I have a different take. His previous development work existed purely in the digital domain. Even if he was successful running huge projects previously I think autonomous driving is a different beast that he was poorly prepared for.
For autonomous driving you need to integrate noisy sensor data from a chaotic real world environment and somehow use that data in a logic process that makes repeatably good decisions and fails gracefully. You need to understand the abilities and limits of the sensors, the range of operating environments and you need to iterate in consultation with the sensor engineers to improve the data delivered into the software in a cost effective way. This is all "get your hands dirty" real world engineering - not his background.
They hired a coder when they should have hired an engineer.
What about non-standard characters? Is the whole ASCII set generally available? Some websites are explicit about which characters are valid but many say nothing.
Most attackers using a rainbow table or brute force would probably not include (Alt-"214") in any of their attack attempts.
Those small Honda inverter gensets barely purr, but they likely need a better power to weight ratio. So basically it's waiting for a better power source... Paging Mr. Stark?
Humans in deep space for any length of time will need serious shielding to avoid the health risks of ionizing radiation (gamma rays, etc). Traditional shielding is heavy and crowds out payload. Without a breakthrough in shielding manned space flight can't leave our planet's protection (the magnetosphere) for any length of time. No moon base - at least on the surface, no Mars missions, no Lagrange point space stations.
Tesla's system is like a two level RAID array for batteries. The car can work around weak cells and failing modules with software and targeted replacement (a RAID6+ of modules employing RAID6+ with the cells). One monolithic battery or even a small number of them would not have the same flexibility or replacebility. Also Tesla's business goals would have them buying a vast fraction of the battery production in short order. It made sense to start with the most widely manufactured 18650 cell. Now that they are building their own factory I imagine you'll see custom packs perfectly suited to their needs.
Lots of projects on Endless-Sphere are now using pouch cells. They have more kwh per unit (~20 Ahr vs 2.2-3Ahr), stack densely and employ tabs that make it easy to connect them in parallel or series.
That's the biggest issue with implants. They seem to cause harm to the adjacent tissue and eventually stop working. We need "softer" implants that last a long time themselves and let the tissue they are connected to last for a long time also. It seems to me that grids of needles are a dead-end for long term viability.
The earth absorbs and sheds massive amounts of (thermal) energy annually. It absolutely dwarfs the heat created by our energy production. Global warming is caused by a by-product of combustion altering the planet's ability to shed heat from sun. The minuscule amount of heat from our coal plants is a drop in the ocean compared to the solar heat trapped by the extra CO2.
This has been an issue for airline pilots. Auto-pilot is great at saving fuel but it makes the cruising portion of the flight so boring that pilots stop paying attention.
Even though Mars is cold the ultra low atmospheric pressure will increase the rate of sublimation for the exterior. Sublimation is direct solid->gas phase change. It's the reason snow piles will get smaller even when the outdoor temps never rise above freezing.
Gasoline seems "safe" because the fuel needs to mix with atmospheric oxygen and that slows the energy release, especially if it's in a container.
A capacitor or lithium-ion battery is more akin to a rocket - fuel and oxidizer held in close proximity, or gun powder where they are literally mixed together.
So print a sacrifical mold and cast the part in aluminum. These are early days... It's not going to change world overnight. But additive manufacture is going to creep up on subtractive. I have a rep rap and it's not so useful, mostly because the i2 design was lousy and the mechatronics are a bit too clockwork. Related - reverse kinetics aren't simple. So what... GE makes turbine blades that are impossible to make via milling or casting. They're top down and enthusiast amateurs do the bottom up. It'll change how we make finished goods in a profound way, just not right away.
Your last point about the Marine requirements is spot on.
They should have designed an Air Force / Navy only version and gotten that fully operational: software bugs worked out, airframe defects fixed, flight/battle tested etc. The costs of the program would be much lower, so more planes, fewer design compromises, more buyers. By that point (which is several years from now) typically there are airframe optimizations and engine upgrades ready that were not available in the initial design. Take those improvements, apply to a dual seat trainer airframe (so you have enough space) and make your STOVL Marine variant. If weight is still a problem consider dropping stealth. Anti SAM and air superiority are a job for the Air Force.
Instead the tail wagged the dog with Marine variant and now we have a big ol' expensive mess.
Re: CAS - I think it is likely to become a drone mission anyhow. Better to have ordinance on a cheap airframe with no pilot. It doesn't matter if it gets shot down and you can have a larger number of cheaper planes. Software flies to and from the battlefield so fewer remote pilots needed - done.
You wouldn't typically use a single neural network from input (LIDAR, video, gyros, accelerometers, etc) to output (steering and pedals)
More typically you'd use different neural networks to tackle steps in the chain. One might identify the borders of the road and the median. Another might pick out cars. Another might project the position of the cars in the future. Each would be easier to test individually. You might also have a "supervisor" that looked for disagreement or inconsistencies between the systems and notify the driver to take over.
Not even close in the US. UK and Europe count the cycles and try to average 60 Hz over minutes/hours/day with minor adjustments. The US has massive drift and variability.
Criminal investigations is a niche use. The broad use and real revenue would be an adult picture of your unborn child or new baby. Future parents already pay $300-400 for 3d sonograms of their fetus. Imagine seeing your new baby's face each year from two to twenty years old.
You could even sequence a couple individually and show the full range of boy and girl facial outcomes with probabilities. Right now they can use some morphing techniques as a kludge but genetics could be MUCH more predictive.
This. RPI2 is a OUYA killer... huge installed base, same access to emulators, can use stock wireless console controllers like PS3 with a ~$5 bluetooth dongle.
It gets good reviews on the Kodi forums. It's Kodi, Netlfix and Amazon video in one small box. The CPU is powerful so theoretically a decent gaming platform though I can't speak to that. I went rpi-2 for Kodi so no Netflix but fortunately my TV and/or tablets do.
Ouya has loads of competition now from ARM "sticks" and media adapters like the Fire, Roku or Cu Box. And each year brings more capable hardware while Ouja stays the same. The new raspberry pi 2 or Amazon Fire are arguably superior in all ways. Certainly both those alternatives make excellent XBMC/Kodi boxes.
And competition has also come from tablets in terms of casual gaming. Tablets benefit from huge economies of scale and large online market ecosystems. Ouja was always going to be a niche market appealing to techies and gamers.
Can run x86 binaries and uses Intel HD graphics, so no gfx driver issues. Roku or other ARM sticks can't run x86 binaries and you have to deal with their binary blob gfx drivers that generally don't play well with new kernels. ARM binaries are vary by core vintage and soft vs hard float. Then there is the variety of gpu cores: MALI, Vivante, PowerVR, Broadcom. By contrast Intel has open source options and Intel supplied binaries that get updated, especially because the desktop chips use the same gpu.
I tried upgrading my headless and slow single core ARM server to an IMX6 quad core. In theory it would have been great - in reality it never worked right and the gfx issues were a nightmare to sort out. I ended up with a J1900 board for not much more and it works like a dream.
Companies have gotten really good a simplifying human jobs so that new hires with few skills can be quickly trained up to replace underperforming or otherwise problematic workers. There are high paying management jobs (a few of them) for producing and optimizing employment manuals, procedures, performance targets and input kiosks so that the absolute lowest common denominator hire can quickly fill a void.
As an example McDonalds "upgraded" their order taking turrets from using words for each food item to pictures for each food item. That meant they could employ people who couldn't read, because I guess literacy was a limiting requirement in their hiring process. McDonalds employs over 400,000 people. Just a small "savings" across that employment base is worth millions. That millions of savings get's split between shareholders and the top tier of management who designs and implements these "process enhancements".
And the new thing is to order online from your smartphone and pick it up at the counter. That gets rid of the order taker entirely and you can staff with mostly "behind the scenes" worker bees that don't even have to speak English. That is until you can get a robot to make the food too.
Call centers have been doing this for years with average call time metrics, flow charts for addressing caller needs, etc... It's happening in lots and lots of industries now.
LEGO molds are some of the best in general manufacturing. If a 3d printer could make a LEGO brick as good as the original it would imply that you could print virtually any plastic part to sufficient tolerance.
Slashdot Mirror
I have a different take. His previous development work existed purely in the digital domain. Even if he was successful running huge projects previously I think autonomous driving is a different beast that he was poorly prepared for.
For autonomous driving you need to integrate noisy sensor data from a chaotic real world environment and somehow use that data in a logic process that makes repeatably good decisions and fails gracefully. You need to understand the abilities and limits of the sensors, the range of operating environments and you need to iterate in consultation with the sensor engineers to improve the data delivered into the software in a cost effective way. This is all "get your hands dirty" real world engineering - not his background.
They hired a coder when they should have hired an engineer.
What about non-standard characters? Is the whole ASCII set generally available? Some websites are explicit about which characters are valid but many say nothing.
Most attackers using a rainbow table or brute force would probably not include (Alt-"214") in any of their attack attempts.
Those small Honda inverter gensets barely purr, but they likely need a better power to weight ratio. So basically it's waiting for a better power source... Paging Mr. Stark?
Humans in deep space for any length of time will need serious shielding to avoid the health risks of ionizing radiation (gamma rays, etc). Traditional shielding is heavy and crowds out payload. Without a breakthrough in shielding manned space flight can't leave our planet's protection (the magnetosphere) for any length of time. No moon base - at least on the surface, no Mars missions, no Lagrange point space stations.
100% this
Tesla's system is like a two level RAID array for batteries. The car can work around weak cells and failing modules with software and targeted replacement (a RAID6+ of modules employing RAID6+ with the cells). One monolithic battery or even a small number of them would not have the same flexibility or replacebility. Also Tesla's business goals would have them buying a vast fraction of the battery production in short order. It made sense to start with the most widely manufactured 18650 cell. Now that they are building their own factory I imagine you'll see custom packs perfectly suited to their needs.
Lots of projects on Endless-Sphere are now using pouch cells. They have more kwh per unit (~20 Ahr vs 2.2-3Ahr), stack densely and employ tabs that make it easy to connect them in parallel or series.
That's the biggest issue with implants. They seem to cause harm to the adjacent tissue and eventually stop working. We need "softer" implants that last a long time themselves and let the tissue they are connected to last for a long time also. It seems to me that grids of needles are a dead-end for long term viability.
The earth absorbs and sheds massive amounts of (thermal) energy annually. It absolutely dwarfs the heat created by our energy production. Global warming is caused by a by-product of combustion altering the planet's ability to shed heat from sun. The minuscule amount of heat from our coal plants is a drop in the ocean compared to the solar heat trapped by the extra CO2.
This has been an issue for airline pilots. Auto-pilot is great at saving fuel but it makes the cruising portion of the flight so boring that pilots stop paying attention.
Even though Mars is cold the ultra low atmospheric pressure will increase the rate of sublimation for the exterior. Sublimation is direct solid->gas phase change. It's the reason snow piles will get smaller even when the outdoor temps never rise above freezing.
Gasoline seems "safe" because the fuel needs to mix with atmospheric oxygen and that slows the energy release, especially if it's in a container.
A capacitor or lithium-ion battery is more akin to a rocket - fuel and oxidizer held in close proximity, or gun powder where they are literally mixed together.
3 to 22 minute communication delay makes it very hard to tele-operate.
So print a sacrifical mold and cast the part in aluminum. These are early days... It's not going to change world overnight. But additive manufacture is going to creep up on subtractive. I have a rep rap and it's not so useful, mostly because the i2 design was lousy and the mechatronics are a bit too clockwork. Related - reverse kinetics aren't simple. So what... GE makes turbine blades that are impossible to make via milling or casting. They're top down and enthusiast amateurs do the bottom up. It'll change how we make finished goods in a profound way, just not right away.
Your last point about the Marine requirements is spot on.
They should have designed an Air Force / Navy only version and gotten that fully operational: software bugs worked out, airframe defects fixed, flight/battle tested etc. The costs of the program would be much lower, so more planes, fewer design compromises, more buyers. By that point (which is several years from now) typically there are airframe optimizations and engine upgrades ready that were not available in the initial design. Take those improvements, apply to a dual seat trainer airframe (so you have enough space) and make your STOVL Marine variant. If weight is still a problem consider dropping stealth. Anti SAM and air superiority are a job for the Air Force.
Instead the tail wagged the dog with Marine variant and now we have a big ol' expensive mess.
Re: CAS - I think it is likely to become a drone mission anyhow. Better to have ordinance on a cheap airframe with no pilot. It doesn't matter if it gets shot down and you can have a larger number of cheaper planes. Software flies to and from the battlefield so fewer remote pilots needed - done.
You wouldn't typically use a single neural network from input (LIDAR, video, gyros, accelerometers, etc) to output (steering and pedals)
More typically you'd use different neural networks to tackle steps in the chain. One might identify the borders of the road and the median. Another might pick out cars. Another might project the position of the cars in the future. Each would be easier to test individually. You might also have a "supervisor" that looked for disagreement or inconsistencies between the systems and notify the driver to take over.
Not even close in the US. UK and Europe count the cycles and try to average 60 Hz over minutes/hours/day with minor adjustments. The US has massive drift and variability.
Criminal investigations is a niche use. The broad use and real revenue would be an adult picture of your unborn child or new baby. Future parents already pay $300-400 for 3d sonograms of their fetus. Imagine seeing your new baby's face each year from two to twenty years old.
You could even sequence a couple individually and show the full range of boy and girl facial outcomes with probabilities. Right now they can use some morphing techniques as a kludge but genetics could be MUCH more predictive.
Creepy, creepy, creepy...
I imagine it would require connecting too much non-x86 R-Pi plumbing but it would be nice if it could run the PC version of Netflix.
I'm on Kubuntu 15.04, NUC with intel graphics. Everything just works including suspend right out of the gate. Love it.
This. RPI2 is a OUYA killer... huge installed base, same access to emulators, can use stock wireless console controllers like PS3 with a ~$5 bluetooth dongle.
It gets good reviews on the Kodi forums. It's Kodi, Netlfix and Amazon video in one small box. The CPU is powerful so theoretically a decent gaming platform though I can't speak to that. I went rpi-2 for Kodi so no Netflix but fortunately my TV and/or tablets do.
Ouya has loads of competition now from ARM "sticks" and media adapters like the Fire, Roku or Cu Box. And each year brings more capable hardware while Ouja stays the same. The new raspberry pi 2 or Amazon Fire are arguably superior in all ways. Certainly both those alternatives make excellent XBMC/Kodi boxes.
And competition has also come from tablets in terms of casual gaming. Tablets benefit from huge economies of scale and large online market ecosystems. Ouja was always going to be a niche market appealing to techies and gamers.
Can run x86 binaries and uses Intel HD graphics, so no gfx driver issues. Roku or other ARM sticks can't run x86 binaries and you have to deal with their binary blob gfx drivers that generally don't play well with new kernels. ARM binaries are vary by core vintage and soft vs hard float. Then there is the variety of gpu cores: MALI, Vivante, PowerVR, Broadcom. By contrast Intel has open source options and Intel supplied binaries that get updated, especially because the desktop chips use the same gpu.
I tried upgrading my headless and slow single core ARM server to an IMX6 quad core. In theory it would have been great - in reality it never worked right and the gfx issues were a nightmare to sort out. I ended up with a J1900 board for not much more and it works like a dream.
Companies have gotten really good a simplifying human jobs so that new hires with few skills can be quickly trained up to replace underperforming or otherwise problematic workers. There are high paying management jobs (a few of them) for producing and optimizing employment manuals, procedures, performance targets and input kiosks so that the absolute lowest common denominator hire can quickly fill a void.
As an example McDonalds "upgraded" their order taking turrets from using words for each food item to pictures for each food item. That meant they could employ people who couldn't read, because I guess literacy was a limiting requirement in their hiring process. McDonalds employs over 400,000 people. Just a small "savings" across that employment base is worth millions. That millions of savings get's split between shareholders and the top tier of management who designs and implements these "process enhancements".
And the new thing is to order online from your smartphone and pick it up at the counter. That gets rid of the order taker entirely and you can staff with mostly "behind the scenes" worker bees that don't even have to speak English. That is until you can get a robot to make the food too.
Call centers have been doing this for years with average call time metrics, flow charts for addressing caller needs, etc... It's happening in lots and lots of industries now.
LEGO molds are some of the best in general manufacturing. If a 3d printer could make a LEGO brick as good as the original it would imply that you could print virtually any plastic part to sufficient tolerance.