc...) uses more power (and pollutes more, btw) than you can yield from it:
That report assumes that the ethanol is refined to 99.8% purity, is produced using the same level of quality we insist on for food, and that fertilization of the fields is necessary.
Fuel cell methanol doesn't need purity, and fertilizier is only added when it's economically viable. Remove subsidies and corn would still be grown, we just wouldn't grow as much per acre.
Don't ignore the reality of food production either. We intentionally overproduce food right now because a food shortage is so disasterous. It's better to convert the surplus to ethanol and store it in a long term reserve than to let the food rot on the ground.
Overall, water+sunlight+carbon -> corn -> methanol -> water+carbon produces an excess of oxygen and carbon.
The idea is well-intentioned, but ludicrous. It would be impossible to implement. This will become another urban legend in time, just as that old chectnut about the US Postal Service charging for modem use or for sending an email.
Bad to impliment perhaps, but not impossible. I already implement something very similar with spamwolf. I could probably hack out an implementation in a week if I wanted to.
It's easy enough for the recipient to maintain a white list if it doesn't have to be complete, and your friends probably wouldn't mind spending a few cents once to email you. The real problem as I see it is big impersonal companies that send important email. Network Solutions isn't likely to spend even a few cents to notify you that your password is being changed, or provide a valid return address that's read by a human, but that's one message you really want to receive.
Unfortunately, this leads to the "Post-It Note Effect", wherein many users leave their passwords on little Post-It notes on their monitors.
As much as I agree that it would be better if they picked passwords that were strong and actually remembered them, If the only choices are weak passwords or strong passwords on postit notes, I'd prefer an office with postit notes.
Postit notes leave you vulnerable to people with physical access to your machine, where as weak passwords leave you vulnerable to everone on the internet. (But you can at least insist that people stick the postit notes under their keyboards or folded over so the passwords aren't casually visible.)
Congratulations! You are the 1 BILLIONTH person to misquote Shakespeare! When looked at in context, this quote actually PROVES THE NECESSITY of lawyers in civilization. The characters were trying to create anarchy - killing the lawyers would remove the rule of law and cause just that.
Congratulations! You have fallen victim to one of the classic miss-miss quotes.
When read in context, The characters were describing their vision of utopia
LEDs are indeed the light of the future. I.e. they aren't the best choice for light today, however they have been improving at close to what Moore's law predicts, so perhaps in 10 years, they will be the ultimate light source. Today however, they are still too expensive (90 dollars vs 90 cents for a standard bulb) and only a little more efficent.
Last year the light/power breakdown was something like this: Lumens/Watt Light Source 100-190 low pressure Sodium (HID) 50-150 High pressure Sodium (HID) 60-140 Metal Halides (HID) 20-60 mercury vapor (HID) 85-95 32 watt T8 fluorescent 60-65 standard F40T12 cool white fluorescent 48-60 compact fluorescents 45-55 Super bright Red/Orange LED 35-45 Super bright Green LED 20 T3 tubular halogen 15-25 bright white LED 5-25 Halogen 17 standard 100 watt incandescent 6 incandescent night light bulb (7w) <6w incandescent flashlight bulbs
The best choice for the home today is almost certainly the cool white tube fluorescent, with an all electronic balast. They don't "whine" or flicker like the older fluorescent lights, the tubes cost about $2.00 US, they are near the top in efficiency, and the last for years. The down side is that they require special fixtures, and they flicker more than incandencances, which is a problem for some people.
Compact flourescents are the next best choice, but they don't have "instant on" which can be annoying for some rooms. In my kitchen, I have an old 3 socket fixture. I use two compact flourecents and a long life incandescent and it works well.
Yes, this is the third time I've made essentially the same post, what's your point?
The functions that ICANN should be performing are minimal, and I'd like to see their replacement deligate as possible. For example, instead of creating a body that resolves international disputes in the name space, I'd much rather that they divied up the domains amongst.cc codes, assigned jurisdiction over disputes of those names to the respective countries, and eliminated all non-cc TLDs. (They can be moved under an existing.cc)
The only thing NewICCAN would need to do is keep the administrative records for a few hundred cc TLDs. If a government wanted to, it could assign control over it's domain space to the current ICANN, but I doubt that any government except the US actually would.
This may not be the crowd that wants to hear this, but some radical changes need to be made in the email protocol to minimize the amount of spam that users deal with these days. Bottom line is that the goal should be for email communications to be as trustworthy as phone calls - sure, there are some telemarketers and crank callers out there, but if the noise level from your phone was as high as in your email, there would be marches on Washington to demand a solution.
I would think the most likely candidate would be to build-in verification of the sender, and bring about the end of anonymous email. That's sure to raise the hackles of many here, but so far, nothing's working.
Before changing a protocol that's worked for over 30 years, I'd like at least some evidence that what you propose would actually address the problem.
It's easy to claim anonymous email is responsible for spam, but I get plenty of spam from companies that don't attempt to hide. (Network solutions for example.)
And I don't see any reason the protocols need to change to implement what you're talking about - You could reject any email that isn't PGP signed right now if you want to. Make it easy for people to sign their email and it might even be reasonable to reject anything unsigned.
Not quite the same thing, but I did hear about a chimney contracter who would put a thin sheet of glass across the chimney blocking it. When he got the final payment he would drop a brick down the chimney.
Slashdot Mirror
So what's the thing that might give me some hope?
I'm not denying that if the premises are true, then the conclusion is true.
I'm denying that the premises are true.
If I claimed that normal bricks can't fly,
and you said "Imagine a brick that could fly.
That's a brick, and it flies, so your claim is clearly false."
I wouldn't deny that your statement is logically correct,
but I would deny that such a brick exists.
Block say "Imagine a Blockhead that passes the Turing test."
and that's what I object to.
In practice, you can't build one. That's easy to prove.
I claim that are infinitely many possible theoretical conversations so a theoretical Blockhead would need to be infinite to contain all possible conversations.
Hence my statement that you can't even build one in theory.
I'm a little unclear on why the theoretical question space must be limited to what is "reasonable" while the theoretical answer space is not so limited, but ok, you don't accept that the number of theoretical conversations is infinite. Fine.
I still claim that a Blockhead that contains a set of answers large enough to pass the Turing test would no longer qualify as "clearly not thinking".
This is because a large enough state tree is indistinguishable from a general purpose computer.
In other words, a Blockhead big enough to need a terabyte of ram just to store it's position in the tree isn't really a Blockhead anymore.
-- this is not a
I agree that that's his point, but I don't agree that it's true.
RISC was a great practical example of why eliminating useless instructions was a good idea.
We could eliminate numbers, but if every program we write uses numbers,
then it's better to have them as intrinsics.
Even if only 90% of the programs we write use them,
it's still better to have to them built in then to recreate them (or include the numbers library) every time we use them.
Any concept and data type should be examined to determine how often it's used,
and if including it costs more than we save by not including it on average.
The "cost" here is the cost in thinking about them, not code space or compile time,
and that should include learning time as well time spent thinking about how to use them in a particular program.
Since numbers are taught in grade school, the thinking time for them is very small.
-- this is not a
The results of a random phone survey show that 100% of New Yorkers have a listed number...
Ah, but it does matter. Approach it from the other side;
The contrapositive of the Turing test says that a machine that doesn't think, can't pass.
Block says "imagine a device that doesn't think, but passes the Turing test anyway."
if such a device could exist, it would certainly invalidate the Turing test.
But if the device can't exist, it proves nothing.
My 1000 coin flips example was a simple one, designed to show that Blockhead can't exist in the real world.
I can extend the example to force the device to be arbitrarily large, but if we're talking theory,
why stop at an arbitrary value?
1.) Roll a die and write down that number.
2.) Unless you rolled a 6, go to step 1.
3.) ask Blockhead what number is 1 larger than that.
To answer, Blockhead needs to be infinite.
Now I claim that you can't even theoretically build an infinitely large device,
but if you want to continue then I'm going to point out that an infinitely large Blockhead would be Turing complete.
This puts it outside the realm of conventional state trees,
so it's no longer an easy matter to dismiss it's ability to think.
I.e. just because conventional state trees don't think,
it doesn't follow that unconventional ones don't either.
-- this is not a
The price they can charge for a song is directly related to the supply.
Dilute the supply, and the optimum selling price goes down.
It's not likely that you'll hear the RIAA complain that piriacy forces them to lower their prices,
but it's a real consequence of the economics.
If you want to do a real analysis of the record companies loss,
this lower price should be part of calculation.
-- this is not a
If you assume that they did understand, then the post is funny.
Perhaps the problem is not in their ability to understand.
-- this is not a
To accept the blockhead as a counter-example you have to accept both that you can build one,
.sig
and that if you built one, it doesn't actually think.
Consider this:
Flip a coin 1000 times, then ask Blockhead how many Heads there were.
To answer this question, Blockhead needs a tree of at least 2^1000 states.
So if Blockhead could exist, it would be larger than the universe.
If you release Blockhead from constraints of physical reality,
then it's relatively simple to give it memory.
(they "pointer" to the location in the tree is stateful)
and from there, we have a general purpose computing device.
If theoretical Blockhead is a general purpose computing device,
then you to answer the question, you must already know the answer.
Block claims that Blockhead obviously doesn't think,
but Saying "computers can't think because obviously computers can't think" isn't very convincing.
-- this is not a
This sounds like a great idea to me. When I was in school, most of my fellow students wasted
huge amounts of time worrying about concealing their grades and attendance records from their parents.
I on the other hand learned very quickly to completely disregard my parents opinions on the matter.
By making it harder for students to snow their parents,
they will be forced to learn a proper measure of disrespect for authority.
-- this is not a
Further discussion is bound to appear on sci.crypt.random-numbers before too long.v al uation.pdf
.85-.99 bits of entropy per bit, .5, probably closer to .3. .1 e-bits/per bit so .3 isn't bad.
.sig
First, the best paper I've been able to find is this one;
http://www.via.com.tw/en/viac3/via_c3_padlock_e
This HRNG is based on sampling a high speed oscillator with a low speed oscillators XORed together.
The output is run through a "von Neumann
Corrector Register" before finally being output.
(The corrector can be bipassed.)
Both oscillators and Von Neumann's method of correcting for bias have a bad history in HRNG,
so this particular HRNG doesn't give me that warm glowy feeling of having been designed by someone who knew what they were doing.
The output of this generator is biased, even in the "corrected" mode.
This isn't damning by itself, but it shows that as usual,
the raw bits are not independant, so the Von Neumann corrector doesn't, and it's IMO overall a waste.
Not a big waste, it has some good properties,
but there are much better things that could have been built out of the realestate.
The interdependance of the bits is typical of an oscillator design,
and we can expect all the usual failures.
In particular, if they ever do a shrink of this chip, they need to redesign the oscillator portions.
Although I'm not particular impressed with this design, and the output needs to be processed before being used,
it's still way better than nothing.
With a raw output of over 3 megabytes a second,
what it lacks in unbiasedness can be more than made up with oversampling.
In the processed mode, it's only 1/8 the speed
so it wouldn't be an acceptable replacement for statistical programs even if the output were acceptably unbiased.
Despite the claims of the paper of
Clearly the amount of entropy in the raw mode is far less than
(If it were as high as claimed, then the corrector wouldn't be as slow as it is.)
The bad entropy estimate is really a failure of the paper though, not the HRNG.
Many HRNG designs produce less than
-- this is not a
Look at last month's Netcraft survey and scroll down to the
.sig
"Totals for Active Servers Across All Domains" graph.
Note that IIS increased until about 2002-04, when it leveled off.
Compare that to Apache which continues it's rocky climb into 2003.
The numbers suggest that neither one is actually displacing the other yet.
I.e. few, if any, are replacing IIS with Apache.
Yes, Apache has always had the lead, and it's lead continues to widen,
but not at the expense of IIS.
-- this is not a
I'm hardly a neutral party,
but I wouldn't trust an anti-spam company that ever had such a spam friendly TOS,
even if the new TOS is "the truth".
-- this is not a
That report assumes that the ethanol is refined to 99.8% purity, is produced using the same level of quality we insist on for food, and that fertilization of the fields is necessary.
Fuel cell methanol doesn't need purity, and fertilizier is only added when it's economically viable.
Remove subsidies and corn would still be grown, we just wouldn't grow as much per acre.
Don't ignore the reality of food production either.
We intentionally overproduce food right now because a food shortage is so disasterous.
It's better to convert the surplus to ethanol and store it in a long term reserve than to let the food rot on the ground.
Overall, water+sunlight+carbon -> corn -> methanol -> water+carbon
produces an excess of oxygen and carbon.
-- this is not a
Bad to impliment perhaps, but not impossible.
I already implement something very similar with spamwolf.
I could probably hack out an implementation in a week if I wanted to.
It's easy enough for the recipient to maintain a white list if it doesn't have to be complete,
and your friends probably wouldn't mind spending a few cents once to email you.
The real problem as I see it is big impersonal companies that send important email.
Network Solutions isn't likely to spend even a few cents to notify you that your password is being changed,
or provide a valid return address that's read by a human, but that's one message you really want to receive.
-- this is not a
I personally like ICF technology.
.sig
(Try googling for Insulating Concrete Forms)
The cost is initially higher than traditional wood,
but it's quite competitive if you take the long view.
-- this is not a
As much as I agree that it would be better if they picked passwords that were strong and actually remembered them,
If the only choices are weak passwords or strong passwords on postit notes,
I'd prefer an office with postit notes.
Postit notes leave you vulnerable to people with physical access to your machine,
where as weak passwords leave you vulnerable to everone on the internet.
(But you can at least insist that people stick the postit notes under their keyboards or folded over so the passwords aren't casually visible.)
-- this is not a
Congratulations!
You have fallen victim to one of the classic miss-miss quotes.
When read in context,
The characters were describing their vision of utopia
-- this is not a
LEDs are indeed the light of the future.
I.e. they aren't the best choice for light today,
however they have been improving at close to what Moore's law predicts,
so perhaps in 10 years, they will be the ultimate light source.
Today however, they are still too expensive (90 dollars vs 90 cents for a standard bulb) and only a little more efficent.
Last year the light/power breakdown was something like this:
Lumens/Watt Light Source
100-190 low pressure Sodium (HID)
50-150 High pressure Sodium (HID)
60-140 Metal Halides (HID)
20-60 mercury vapor (HID)
85-95 32 watt T8 fluorescent
60-65 standard F40T12 cool white fluorescent
48-60 compact fluorescents
45-55 Super bright Red/Orange LED
35-45 Super bright Green LED
20 T3 tubular halogen
15-25 bright white LED
5-25 Halogen
17 standard 100 watt incandescent
6 incandescent night light bulb (7w)
<6w incandescent flashlight bulbs
The best choice for the home today is almost certainly the cool white tube fluorescent, with an all electronic balast.
They don't "whine" or flicker like the older fluorescent lights,
the tubes cost about $2.00 US, they are near the top in efficiency, and the last for years.
The down side is that they require special fixtures, and they flicker more than incandencances, which is a problem for some people.
Compact flourescents are the next best choice,
but they don't have "instant on" which can be annoying for some rooms.
In my kitchen, I have an old 3 socket fixture.
I use two compact flourecents and a long life incandescent and it works well.
Yes, this is the third time I've made essentially the same post, what's your point?
Windows 95 will not allow more than 32767 files in a single directory.
.sig
It's probable that later versions also have that limitation.
-- this is not a
There might be a system running a hundred processes, but a typical windows system runs more like 8.
And very few of those 8 actually share the dlls they load.
Most running programs are less than 50% library code.
Trivial apps aren't, but how many trival apps do you run at the same time?
-- this is not a
Maybe when you're dead, the prospect of being buried doesn't look so bad.
-- this is not a
Why do we need any generic international domains at all?
The functions that ICANN should be performing are minimal,
and I'd like to see their replacement deligate as possible.
For example, instead of creating a body that resolves international disputes in the name space,
I'd much rather that they divied up the domains amongst
assigned jurisdiction over disputes of those names to the respective countries,
and eliminated all non-cc TLDs. (They can be moved under an existing
The only thing NewICCAN would need to do is keep the administrative records for a few hundred cc TLDs.
If a government wanted to, it could assign control over it's domain space to the current ICANN,
but I doubt that any government except the US actually would.
-- this is not a
Before changing a protocol that's worked for over 30 years,
I'd like at least some evidence that what you propose would actually address the problem.
It's easy to claim anonymous email is responsible for spam,
but I get plenty of spam from companies that don't attempt to hide.
(Network solutions for example.)
And I don't see any reason the protocols need to change to implement what you're talking about -
You could reject any email that isn't PGP signed right now if you want to.
Make it easy for people to sign their email and it might even be reasonable to reject anything unsigned.
-- this is not a
Not quite the same thing, but I did hear about a chimney contracter who would put a thin sheet of glass across the chimney blocking it.
When he got the final payment he would drop a brick down the chimney.
-- this is not a