The guy is just trolling for hits on his web site, with these so-called "stories". He wants to be
George Gilder, but he's not good enough.
Twenty years ago, you could detach from a job on TOPS-20 and reattach to it from another terminal later. Many centralized systems have had features like that. It wouldn't be all that hard to do with X and Linux, if anybody wanted to.
Most of the people proposing stuff like this are desperately trying to lock customers in to some service for which they can charge a monthly fee. With add-ons! Remember Application Service Providers?
We had this gizmo for a while, before CDs fully supplanted cassettes, and before burnable CDs were de rigeur
Oh, yes, that thing. That CD jukebox showed up in Silicon Valley surplus stores for years. It was a 3x audio CD player, and people were struggling to modify them to run at 1x.
Yeah, and 40% of apparel is eventually sold at a discount.
The big breakthrough in the rag trade has been better supply chain and inventory management. Most of the big players finally have this down. Now, bell-bottom production gets turned off before you have three container ships of the things in transit from Mayalasia to Long Beach.
Bad supply chain management leads to stuff like this:
Used Clothing: SUPER PRICE FOR A LIMITED TIME ONLY! A huge selection of mixed men's, women's and children's clothing sold by the bail. Bails are made into 1,000 pounds each. Your cost is as low as $0.39 per pound. Minimum order is approximately 20,000 pounds (20' container).
While on Slashdot we hear endlessly about Red Hat, Debian, etc., the volume manufacturers are going with Lindows, Linspire, Thiz, and in China, Red Flag. Maybe those should get more coverage. What do the installed base figures look like?
Lesson 2: Opening a "MS-DOS" window on Windows 95/98. (Not an NT-family OS, even though this is a corporate networking class.)
Lesson 3: Installing a network card. ("Try to see how a Token Ring NIC differs from an Ethernet NIC.")
A little further along, there are chapters on binary arithmetic, hex arithmetic, IP addressing, and the symbols Cisco uses in their manuals.
Then, immediately after the chapter on IP addressing, things suddenly get complicated:
You are the network administrator for an upstart website publishing company. They have
offices in two adjacent buildings on different floors. Lately, they have realized the costs
of their individual Internet accounts far exceeds the costs of installing and maintaining a
T-1 line. As the network guru you are to design a network that will utilize FDDI between
the buildings. The west building uses floors 3, 4, and 5 for the sales and admin staff.
Here you will want to use a CISCO Catalyst 5000 with a FDDI module, a management
module, and a 24-port switch module. From there each floor will distribute access via a
CISCO 1924 switch to each of its 20 nodes (workstations, servers, and printers). The east
building uses floors 1 through 5 for the design and engineering staff. Here you will want
to use a CISCO Catalyst 5500 with a FDDI module, a management module, and a 24-port
switch module. You will also have a CISCO 2610 router with T-1 module, and a
Kentrox CSU/DSU for your full T-1 line. Your ISP, ComBase has sold you two blocks
of 62 IP addresses: 198.74.56.x (1-62) and (65-126). Combase will also provide the DNS
services, unlike most ISP's where more than 24 IP's are ordered. Design your network,
including cabling and grounds, to include all IP's, subnet masks, gateways, and anything
else you need to include.
This is before they've mentioned how to configure, operate or use any of that stuff. Wierd.
Some quotes:
"Supercomputer--See Nasa, Berkely, MIT, etc. Kind of
like the W.O.P.R. in Wargames."
Genetic algorithms are methods for optimization in bumpy spaces. The basic goal is "find x such that f(x) is maximized".
As optimization algorithms, they should always be tested against the two simple optimization algorithms - basic hill climbing, and random search.
If the search space is not dominated by local maxima, basic hill climbing (go for the best neighboring value) will work. And it will be fast. If the function is differentiable, it can be orders of magnitude faster than other methods,
because you can use a variant of Newton's Method.
If the search space is small, random search (just guessing) will work by exhaustively searching the space. This is obvious, but tends to be ignored in academic papers all too often.
This discussion also applies to neural nets and simulated annealing.
Now this article at least describes a problem for which a GA might actually be useful. Many such articles don't. But they haven't demonstrated that you need a bumpy hill-climbing algorithm.
This is why, despite all the hype, GAs, neural nets, and such aren't used all that much. The search space has to have the right properties.
Not too small, not too big, bumpy, but not too bumpy.
Anybody with a spectrum analyzer can see what's going on in a case like this. Many advanced hams have one. Cell phone and cable TV service operations usually have one around. Even a handheld multiband radio with a signal strength meter is enough to get a clue. Ideally, you'd want one of these.
Anybody with a sizable WLAN operation probably should have one of those around.
It's not like RF interference isn't well understood.
If you're getting interference with a keyless entry device at very short range, the interference source is probably nearby. Very nearby, like tens of meters. There's an inverse square law, remember. Somebody in that parking lot has something that's emitting.
Sure, an Aegis battlecruser could point its phased array radar in your direction, hold the beam stationary. and send a few megawatts down a narrow beam out to the horizon, but that's unlikely.
Few smaller radars have that kind of power, directionality, and steerability. You still have to have near line of sight, anyway.
Get a directional antenna and a signal strength meter, and you'll find the source.
MIT used to have a Cyc-based system on line, but it was so lame they took it down.
They'd loaded it up with information about the MIT/Cambridge area, and information about the Middle East. Suggested queries were things like "Who is the king of Jordan", and "Is MIT in Cambridge?".
So I tried queries like "Who is the king of Israel",
which returned the name of the premier of Israel.
Inference was broken. I asked "Is MIT in Cambridge" - Yes. "Is Cambridge in Massachusetts?" - Yes. "Is MIT in Massachusetts?" - Don't know.
That was lame. It was really no smarter than a search engine. It seemed to be on a par with Ask Jeeves. That's embarassing. Cyc is supposed to be able to do simple inference.
As a promotion, it would make sense to offer many films as free 320x240 downloads with mono audio. If it's any good, you'll want to see the film version.
It's fun watching the XML kiddies re-invent concepts from LISP. They just re-invented property lists,
"is-a" links, and much of the baggage that made SGML painful.
Knowledge representation via "is-a" links has been tried, and it breaks down rather quickly.
Read "Artificial Intelligence meets Natural Stupidity", by Drew McDermott, for a 20 year old critique of this concept. It's overkill for searching, and not powerful enough for reliable automated question answering.
The Cyc debacle illustrates how much work you have to put into tagging to get very little out. After twenty years of that money sink, it's still useless.
Very true. Although, as with a web browser, you could have a TV with multiple "channels" open simultaneously. "Picture in Picture" is a primitive form of this.
When you think about it, broadcasting is "push technology". Push technology lost out decisively to the "pull technology" of the web. Eventually this will happen to TV.
The concept of "channels" will become more of a marketing concept than a technical one. AOL has "channels". MSN used to (do they still?). But they have no real existence other than as a naming convention.
This has already happened for cell phones. Some cell phone providers are now selling "multiple line" cell phones. This is really call waiting with a different user interface. There's only one talk path. But they can sell those "extra lines" for $10/month.
Many complaints. Litigation. Suspicion of product tampering. Worst case was when one was opened on an aircraft, a bomb was suspected, and an emergency landing was made. The annoyed airline switched from Coke to Pepsi.
This sounds like a multiplayer gamer configuration.
Unlike most shared-CPU systems, everybody has a 3D video card, although they have to be PCI boards. With everybody on the same CPU, latency is a non-issue. Fast FPS games should synchronize perfectly. That tightly synchronized feel will make for much better head to head gameplay.
This particular company made a bad partnering decision, though; their LCD panel vendor didn't get funding and tanked. (Venture capitalists refer to this as a deal with "too many moving parts").
As someone mentioned above, Humiseal is an option. They have some heavier coatings than Fine-L-Kote.
Encapsulating a the whole board in a plastic brick is possible. Dow Corning has products for doing this. It's not at all unusual. Many industrial electronics boards get this kind of treatment.
What makes this so great? They still have to have a light source and enough power to drive it.
The actual imaging component of a projector isn't that big. Look at the TI DLP chip. Their projectors are already down to 2.2 pounds.
Color is a problem. Currently, you need either a color wheel for field-sequential color or three imaging chips, which looks better. This new "holographic" display has the same problem. Note that their demo image is greyscale.
What's really needed are powerful LED arrays as the light source. If you could change the light source color at a few KHz, which LEDs can easily do, a one-chip DLP projector without a color wheel would work. With an LED light source, you could do some other obvious power-saving tricks, too.
You need no more light output than the brightest pixel in that color in that frame. With sectional lighting, maybe less.
LEDs with enough light output for this are not far off. LEDs have taken over automotive taillights, and white LED automotive headlamps are expected in 2006. Toyota showed a car with LED headlamps in 2003.
That direction is more likely to result in smaller projectors than this "holographic" thing.
I wish. Automotive electronic units are routinely conformal coated, but most consumer devices are not. Many blank boards have a masked insulating layer on top, but that's different than a conformal coat. A conformal coat is applied after the parts are mounted, so the whole board becomes a sealed unit.
It's a flammable, toxic chemical mixture until it dries; you need gloves, goggles, a respirator mask, and proper flammable liquid storage. Cover connectors with masking tape before spraying.
It's a clear coat, but glows in UV, so you can check for missed spots.
That's life for most women who model. Modelling is a good paying job for a few hundred women in NY and LA. Outside of that, models need day jobs.
If you spend any time in LA, you'll meet actress/model/waitress types. I've had friends in that trap - a minor screen credit, a few TV commercials, occasional extra work, and a day job as a waitress. It's sad when they have the acting bug but can't break into the industry. Some spend years in that state, until they're over the hill and stuck in a low-paying job in West Hollywood. Even the ones with SAG or AFTRA cards don't typically make all that much.
Trade show work is one way women in that situation make a few extra bucks. It's not a full time job.
Ask trade-show models if they've done any film or TV work. Either they have, or they want to.
SCOX has been trading around $5 ± 0.50 for about a month, with low trading volume. Below $5, you can't short a NASDAQ stock, so the bears have dropped out of the market. (Two German exchanges with different short-selling rules recently listed SCOX, to get around the NASDAQ rule. This came up in the SCO conference call, and Darl was annoyed about it.)
Last Wednesday, around 11 am, SCOX suddenly went up to $6.25, in heavy trading. Somebody made that happen. It wasn't a random event. But whomever did it (and it could be the announced SCO stock buyback program) didn't put in enough money to keep the price up, and it started sliding back down that afternoon. Today, Friday, it's back to where it was at the beginning of the week, close to $5. The trading volume is back down, too.
We've seen a few attempts by SCO to prop up their stock. Each one has failed. Except to correct for the distortions caused by the buyback program, the market is ignoring SCOX. Volume is far below what it was last year. When the legal mess starts to generate some decisions, we'll see some activity, but until then, it looks like nothing much will happen.
Here's the company web site. No useful info, just a press release and a note about financing. This company may just be the one guy named Aaron Aaron, who's mentioned in the press report.
Also, the company's web site says "is developing", not "has developed". So it doesn't work yet.
There's no fundamental reason this can't work, but it might take a big active antenna array to do it.
Beam steering takes space, and fine beam steering takes lots of active antenna elements. Phased arrays won't work for ultrawideband (think about it), so you probably have to emit a nondirectional signal and do all the processing on the receive side. Or you can move the sensor around and build up a picture, an approach used for prototypes of land mine detection systems. There's considerable interest in ground-penetrating radar of this type.
Slashdot Mirror
There's a project to do this, leaving behind a memorial to Ronald Reagan at each stop.
Visiting every Starbucks is like visiting every Burger King.
Twenty years ago, you could detach from a job on TOPS-20 and reattach to it from another terminal later. Many centralized systems have had features like that. It wouldn't be all that hard to do with X and Linux, if anybody wanted to.
Most of the people proposing stuff like this are desperately trying to lock customers in to some service for which they can charge a monthly fee. With add-ons! Remember Application Service Providers?
Oh, yes, that thing. That CD jukebox showed up in Silicon Valley surplus stores for years. It was a 3x audio CD player, and people were struggling to modify them to run at 1x.
The big breakthrough in the rag trade has been better supply chain and inventory management. Most of the big players finally have this down. Now, bell-bottom production gets turned off before you have three container ships of the things in transit from Mayalasia to Long Beach.
Bad supply chain management leads to stuff like this:
While on Slashdot we hear endlessly about Red Hat, Debian, etc., the volume manufacturers are going with Lindows, Linspire, Thiz, and in China, Red Flag. Maybe those should get more coverage. What do the installed base figures look like?
-
Lesson 1: Finding CISCO's web site.
-
Lesson 2: Opening a "MS-DOS" window on Windows 95/98. (Not an NT-family OS, even though this is a corporate networking class.)
-
Lesson 3: Installing a network card. ("Try to see how a Token Ring NIC differs from an Ethernet NIC.")
-
A little further along, there are chapters on binary arithmetic, hex arithmetic, IP addressing, and the symbols Cisco uses in their manuals.
Then, immediately after the chapter on IP addressing, things suddenly get complicated:
-
You are the network administrator for an upstart website publishing company. They have
offices in two adjacent buildings on different floors. Lately, they have realized the costs
of their individual Internet accounts far exceeds the costs of installing and maintaining a
T-1 line. As the network guru you are to design a network that will utilize FDDI between
the buildings. The west building uses floors 3, 4, and 5 for the sales and admin staff.
Here you will want to use a CISCO Catalyst 5000 with a FDDI module, a management
module, and a 24-port switch module. From there each floor will distribute access via a
CISCO 1924 switch to each of its 20 nodes (workstations, servers, and printers). The east
building uses floors 1 through 5 for the design and engineering staff. Here you will want
to use a CISCO Catalyst 5500 with a FDDI module, a management module, and a 24-port
switch module. You will also have a CISCO 2610 router with T-1 module, and a
Kentrox CSU/DSU for your full T-1 line. Your ISP, ComBase has sold you two blocks
of 62 IP addresses: 198.74.56.x (1-62) and (65-126). Combase will also provide the DNS
services, unlike most ISP's where more than 24 IP's are ordered. Design your network,
including cabling and grounds, to include all IP's, subnet masks, gateways, and anything
else you need to include.
Some quotes:This is before they've mentioned how to configure, operate or use any of that stuff. Wierd.
"Supercomputer--See Nasa, Berkely, MIT, etc. Kind of like the W.O.P.R. in Wargames."
If the search space is not dominated by local maxima, basic hill climbing (go for the best neighboring value) will work. And it will be fast. If the function is differentiable, it can be orders of magnitude faster than other methods, because you can use a variant of Newton's Method.
If the search space is small, random search (just guessing) will work by exhaustively searching the space. This is obvious, but tends to be ignored in academic papers all too often.
This discussion also applies to neural nets and simulated annealing.
Now this article at least describes a problem for which a GA might actually be useful. Many such articles don't. But they haven't demonstrated that you need a bumpy hill-climbing algorithm.
This is why, despite all the hype, GAs, neural nets, and such aren't used all that much. The search space has to have the right properties. Not too small, not too big, bumpy, but not too bumpy.
Garage.com has been doing this for years. They have some modest successes.. Guy Kawasaki ran Garage.com for a while, after he quit Apple.
If you're getting interference with a keyless entry device at very short range, the interference source is probably nearby. Very nearby, like tens of meters. There's an inverse square law, remember. Somebody in that parking lot has something that's emitting.
Sure, an Aegis battlecruser could point its phased array radar in your direction, hold the beam stationary. and send a few megawatts down a narrow beam out to the horizon, but that's unlikely. Few smaller radars have that kind of power, directionality, and steerability. You still have to have near line of sight, anyway.
Get a directional antenna and a signal strength meter, and you'll find the source.
They'd loaded it up with information about the MIT/Cambridge area, and information about the Middle East. Suggested queries were things like "Who is the king of Jordan", and "Is MIT in Cambridge?". So I tried queries like "Who is the king of Israel", which returned the name of the premier of Israel.
Inference was broken. I asked "Is MIT in Cambridge" - Yes. "Is Cambridge in Massachusetts?" - Yes. "Is MIT in Massachusetts?" - Don't know.
That was lame. It was really no smarter than a search engine. It seemed to be on a par with Ask Jeeves. That's embarassing. Cyc is supposed to be able to do simple inference.
As a promotion, it would make sense to offer many films as free 320x240 downloads with mono audio. If it's any good, you'll want to see the film version.
Seen a console with multiple displays yet?
Knowledge representation via "is-a" links has been tried, and it breaks down rather quickly. Read "Artificial Intelligence meets Natural Stupidity", by Drew McDermott, for a 20 year old critique of this concept. It's overkill for searching, and not powerful enough for reliable automated question answering.
The Cyc debacle illustrates how much work you have to put into tagging to get very little out. After twenty years of that money sink, it's still useless.
When you think about it, broadcasting is "push technology". Push technology lost out decisively to the "pull technology" of the web. Eventually this will happen to TV.
The concept of "channels" will become more of a marketing concept than a technical one. AOL has "channels". MSN used to (do they still?). But they have no real existence other than as a naming convention.
This has already happened for cell phones. Some cell phone providers are now selling "multiple line" cell phones. This is really call waiting with a different user interface. There's only one talk path. But they can sell those "extra lines" for $10/month.
Many complaints. Litigation. Suspicion of product tampering. Worst case was when one was opened on an aircraft, a bomb was suspected, and an emergency landing was made. The annoyed airline switched from Coke to Pepsi.
This sounds like a multiplayer gamer configuration. Unlike most shared-CPU systems, everybody has a 3D video card, although they have to be PCI boards. With everybody on the same CPU, latency is a non-issue. Fast FPS games should synchronize perfectly. That tightly synchronized feel will make for much better head to head gameplay.
This particular company made a bad partnering decision, though; their LCD panel vendor didn't get funding and tanked. (Venture capitalists refer to this as a deal with "too many moving parts").
Encapsulating a the whole board in a plastic brick is possible. Dow Corning has products for doing this. It's not at all unusual. Many industrial electronics boards get this kind of treatment.
The actual imaging component of a projector isn't that big. Look at the TI DLP chip. Their projectors are already down to 2.2 pounds.
Color is a problem. Currently, you need either a color wheel for field-sequential color or three imaging chips, which looks better. This new "holographic" display has the same problem. Note that their demo image is greyscale.
What's really needed are powerful LED arrays as the light source. If you could change the light source color at a few KHz, which LEDs can easily do, a one-chip DLP projector without a color wheel would work. With an LED light source, you could do some other obvious power-saving tricks, too. You need no more light output than the brightest pixel in that color in that frame. With sectional lighting, maybe less.
LEDs with enough light output for this are not far off. LEDs have taken over automotive taillights, and white LED automotive headlamps are expected in 2006. Toyota showed a car with LED headlamps in 2003.
That direction is more likely to result in smaller projectors than this "holographic" thing.
You can conformal-coat boards yourself, using Fine-L-Kote spray. We use this stuff on the Overbot.
It's a flammable, toxic chemical mixture until it dries; you need gloves, goggles, a respirator mask, and proper flammable liquid storage. Cover connectors with masking tape before spraying. It's a clear coat, but glows in UV, so you can check for missed spots.
If you spend any time in LA, you'll meet actress/model/waitress types. I've had friends in that trap - a minor screen credit, a few TV commercials, occasional extra work, and a day job as a waitress. It's sad when they have the acting bug but can't break into the industry. Some spend years in that state, until they're over the hill and stuck in a low-paying job in West Hollywood. Even the ones with SAG or AFTRA cards don't typically make all that much.
Trade show work is one way women in that situation make a few extra bucks. It's not a full time job.
Ask trade-show models if they've done any film or TV work. Either they have, or they want to.
Last Wednesday, around 11 am, SCOX suddenly went up to $6.25, in heavy trading. Somebody made that happen. It wasn't a random event. But whomever did it (and it could be the announced SCO stock buyback program) didn't put in enough money to keep the price up, and it started sliding back down that afternoon. Today, Friday, it's back to where it was at the beginning of the week, close to $5. The trading volume is back down, too.
We've seen a few attempts by SCO to prop up their stock. Each one has failed. Except to correct for the distortions caused by the buyback program, the market is ignoring SCOX. Volume is far below what it was last year. When the legal mess starts to generate some decisions, we'll see some activity, but until then, it looks like nothing much will happen.
Also, the company's web site says "is developing", not "has developed". So it doesn't work yet.
There's no fundamental reason this can't work, but it might take a big active antenna array to do it. Beam steering takes space, and fine beam steering takes lots of active antenna elements. Phased arrays won't work for ultrawideband (think about it), so you probably have to emit a nondirectional signal and do all the processing on the receive side. Or you can move the sensor around and build up a picture, an approach used for prototypes of land mine detection systems. There's considerable interest in ground-penetrating radar of this type.