Russian TV
has video, with more technical details.
They put a drag chute on the bomb, which they need if it's a fuel-air explosive - those don't work well if they're moving too fast. It doesn't seem to have a guidance system, so it's more of an area weapon than a bunker-buster. It's dropped from a bomber, so it can be delivered into hostile airspace.
The US's GBU-43/B bomb is GPS/inertial guided, using fins but no drag chute. It's not quite as powerful, but usually hits on target. It's dropped from a C-130 transport, which is not too hard to shoot down, so you have to have total air superiority to use the thing.
Neither has anywhere near the power of even a small tactical nuclear weapon.
The problem is, QNX management has said that before:
"The new QNX initiative consists of several key elements . . .
In recognition of the ease with which developers can obtain and begin development with Linux, QNX has decided to make their new QNX Realtime Platform free "for non-commercial use". Developers can now download the software and its associated development directly from the QNX website.
In response to the growing desire for source code that has resulted from the exploding popularity of open-source Linux, QNX will soon release the source code for many QNX applications, drivers, and libraries."
Last time, the free version of QNX stayed around just long enough that
free software developers ported their major packages to QNX. Then
QNX management yanked it away.
it'd be unwise to assume that nobody else has got one. (large phased-array radars)
There's been less progress in large phased-array radars than one might expect. One would think that air traffic control and weather radar installations would be using them by now, but no. They're still using rotating dishes. The Navy's AEGIS SPY-1 radar, a good phased array radar from the early 1980s, is still roughly the state of the art. There have been upgrades to the signal processing, but the transmitters are still vacuum-tube powered, with tubes for each cell of the phased array antenna. That's why the things cost so much.
This is one of the few vacuum tube applications left, and since nobody is doing R&D on tubes any more, there's not much progress. Semiconductors still don't quite have the power handling capacity for those multi-megawatt nanosecond pulses.
The US has had the Ground Based-Electro-Optical Deep Space Surveillance system since the early 1980s.
GEODSS is an automated sky search telescope system. Multiple sites with multiple 40-inch telescopes search the sky automatically every night, looking for anything that isn't in the catalogues. GEODSS will even detect dark objects that occult stars.
Everybody has automated astronomy now, but it started with GEODSS, around 1980.
GEODSS has an unusual feature for a telescope - illumination. The system can use one of the telescopes at a site to aim a laser light source, while the other telescope looks at the target with the imager. This allows a good look at low-orbit satellites.
The original test installation for GEODSS, at White Sands, NM, is now used by MIT to look for near-Earth objects. They've found 1622 so far. It wouldn't hurt to have more systems working on that problem. A French version of GEODSS would be a win for everyone.
Funny how the hemp promoters are uninterested in other coarse-fiber crops, like jute, sisal, kenaf, and manila. Or in other low-cost sources of cellulose, like straw, bagasse (sugar cane after sugar extraction), and similar agricultural waste.
No, somehow they're attracted only to hemp.
CustomizeGoogle is a special-purpose ad blocker for Google search results. You can turn off "sponsored results", for example. If Google goes over to the dark side, tools like CustomizeGoogle can be used to filter out the ads and dreck.
So there's a backup plan in place if they go evil.
Logic-based AI AI looked so close in the 1960s, once it was realized that you could get a computer to do mathematical logic. All that was necessary was to express the real world in predicate calculus and prove theorems. After all, that's how logicians and philosophers all the way back to Aristotle said thinking worked. Well, no. We understand now that setting up the problem in a formal way is the hard part. That's the part that takes intelligence. Crunching out a solution by theorem proving is easily mechanized, but not too helpful. That formalism is too brittle, because it deals in absolutes.
Expert systems Today, it's clear that they're no smarter than the rules somebody puts in. But back in the 1980s, when I went through Stanford, people like Prof. Ed Feigenbaum were promising Strong AI Real Soon Now from rule based systems. The claims were embarrassing; at least some of that crowd knew better. All their AI startups went bust, the "AI Winter" of low funding followed, and the whole field was stuck until that crowd was pushed aside.
Neural nets / genetic algorithms / learning systems These all belong to the family of hill-climbing optimizers. These approaches work on problems where continuous improvement via tweaking is helpful, but usually max out after a while.
We still don't really understand how evolution makes favorable jumps. I once said to Koza's crowd that there's a Nobel Prize waiting for whomever figures that out. Nobody has won it yet.
Bayesian statistics Now used to do many of the things that used to be done with neural nets, but with a better understanding of what's going on inside. Lots of practical problems in AI, from spam filtering to robot navigation, are yielding to modern statistical approaches. Compute power helps here; these approaches take much floating point math. These methods also play well with data mining. Progress continues.
AI is one of those fields, like fusion power, where the delivery date keeps getting further away. For this conference, the claim is "some time in the next century". Back in the 1980s, people in the field were saying 10-15 years.
We're probably there on raw compute power, even though we don't know how to use it.
Any medium-sized server farm has more storage capacity that the human brain. If we had a clue how to build a brain, the hardware wouldn't be the problem.
For those not understanding what's going on: Kimball ruled that the copyrights on UNIX System V don't belong to SCO, they belong to Novell, because the Asset Purchase Agreement, signed between Novell and Santa Cruz never transferred the copyrights, just the business.
Right. That issue was examined during discovery. SCO wanted the copyrights transferred when they bought the UNIX business, but Novell insisted on full payment before transferring the copyrights. SCO couldn't come up with the up-front cash, so the final deal called for royalty payments over time with Novell holding the copyrights. That's a common way to structure intellectual property deals, especially if there's any possibility that the buyer might go bankrupt.
This didn't happen by accident. Novell was protecting themselves against the possibility that SCO might not pay in full and might go bankrupt. Which, right now, looks like the likely outcome.
Remember Amit Yoran? He was "cyber-security czar" at the US Department of Homeland Security. He started talking about the vulnerabilities implicit in Microsoft's software. His position was downgraded and he resigned in 2004.
Yoran's successor, Gregory Garcia, was a professional lobbyist, not a security expert.
Read the actual decision. (PDF) What the court ruled was that the "gag rule" associated with "National Security Letters" was fundamentally unconstitutional as a First Amendment violation.
The issue is that the FBI can't impose a "gag order" on someone without court approval.
The previous issue, issuance of National Security Letters without court approval as a Fourth Amendment violation, was dealt with when Congress revised the Patriot Act last year to allow recipients of a National Security Letter to challenge them in court, like a subpoena.
As a classic rule of First Amendment jurisprudence, when the Court finds a First Amendment violation, they strike down the entire statute, rather than trying to patch it. That's what the court did here. The court also stayed the execution of the order pending an appeal, which is likely.
It's a narrow holding. The FBI can still issue National Security Letters without going to court first, but anyone who receives one is now in a much stronger position to argue about it. As a practical matter, if you work for an ISP or telco and get a National Security Letter, your response is "This has to go through our lawyers."
How aggressive should systems be about downgrading ratings for web sites? We've been struggling with this for SiteTruth. In addition to SiteTruth's main function, checking business identity, we have some basic phishing checks. We download the PhishTank database every few hours. PhishTank has lists of bad URLs, but now that the smarter phishing sites change URL and even subdomain in each spam e-mail, blocking by URL is no longer effective. So we now flag the entire base domain.
When a site with an open redirector plugs the hole, PhishTank will downgrade those "active phishes" to inactive. We'll then pick that up and rerate them within hours. But until they do, they're in the tank. The whole site.
Too harsh? Realistic? Evolution in action? Comments?
First, get past the blogodreck to the actual work. (Slashdot editors missed a blog troll again.)
Also, this work is several years old. The papers are from 2004 to 2006.
The original article says that the robot has "tilt and angle sensors in all its joints", but that's wrong. It only has one central tilt sensor. That's significant, because if it did have tilt sensors at each joint, system identification would be easier. The algorithm is doing better than one might expect.
This thing is doing what controls people call "automatic system identification". You have some set of sensor inputs and some set of control outputs, and the control system has to figure out how they relate. It does this by adjusting the outputs and watching what happens. There are various statistical techniques for doing this. Calling this "introspection" isn't really correct.
After system identification, the model is inverted, or solved for the inputs in terms of the outputs. The inverted model can then be used as a controller. Given desired outputs, the inputs needed to achieve them can be computed.
The novel result here is that a reasonably decent system identification for a nonlinear system is being performed with a small number of physical tries. That's an improvement over previous methods, which tended to "learn" very slowly. I'd looked at approaches like this for legged locomotion in the past, but the available system identification algorithms weren't good enough. This looks promising.
Moller has been claiming he'd have a flying car Real Soon Now since the 1960s. Here's his 1974 brochure. The schedule back then was "December 31, 1974 - Preliminary test flights complete - December 31, 1976 - Full-scale production begins". Thirty years later...
In the words of the Securities and Exchange Commission investigation (Moller had a bit of trouble about selling unregistered stock, making false and misleading claims, and other securities law violations), "As of late 2002, MI's approximately 40 years' of development has resulted in a prototype Skycar capable of hovering about fifteen feet above the ground."
What Moller is talking about now, the "M200G", is closer to a hovercraft than a flying car. This isn't the big red "flying car" prototype he's been touting for the last ten years; it's a new, simpler model. He does have a prototype flying, at least while tethered to a crane.
Performance is worse than the AvroCar, circa 1960.
It's not that a flying car is impossible. It's that Moller isn't good enough to bring it off. If somebody like Burt Rutan was doing this, it would be flying in a year or two.
The basic problem with a flying car today is that nobody ever found a way to make a cheap, reliable, small jet engine. It's possible to make a small jet engine, but below bizjet size, they don't get much cheaper. That's why general aviation is still mostly piston-powered. Piston-powered VTOL, which is what Moller is trying, is a marginal idea. It was tried extensively in the 1950s, and the power to weight ratio just isn't good enough. Moller is using Wankel engines, which helps a little, but not enough.
One big problem with Wikipedia has been that editor status, and promotion to "adminship", is based on edit counts, the number of times someone has changed something. The editors with huge edit counts don't generally write much; it takes too long. Slashdot karma is a more useful metric than edit counts, but Wikipedia doesn't have anything like karma.
I'd suggested on Wikipedia that we needed a metric for editors like "amount of new text that lasted at least 90 days without deletion". This UCSC thing is a similar metric.
Somewhere in Bentonville, Arkansas, a Wal-Mart executive is deciding how to respond to this pricing move. When the decision is made, calls will go out to record companies, telling them what Wal-Mart is willing to pay. That's what really scares the RIAA.
Other device interrupts are blocked while ISRs run, so ISRs typically do some device book-keeping and then perform the more lengthy transfer of data to or from their device in a Deferred Procedure Call (DPC) that runs with device interrupts enabled. While DPCs execute with interrupts enabled, they take precedence over all thread execution, regardless of priority, on the processor on which they run, and can therefore impede media playback threads.
OK, we're making some progress. At least the kernel isn't actually doing much at interrupt level any more. But apparently the "DPC" mechanism in Windows doesn't go through the regular dispatch mechanism, so it's not subject to normal priority control.
Not so good.
In QNX, where everything that takes any significant time is in user threads, you can get stuff like this to work right by adjusting priorities. You can run your real time code at a higher priority than some drivers, and that's routinely done in demanding applications. But QNX users are forever fussing over thread priorities. You see QNX bug reports like "We did 100 million thread activations from interrupts and two of them took more than 100us before a response. What's wrong?" That sort of complaint gets taken seriously, too. So that's how it's done in the hard real time world.
What's really needed to make this user-friendly is explicit sporadic scheduling. A "sporadic scheduler" lets a thread request NN ms of CPU time every NNN ms. Sporadic scheduling requests are rejected if there are insufficient resources currently uncommitted. If the request is accepted, the thread is guaranteed its slot. Threads that overrun their slots drop in priority and may be preempted.
Applications like games and media players have to calculate how much resources they need and request them, or tell the user "Can't open another media window - computer too busy".
If the app overruns its time slot, the app mis-estimated, and it's the app's fault. This requires designing media apps for predictable CPU consumption, which is standard hard real time program design.
QNX has sporadic scheduling, although it's not used much. It's most useful when the load is somewhat unknown, as on the desktop, rather than in embedded systems. Something to think about for Linux. I wonder if Apple's OSs have sporadic scheduling; I'll have to ask some people I know there. They hired some QNX people for the iPhone, because they needed hard real time expertise.
The V1 was so dumb that defenses only had to be placed on the direct line between the fixed launching ramps and the regular targets. It won't be that easy next time around.
The V1 of WWII was a very simple device, built cheaply out of sheet metal with a crude engine. Range of several hundred miles. Moderately reliable airframe. But the guidance systems of that era had trouble finding London, and hitting a specific military target was hopeless. The same airframe with modern guidance could hit specific buildings. It could become the Third World's answer to US bombing strikes - the AK-47 of air warfare. So far, no one has bothered.
Soldering on jumper wires is only for experimenting. If it really can't be done from software alone, what's needed is a a PC board test fixture. Once you have one of those set up for the job, connecting to the right spots is easy. With that, and a control PC programmed to do all the necessary downloads and updates, the whole operation can be done in a minute or two per phone.
Coming soon to the back room of an off-brand cell phone store near you.
Slashdot Mirror
Russian TV has video, with more technical details. They put a drag chute on the bomb, which they need if it's a fuel-air explosive - those don't work well if they're moving too fast. It doesn't seem to have a guidance system, so it's more of an area weapon than a bunker-buster. It's dropped from a bomber, so it can be delivered into hostile airspace.
The US's GBU-43/B bomb is GPS/inertial guided, using fins but no drag chute. It's not quite as powerful, but usually hits on target. It's dropped from a C-130 transport, which is not too hard to shoot down, so you have to have total air superiority to use the thing.
Neither has anywhere near the power of even a small tactical nuclear weapon.
"Bully" is like playing a Harry Potter game as Neville Longbottom. With that as the only option.
The problem is, QNX management has said that before:
"The new QNX initiative consists of several key elements . . .
That's from a press release back in 2000.
Last time, the free version of QNX stayed around just long enough that free software developers ported their major packages to QNX. Then QNX management yanked it away.
Despite the new press releases, the QNX CVS source repository hasn't been updated in six years.
it'd be unwise to assume that nobody else has got one. (large phased-array radars)
There's been less progress in large phased-array radars than one might expect. One would think that air traffic control and weather radar installations would be using them by now, but no. They're still using rotating dishes. The Navy's AEGIS SPY-1 radar, a good phased array radar from the early 1980s, is still roughly the state of the art. There have been upgrades to the signal processing, but the transmitters are still vacuum-tube powered, with tubes for each cell of the phased array antenna. That's why the things cost so much.
This is one of the few vacuum tube applications left, and since nobody is doing R&D on tubes any more, there's not much progress. Semiconductors still don't quite have the power handling capacity for those multi-megawatt nanosecond pulses.
The US has had the Ground Based-Electro-Optical Deep Space Surveillance system since the early 1980s. GEODSS is an automated sky search telescope system. Multiple sites with multiple 40-inch telescopes search the sky automatically every night, looking for anything that isn't in the catalogues. GEODSS will even detect dark objects that occult stars. Everybody has automated astronomy now, but it started with GEODSS, around 1980.
GEODSS has an unusual feature for a telescope - illumination. The system can use one of the telescopes at a site to aim a laser light source, while the other telescope looks at the target with the imager. This allows a good look at low-orbit satellites.
The original test installation for GEODSS, at White Sands, NM, is now used by MIT to look for near-Earth objects. They've found 1622 so far. It wouldn't hurt to have more systems working on that problem. A French version of GEODSS would be a win for everyone.
To use hemp ropes on ships, they had to coat it in tar across its entire surface just to keep it from rotting.
Yes. Not only that, hemp rope rots from the inside out, so the rope looks good until it breaks under load.
Funny how the hemp promoters are uninterested in other coarse-fiber crops, like jute, sisal, kenaf, and manila. Or in other low-cost sources of cellulose, like straw, bagasse (sugar cane after sugar extraction), and similar agricultural waste. No, somehow they're attracted only to hemp.
That's right. I was expecting to read about how somebody solved the problem. But they're just blithering about it. No useful content here, move along.
CustomizeGoogle is a special-purpose ad blocker for Google search results. You can turn off "sponsored results", for example. If Google goes over to the dark side, tools like CustomizeGoogle can be used to filter out the ads and dreck.
So there's a backup plan in place if they go evil.
OK. here's where we are:
AI is one of those fields, like fusion power, where the delivery date keeps getting further away. For this conference, the claim is "some time in the next century". Back in the 1980s, people in the field were saying 10-15 years.
We're probably there on raw compute power, even though we don't know how to use it. Any medium-sized server farm has more storage capacity that the human brain. If we had a clue how to build a brain, the hardware wouldn't be the problem.
For those not understanding what's going on: Kimball ruled that the copyrights on UNIX System V don't belong to SCO, they belong to Novell, because the Asset Purchase Agreement, signed between Novell and Santa Cruz never transferred the copyrights, just the business.
Right. That issue was examined during discovery. SCO wanted the copyrights transferred when they bought the UNIX business, but Novell insisted on full payment before transferring the copyrights. SCO couldn't come up with the up-front cash, so the final deal called for royalty payments over time with Novell holding the copyrights. That's a common way to structure intellectual property deals, especially if there's any possibility that the buyer might go bankrupt.
This didn't happen by accident. Novell was protecting themselves against the possibility that SCO might not pay in full and might go bankrupt. Which, right now, looks like the likely outcome.
Remember Amit Yoran? He was "cyber-security czar" at the US Department of Homeland Security. He started talking about the vulnerabilities implicit in Microsoft's software. His position was downgraded and he resigned in 2004.
Yoran's successor, Gregory Garcia, was a professional lobbyist, not a security expert.
Read the actual decision. (PDF) What the court ruled was that the "gag rule" associated with "National Security Letters" was fundamentally unconstitutional as a First Amendment violation. The issue is that the FBI can't impose a "gag order" on someone without court approval.
The previous issue, issuance of National Security Letters without court approval as a Fourth Amendment violation, was dealt with when Congress revised the Patriot Act last year to allow recipients of a National Security Letter to challenge them in court, like a subpoena.
As a classic rule of First Amendment jurisprudence, when the Court finds a First Amendment violation, they strike down the entire statute, rather than trying to patch it. That's what the court did here. The court also stayed the execution of the order pending an appeal, which is likely.
It's a narrow holding. The FBI can still issue National Security Letters without going to court first, but anyone who receives one is now in a much stronger position to argue about it. As a practical matter, if you work for an ISP or telco and get a National Security Letter, your response is "This has to go through our lawyers."
there is a good chance the Republicans will not be in power..
That may happen sooner rather than later.
Bush's old Texas crowd is almost gone.
How aggressive should systems be about downgrading ratings for web sites? We've been struggling with this for SiteTruth. In addition to SiteTruth's main function, checking business identity, we have some basic phishing checks. We download the PhishTank database every few hours. PhishTank has lists of bad URLs, but now that the smarter phishing sites change URL and even subdomain in each spam e-mail, blocking by URL is no longer effective. So we now flag the entire base domain.
This can have broad effects. Right now, we're blacklisting all of AOL (SiteTruth report) and all of "live.com" (SiteTruth report). Both AOL and Microsoft Live have redirectors which are being actively exploited by phishing sites. We can't tell their safe URLs from their unsafe URLs, so we have to blacklist the whole domain.
When a site with an open redirector plugs the hole, PhishTank will downgrade those "active phishes" to inactive. We'll then pick that up and rerate them within hours. But until they do, they're in the tank. The whole site.
Too harsh? Realistic? Evolution in action? Comments?
First, get past the blogodreck to the actual work. (Slashdot editors missed a blog troll again.) Also, this work is several years old. The papers are from 2004 to 2006.
The original article says that the robot has "tilt and angle sensors in all its joints", but that's wrong. It only has one central tilt sensor. That's significant, because if it did have tilt sensors at each joint, system identification would be easier. The algorithm is doing better than one might expect.
This thing is doing what controls people call "automatic system identification". You have some set of sensor inputs and some set of control outputs, and the control system has to figure out how they relate. It does this by adjusting the outputs and watching what happens. There are various statistical techniques for doing this. Calling this "introspection" isn't really correct.
After system identification, the model is inverted, or solved for the inputs in terms of the outputs. The inverted model can then be used as a controller. Given desired outputs, the inputs needed to achieve them can be computed.
The novel result here is that a reasonably decent system identification for a nonlinear system is being performed with a small number of physical tries. That's an improvement over previous methods, which tended to "learn" very slowly. I'd looked at approaches like this for legged locomotion in the past, but the available system identification algorithms weren't good enough. This looks promising.
Good robotics work, crap Slashdot article.
Moller has been claiming he'd have a flying car Real Soon Now since the 1960s. Here's his 1974 brochure. The schedule back then was "December 31, 1974 - Preliminary test flights complete - December 31, 1976 - Full-scale production begins". Thirty years later...
In the words of the Securities and Exchange Commission investigation (Moller had a bit of trouble about selling unregistered stock, making false and misleading claims, and other securities law violations), "As of late 2002, MI's approximately 40 years' of development has resulted in a prototype Skycar capable of hovering about fifteen feet above the ground."
What Moller is talking about now, the "M200G", is closer to a hovercraft than a flying car. This isn't the big red "flying car" prototype he's been touting for the last ten years; it's a new, simpler model. He does have a prototype flying, at least while tethered to a crane. Performance is worse than the AvroCar, circa 1960.
It's not that a flying car is impossible. It's that Moller isn't good enough to bring it off. If somebody like Burt Rutan was doing this, it would be flying in a year or two.
The basic problem with a flying car today is that nobody ever found a way to make a cheap, reliable, small jet engine. It's possible to make a small jet engine, but below bizjet size, they don't get much cheaper. That's why general aviation is still mostly piston-powered. Piston-powered VTOL, which is what Moller is trying, is a marginal idea. It was tried extensively in the 1950s, and the power to weight ratio just isn't good enough. Moller is using Wankel engines, which helps a little, but not enough.
One big problem with Wikipedia has been that editor status, and promotion to "adminship", is based on edit counts, the number of times someone has changed something. The editors with huge edit counts don't generally write much; it takes too long. Slashdot karma is a more useful metric than edit counts, but Wikipedia doesn't have anything like karma.
I'd suggested on Wikipedia that we needed a metric for editors like "amount of new text that lasted at least 90 days without deletion". This UCSC thing is a similar metric.
This should come up in the confirmation hearings for the new Attorney General.
Somewhere in Bentonville, Arkansas, a Wal-Mart executive is deciding how to respond to this pricing move. When the decision is made, calls will go out to record companies, telling them what Wal-Mart is willing to pay. That's what really scares the RIAA.
Other device interrupts are blocked while ISRs run, so ISRs typically do some device book-keeping and then perform the more lengthy transfer of data to or from their device in a Deferred Procedure Call (DPC) that runs with device interrupts enabled. While DPCs execute with interrupts enabled, they take precedence over all thread execution, regardless of priority, on the processor on which they run, and can therefore impede media playback threads.
OK, we're making some progress. At least the kernel isn't actually doing much at interrupt level any more. But apparently the "DPC" mechanism in Windows doesn't go through the regular dispatch mechanism, so it's not subject to normal priority control. Not so good.
In QNX, where everything that takes any significant time is in user threads, you can get stuff like this to work right by adjusting priorities. You can run your real time code at a higher priority than some drivers, and that's routinely done in demanding applications. But QNX users are forever fussing over thread priorities. You see QNX bug reports like "We did 100 million thread activations from interrupts and two of them took more than 100us before a response. What's wrong?" That sort of complaint gets taken seriously, too. So that's how it's done in the hard real time world.
What's really needed to make this user-friendly is explicit sporadic scheduling. A "sporadic scheduler" lets a thread request NN ms of CPU time every NNN ms. Sporadic scheduling requests are rejected if there are insufficient resources currently uncommitted. If the request is accepted, the thread is guaranteed its slot. Threads that overrun their slots drop in priority and may be preempted. Applications like games and media players have to calculate how much resources they need and request them, or tell the user "Can't open another media window - computer too busy". If the app overruns its time slot, the app mis-estimated, and it's the app's fault. This requires designing media apps for predictable CPU consumption, which is standard hard real time program design.
QNX has sporadic scheduling, although it's not used much. It's most useful when the load is somewhat unknown, as on the desktop, rather than in embedded systems. Something to think about for Linux. I wonder if Apple's OSs have sporadic scheduling; I'll have to ask some people I know there. They hired some QNX people for the iPhone, because they needed hard real time expertise.
The V1 was so dumb that defenses only had to be placed on the direct line between the fixed launching ramps and the regular targets. It won't be that easy next time around.
Views of 2029:
Imagine if someone decided to design an open source cruise missile. ... DIY cruise missile
That guy was developing something that some strategic intel people have been expecting for years - a simple V1-like UAV, but with modern guidance.
The V1 of WWII was a very simple device, built cheaply out of sheet metal with a crude engine. Range of several hundred miles. Moderately reliable airframe. But the guidance systems of that era had trouble finding London, and hitting a specific military target was hopeless. The same airframe with modern guidance could hit specific buildings. It could become the Third World's answer to US bombing strikes - the AK-47 of air warfare. So far, no one has bothered.
Soldering on jumper wires is only for experimenting. If it really can't be done from software alone, what's needed is a a PC board test fixture. Once you have one of those set up for the job, connecting to the right spots is easy. With that, and a control PC programmed to do all the necessary downloads and updates, the whole operation can be done in a minute or two per phone.
Coming soon to the back room of an off-brand cell phone store near you.