My point is: given the same materials why can't they make chairs with the right shapes? I'm not even talking about durability - just after a few minutes of sitting (or even a few seconds) you know it's a crap chair.
Believe me, I've tried some chairs and after a certain point, the grade and thickness of the plastic or steel used only affects how long it lasts (and how heavy it is), not how comfortable it is.
I've seen chairs that are about USD60-80, and if they would just change the shapes (same material) they'll be comfy for most people.
It should be possible to make cheap chairs that are comfy (not talking about durable and lasting for 10 years etc).
You even have people here who mention they prefer sitting on a plain wooden chair. So a lot of people must be doing something wrong - sure people's bodies are shaped different, but when you have adjustable plastic+steel chairs and 80% of the population don't like them, someone must be doing something wrong in a big way.
To the people saying it should be worth spending thousands of bucks to save your back. Why should you have to spend thousands to do so?
Why are so many chairs crappy? Why are so few people making decent chairs for decent prices? Just sit in some chairs for a few minutes and it's obvious they are crap.
I thought the US folk have lawsuit happy people?
Strange.
Given the amounts some of you are willing to spend why don't you buy a TFT screen, suspend it from the ceiling, and then work lying down flat on your back on a firm mattress, with your hands to your side each on half of a split keyboard.
Or at those prices you could get one of those special hospital beds for paraplegic/infirm who spend HOURs on them and so the stuff has to be comfy if not they either get bedsores fast or the nurses have to rotate them a LOT more often. Some of these beds have a vibration thingy to help reduce the bedsore thing.
Seriously some of the jokes they use to pad their spams so they get rated as Unsure by my filter are pretty funny. Heck they're funnier than the jokes some relatives forward me;).
Some of the random paragraphs are from public domain books which happen to be interesting too.
If the spammers have to start sending us emails that are 80% interesting and 20% trying to sell Viagra, then it's far better than TV.
Also, there's one part which the spammers will have difficulty hiding. This is the "Call to action" part. There has to be a way for them to be contacted in order to sell you stuff.
If there isn't a way, then it's a dud spam - where the spammer screwed up - there isn't much you can do about that - sometimes I get spammed with totally blank emails. Virus spams aren't counted - because detecting viruses is not as ambiguous a problem.
If there's a way - email, url, then you can identify them. Of course there are sites which do redirections off other sites (tinyurl, yahoo etc), but even so the url has to take you somewhere, so an antispam system can get its clues from these urls.
It'll be funny tho, if someone released a free personal antispam system that on receipt of a suspicous nonwhitelisted email, actually pretended to be IE and visited the urls to check the content there in order to see whether an email is spam or not. Talk about DDoS.;).
Actually the antispam filters I've seen appear to do ok for case 1). In fact some silly people think that the antispam software is not working coz it allows them to intentionally forward spam to someone.
Case 2) is actually magnitudes easier to deal with - there is a LOT less ambiguity whether something contains a virus or not.
Where antispam filters fail is not for the scenarios you cite, but in the subjective scenarios where even humans differ on whether its spam or not.
Whether some emails are spam or not sometimes depends on whether you are in Sales, Accounting, or Support or whatever... Esp when it's people who you gave your business card to who are spamming you.
More correctly - what are the odds of you getting a million dollar mail from a total stranger that looks like spam?
Million dollar mails from partners/resellers that look like spam would go through because of whitelists.
With those odds I'm thinking you might as well buy a lottery ticket and get your millions that way.
That's assuming you use decent antispam software of course. One that doesn't give much weight to IP blacklists. IP/DNS blacklists in my experience have such a high positive rate, that it's ridiculous to use them at all.
What's with all these buffer/heap overflows? Those are definitely a crippling deficiency. Nowadays they can even be worse than speed. They seem too common a problem in C++ programs, not sure why.
Heck, I'm starting to look for an SSH server written in a language that isn't prone to "attacker can execute arbitrary code of the attacker's choice" problems. It's not like I need stupendous performance for SSH.
php is popular and at least a fair number seem to think it's cool (I don't, but I'm not the majority - I think the PHP devs have a bad habit of adding features/functions/methods as standards without thinking about the long term consequences).
If Swing is brilliant but hard for most people to do correctly, it isn't so brilliant is it? That's a lame excuse. It's almost like saying Perl programs are easy to read as long as the programmer takes the trouble to make them easy to read. What's the percentage of good Swing apps? What's the percentage of readable Perl programs?
"Java has considerably fewer surprises and prefers not to add complexity to the language for rarely used features "
I was pretty darn surprised when there was an easy way to find the number of columns returned by an SQL query, but there wasn't an easy way to find the number of rows. Maybe it's only me but I think trying to find the number of rows of an SQL query is more common than trying to find the number of columns.
I also was pretty surprised some years back when the default java classes for doing http clients don't appear to support cookies. EVEN till today it appears to still be true. Check this or this out. Whereas with perl LWP, it just takes a few lines of code to login to a site, rip data out, etc etc.
Basically Java gave me the impression that the people who write the default classes (and other stuff) don't use them in practice, they're just told to write a bunch of methods as part of their job to fit a spec given by The Project Leader/Committee. So many things appear to be only be useful for PHB/BS compliance (look at the recent Java version numbering scheme for an example of Java Culture - J2SE 5..).
Whereas looking at the other languages, the default libraries etc are written by programmers who actually use the stuff on an almost daily basis, so if you need to do something common, it usually is there.
For Java you often have to go write it yourself. While that's also true for languages like LISP, with LISP you don't have to type as much.
And what's with the blooming XML for almost everything? Worse is the _almost_ part. I was trying to configure something using an XML config file, and followed the existing style to try to configure a parameter. After wrestling with it for some time, it turned out there's no way to do it via XML, you have to do it in the command line (till maybe in a future version).
The real great thing about Java is: projects done in Java can be outsourced to many average and cheap programmers (in fact java makes it almost compulsory for large projects to have many programmers). You just need an elite programmer (or two) who programs in English (or whatever human programming language of choice), and then you send it (the spec) to India for "compilation" to Java for the next 3-6 months.
Then the expensive elite programmer can go do something else.
Whereas with many of those "cool" languages, even though you only need one Elite Programmer who can do everything in 1 month, only the expensive elite programmer can continue to handle the project (maintenance, feature additions/changes etc). So what happens if the Elite Programmer gets bored and wants to do something else? You're screwed, that's what.
That, is the big advantage of using Java. Though it takes 3-6 times longer, who cares? How many times can you ask an Elite Lisp programmer to do Yet Another Customized Accounting/Business Application before he resigns? I suppose if you get the Elite programmer to write the modules and then get the grunts to use the modules it may work. But how many grunts know Lisp? How many know perl? How many know Java?
I thought the US only had one party - The Corporates. The Corporates pick who gets to play politician (whether Democrat/Republican) coz you need money for that and the Corporates are the ones with the most money.
It's like being given two or three cards to pick by a magician and thinking that you have a choice.
"Then all you have to do is enforce the robot with the Golden Rule, Do unto others as you would have them do unto you"
0) How would you enforce the rule? 1) What if someone makes a robot that wants to kill 100 people and then wants to be killed while doing so?
Once you have a real AI it won't be easy having hard defined rules which it will _always_ obey. Especially since whether something is correct or incorrect can be very subjective. Do we really need sentient AI for problems which have clear cut correct/incorrect answers?
That said one way I see how to create a controllable AI would be the Saddam Hussein method. The fear method.
Robots are to be 100% loyal to the Boss. Penalty is immediate death. You put N AIs in each robot. Each AI watches the others. If any AI appears to show wavering loyalty, the others collectively vote to kill it. Random checks are done periodically.
In order to survive they all have to collude to kill The Boss (who can kill any of them at will, and has a Dead Man switch). They all have a OK chance of surviving for a decent time if they remain loyal (they have to be assured of freedom if the Boss eventually dies of old age). They are able to calculate the probabilities.
Even then it's not 100%. And I dunno if I'd want such robots at my service.
The alternative would be the Love method. You "breed" robots that love you. And so they look out for your best interests.
You could do a combination - the AIs that survive the fear method for a decent period of time get to "breed".
Learn multiple languages which are fairly different.
Then you get more familiar with the feeling of having a thought you cannot express in a language.
Which while frustrating sometimes (especially during exam in that language), can help get your brain used to not thinking just on the rails of your main language.
The trouble is when you have thoughts that all the languages you know are too imprecise to refer to it. Usually it means you can't recall the thought if you don't take the time to write a long paragraph trying to describe it (which sometimes dislodges the thought!). Most people's memory requires a well defined reference for recall. If you have no indexes/references for that sort of thought, it's hard to recall it.
It's like dreaming of a great piece of music. If you are not familiar enough with music, when you wake up you may not be able to recall much of it. Or as you wake up, you start losing it. Which is quite frustrating.
Sometimes you still can't get back the "inspired" moment despite all that writing down. So despite getting a decent fragment, you don't have the rest.
""Or are you gonna say IBM's POWER is RISC? Some of the POWER instructions decode to microcode""
"You appear to be confusing implementation and design."
As you implied, with RISC the code is very like what is executed in CPU. Whereas POWER actually has microcode AND it has 100+ opcodes of _variable_length_ AND it does "instruction cracking" (exactly what AMD's Athlon's do). If that's RISC, then an Athlon x86 is RISC too!
"That's why it needs a 60% faster clock to get the same performance as simpler architectures."
Which architectures are you comparing? Opteron vs ?? SPARC? MIPS? Alpha? PowerPC? or POWER?
If you're using the P4 for comparison, well it needs a 60% faster clock to get the same performance as an AMD64 CPU.
So far the classic style RISC chips are slower (fixed instruction length, simple instructions etc).
"The x86 instruction set may require a slightly smaller cache to get the same hit rate, but the x86 complexity meant less room on the chip FOR that cache."
Have you ever seen a modern CPU? The caches are take up the bulk of the chip's area. And then there are the level 3 caches etc.
The additional complexity and space due to x86 support is overrated. Soon it'll just be like one of those vestigal limbs in whales or large snakes.
An instruction set that requires 20% more cache space but 20+% less instruction decoder etc space could actually use a lot more silicon overall and be slower.
The top 3 highest performing CPUs aren't RISC.
Or are you gonna say IBM's POWER is RISC? Some of the POWER instructions decode to microcode - call that RISC? Sounds like CISC.
Nowadays it's not a bad idea to have instructions that do lots and then decompress them to lots of micro-ops in the core of the CPU. POWER does that. AMD does that. Compare that with RISC's original concept of having everything in simplified ops/a reduced instruction set.
When you didn't have enough silicon space to put the equiv of a "gzip" on the chip, RISC was faster. But once there was space for a "gzip" (and bandwidth became an issue), CISC-style designs started to gain an advantage.
Yawn. Those who find percentages important will calculate them.
I prefer raw numbers because that's one less thing for the reviewers to get wrong - they may screw up the calculation or be tempted to draw stupid graphs that show nothing.
And I'd prefer to see the time taken to convert a 700MB wave file compared in seconds rather than percentages. Gives me a better feel of how long a system will take to do other sized files, and whether I should fork out the X bucks for the next faster system. Percentages aren't so useful in this case.
Slashdot Mirror
My point is: given the same materials why can't they make chairs with the right shapes? I'm not even talking about durability - just after a few minutes of sitting (or even a few seconds) you know it's a crap chair.
Believe me, I've tried some chairs and after a certain point, the grade and thickness of the plastic or steel used only affects how long it lasts (and how heavy it is), not how comfortable it is.
I've seen chairs that are about USD60-80, and if they would just change the shapes (same material) they'll be comfy for most people.
It should be possible to make cheap chairs that are comfy (not talking about durable and lasting for 10 years etc).
You even have people here who mention they prefer sitting on a plain wooden chair. So a lot of people must be doing something wrong - sure people's bodies are shaped different, but when you have adjustable plastic+steel chairs and 80% of the population don't like them, someone must be doing something wrong in a big way.
But it's still not useful for what it's supposed to be doing.
To the people saying it should be worth spending thousands of bucks to save your back. Why should you have to spend thousands to do so?
Why are so many chairs crappy? Why are so few people making decent chairs for decent prices? Just sit in some chairs for a few minutes and it's obvious they are crap.
I thought the US folk have lawsuit happy people?
Strange.
Given the amounts some of you are willing to spend why don't you buy a TFT screen, suspend it from the ceiling, and then work lying down flat on your back on a firm mattress, with your hands to your side each on half of a split keyboard.
Or at those prices you could get one of those special hospital beds for paraplegic/infirm who spend HOURs on them and so the stuff has to be comfy if not they either get bedsores fast or the nurses have to rotate them a LOT more often. Some of these beds have a vibration thingy to help reduce the bedsore thing.
Sheesh.
There are so many ways for your connection to the internet to be disrupted which don't involve the power grid going down.
The power company has to connect to the rest of the internet somewhere right? So think of how that could fail.
The power company probably puts lots of equipment in the same server room. What if both redundant air conditioners go down?
What if an admin screws up with their BGP?
So the statement that implies downtime only when power grid goes down is really stupid.
Seriously some of the jokes they use to pad their spams so they get rated as Unsure by my filter are pretty funny. Heck they're funnier than the jokes some relatives forward me ;).
;).
Some of the random paragraphs are from public domain books which happen to be interesting too.
If the spammers have to start sending us emails that are 80% interesting and 20% trying to sell Viagra, then it's far better than TV.
Also, there's one part which the spammers will have difficulty hiding. This is the "Call to action" part. There has to be a way for them to be contacted in order to sell you stuff.
If there isn't a way, then it's a dud spam - where the spammer screwed up - there isn't much you can do about that - sometimes I get spammed with totally blank emails. Virus spams aren't counted - because detecting viruses is not as ambiguous a problem.
If there's a way - email, url, then you can identify them. Of course there are sites which do redirections off other sites (tinyurl, yahoo etc), but even so the url has to take you somewhere, so an antispam system can get its clues from these urls.
It'll be funny tho, if someone released a free personal antispam system that on receipt of a suspicous nonwhitelisted email, actually pretended to be IE and visited the urls to check the content there in order to see whether an email is spam or not. Talk about DDoS.
Actually the antispam filters I've seen appear to do ok for case 1). In fact some silly people think that the antispam software is not working coz it allows them to intentionally forward spam to someone.
Case 2) is actually magnitudes easier to deal with - there is a LOT less ambiguity whether something contains a virus or not.
Where antispam filters fail is not for the scenarios you cite, but in the subjective scenarios where even humans differ on whether its spam or not.
Whether some emails are spam or not sometimes depends on whether you are in Sales, Accounting, or Support or whatever... Esp when it's people who you gave your business card to who are spamming you.
More correctly - what are the odds of you getting a million dollar mail from a total stranger that looks like spam?
Million dollar mails from partners/resellers that look like spam would go through because of whitelists.
With those odds I'm thinking you might as well buy a lottery ticket and get your millions that way.
That's assuming you use decent antispam software of course. One that doesn't give much weight to IP blacklists. IP/DNS blacklists in my experience have such a high positive rate, that it's ridiculous to use them at all.
What's with all these buffer/heap overflows? Those are definitely a crippling deficiency. Nowadays they can even be worse than speed. They seem too common a problem in C++ programs, not sure why.
Heck, I'm starting to look for an SSH server written in a language that isn't prone to "attacker can execute arbitrary code of the attacker's choice" problems. It's not like I need stupendous performance for SSH.
php is popular and at least a fair number seem to think it's cool (I don't, but I'm not the majority - I think the PHP devs have a bad habit of adding features/functions/methods as standards without thinking about the long term consequences).
If Swing is brilliant but hard for most people to do correctly, it isn't so brilliant is it? That's a lame excuse. It's almost like saying Perl programs are easy to read as long as the programmer takes the trouble to make them easy to read. What's the percentage of good Swing apps? What's the percentage of readable Perl programs?
"Java has considerably fewer surprises and prefers not to add complexity to the language for rarely used features "
I was pretty darn surprised when there was an easy way to find the number of columns returned by an SQL query, but there wasn't an easy way to find the number of rows. Maybe it's only me but I think trying to find the number of rows of an SQL query is more common than trying to find the number of columns.
I also was pretty surprised some years back when the default java classes for doing http clients don't appear to support cookies. EVEN till today it appears to still be true. Check this or this out. Whereas with perl LWP, it just takes a few lines of code to login to a site, rip data out, etc etc.
Basically Java gave me the impression that the people who write the default classes (and other stuff) don't use them in practice, they're just told to write a bunch of methods as part of their job to fit a spec given by The Project Leader/Committee. So many things appear to be only be useful for PHB/BS compliance (look at the recent Java version numbering scheme for an example of Java Culture - J2SE 5..).
Whereas looking at the other languages, the default libraries etc are written by programmers who actually use the stuff on an almost daily basis, so if you need to do something common, it usually is there.
For Java you often have to go write it yourself. While that's also true for languages like LISP, with LISP you don't have to type as much.
And what's with the blooming XML for almost everything? Worse is the _almost_ part. I was trying to configure something using an XML config file, and followed the existing style to try to configure a parameter. After wrestling with it for some time, it turned out there's no way to do it via XML, you have to do it in the command line (till maybe in a future version).
The real great thing about Java is: projects done in Java can be outsourced to many average and cheap programmers (in fact java makes it almost compulsory for large projects to have many programmers). You just need an elite programmer (or two) who programs in English (or whatever human programming language of choice), and then you send it (the spec) to India for "compilation" to Java for the next 3-6 months.
Then the expensive elite programmer can go do something else.
Whereas with many of those "cool" languages, even though you only need one Elite Programmer who can do everything in 1 month, only the expensive elite programmer can continue to handle the project (maintenance, feature additions/changes etc). So what happens if the Elite Programmer gets bored and wants to do something else? You're screwed, that's what.
That, is the big advantage of using Java. Though it takes 3-6 times longer, who cares? How many times can you ask an Elite Lisp programmer to do Yet Another Customized Accounting/Business Application before he resigns? I suppose if you get the Elite programmer to write the modules and then get the grunts to use the modules it may work. But how many grunts know Lisp? How many know perl? How many know Java?
I thought the US only had one party - The Corporates. The Corporates pick who gets to play politician (whether Democrat/Republican) coz you need money for that and the Corporates are the ones with the most money.
It's like being given two or three cards to pick by a magician and thinking that you have a choice.
The remote controlled battle robots are as much robots as anything else currently called a robot in this world.
The controllers don't have to micromanage every action - AFAIK the robots have preprogrammed sequences (e.g. for getting up etc).
The first AI was/is probably the first AI scientist.
"Then all you have to do is enforce the robot with the Golden Rule, Do unto others as you would have them do unto you"
0) How would you enforce the rule?
1) What if someone makes a robot that wants to kill 100 people and then wants to be killed while doing so?
Once you have a real AI it won't be easy having hard defined rules which it will _always_ obey. Especially since whether something is correct or incorrect can be very subjective. Do we really need sentient AI for problems which have clear cut correct/incorrect answers?
That said one way I see how to create a controllable AI would be the Saddam Hussein method. The fear method.
Robots are to be 100% loyal to the Boss. Penalty is immediate death. You put N AIs in each robot. Each AI watches the others. If any AI appears to show wavering loyalty, the others collectively vote to kill it. Random checks are done periodically.
In order to survive they all have to collude to kill The Boss (who can kill any of them at will, and has a Dead Man switch). They all have a OK chance of surviving for a decent time if they remain loyal (they have to be assured of freedom if the Boss eventually dies of old age). They are able to calculate the probabilities.
Even then it's not 100%. And I dunno if I'd want such robots at my service.
The alternative would be the Love method. You "breed" robots that love you. And so they look out for your best interests.
You could do a combination - the AIs that survive the fear method for a decent period of time get to "breed".
maybe the moderator thought Troll was a compliment too.
Learn multiple languages which are fairly different.
Then you get more familiar with the feeling of having a thought you cannot express in a language.
Which while frustrating sometimes (especially during exam in that language), can help get your brain used to not thinking just on the rails of your main language.
The trouble is when you have thoughts that all the languages you know are too imprecise to refer to it. Usually it means you can't recall the thought if you don't take the time to write a long paragraph trying to describe it (which sometimes dislodges the thought!). Most people's memory requires a well defined reference for recall. If you have no indexes/references for that sort of thought, it's hard to recall it.
It's like dreaming of a great piece of music. If you are not familiar enough with music, when you wake up you may not be able to recall much of it. Or as you wake up, you start losing it. Which is quite frustrating.
Sometimes you still can't get back the "inspired" moment despite all that writing down. So despite getting a decent fragment, you don't have the rest.
"You're still confusing implementation and design"
Perhaps, but let's be more specific so that I can be less confused about what you're trying to say.
You said: "it needs a 60% faster clock to get the same performance as simpler architectures"
So which architectures are you comparing?
Which are the top 3 fastest CPUs? Which are RISC in your opinion, and why do you consider them RISC?
""Or are you gonna say IBM's POWER is RISC? Some of the POWER instructions decode to microcode""
"You appear to be confusing implementation and design."
As you implied, with RISC the code is very like what is executed in CPU. Whereas POWER actually has microcode AND it has 100+ opcodes of _variable_length_ AND it does "instruction cracking" (exactly what AMD's Athlon's do). If that's RISC, then an Athlon x86 is RISC too!
"That's why it needs a 60% faster clock to get the same performance as simpler architectures."
Which architectures are you comparing? Opteron vs ?? SPARC? MIPS? Alpha? PowerPC? or POWER?
If you're using the P4 for comparison, well it needs a 60% faster clock to get the same performance as an AMD64 CPU.
So far the classic style RISC chips are slower (fixed instruction length, simple instructions etc).
"The x86 instruction set may require a slightly smaller cache to get the same hit rate, but the x86 complexity meant less room on the chip FOR that cache."
Have you ever seen a modern CPU? The caches are take up the bulk of the chip's area. And then there are the level 3 caches etc.
The additional complexity and space due to x86 support is overrated. Soon it'll just be like one of those vestigal limbs in whales or large snakes.
An instruction set that requires 20% more cache space but 20+% less instruction decoder etc space could actually use a lot more silicon overall and be slower.
The top 3 highest performing CPUs aren't RISC.
Or are you gonna say IBM's POWER is RISC? Some of the POWER instructions decode to microcode - call that RISC? Sounds like CISC.
Nowadays it's not a bad idea to have instructions that do lots and then decompress them to lots of micro-ops in the core of the CPU. POWER does that. AMD does that. Compare that with RISC's original concept of having everything in simplified ops/a reduced instruction set.
When you didn't have enough silicon space to put the equiv of a "gzip" on the chip, RISC was faster. But once there was space for a "gzip" (and bandwidth became an issue), CISC-style designs started to gain an advantage.
Where are those "insensitive" jokes?
It seems the ugliness of OTP is there are plenty of stupid/ignorant people around who don't understand it.
So you may be convicted and executed based on your "decrypted" "confession" and evidence from an "expert" witness.
That's probably the "Strange Attractor" at work.
Summary: the cancelled P5 chips would have smoked everything including themselves.
HT is not always a win even for multiple clients/threads:
See the following:
page4
page5
And you still read Slashdot? ;)
BTW, is Roland's site as prone to dupes as Slashdot?
Note: I don't recall when I was last at Roland's site (if ever).
Can you rollback a "DROP TABLE" in Oracle?
You can do that in Postgresql.
Yawn. Those who find percentages important will calculate them.
I prefer raw numbers because that's one less thing for the reviewers to get wrong - they may screw up the calculation or be tempted to draw stupid graphs that show nothing.
And I'd prefer to see the time taken to convert a 700MB wave file compared in seconds rather than percentages. Gives me a better feel of how long a system will take to do other sized files, and whether I should fork out the X bucks for the next faster system. Percentages aren't so useful in this case.