Eventually the message body won't count either. C'mon guys, give it up. It's 2016. Can we afford maybe 300 characters, when the payload of a tweet's metadata is already several kilobytes?
I respect geohotz, but I don't think a 26-year old kid completely grasps the ramifications of building and shipping a technology whose goal is arguably to "kill fewer people". Even if he does kill fewer people, people are going to die using this or any autonomous driving platform. Is he ready to deal with the fallout from that?
Also, we won't know until millions or hundreds of millions of miles have been driven how the technology compares in terms of safety to human drivers. It sounds like he's using low-end electronics, so chances are he's just banking on deep learning solving all the driving problems. This is a very, very dangerous approach to take, when deep learning has a strong tendency to exhibit pathological failure cases (see "adversarial nets"), and when it's usually impossible to fully understand, explain, justify or characterize the strengths and weaknesses of a network trained on a given task.
he was mystified by the reluctance of some portfolio companies to avoid a stock market flotation
Maybe some people have wised up to the fact that market valuations are completely artificial -- they are numbers picked out of the air, and being listed on the stock market places you at the mercy of mass psychology, media spin, gross subjectivity, market volatility, and trading algorithms that solely exist to milk profits from the market.
This probably happens all the time in real banks, given how antiquated their IT systems are. You just don't hear about it, because the bank doesn't want to undermine your confidence, and can ask the Federal Reserve to bail them out. Not so with Bitcoin.
That's because there's no such thing as dark matter or dark energy. In 50 years, we'll look back at this theory (and "junk DNA") as a quaint historical silliness, the way we look at "the ether/aether" today.
This is presumably to create a shield of plausible deniability for priests that want to download and view illegal content. "It must have been some random member of the public sitting outside using their laptop in their car!".
Yes, of course it does, the C++ API is the one that Python uses. But there is a lot of functionality in Python that is not available in the C++ API, because a lot of rich features were quickly added in Python, rather than taking the time to develop them in C++: the Python API is not just a thin layer over the C++ API.
A lot of the TensorFlow functionality is written in Python right now, for whatever bizarre reason (many TensorFlow pipelines take a huge performance hit by dropping out to Python after every training batch, in order to feed the next batch from a Python data structure). The TensorFlow team eventually plans to push more of the Python functionality down into C++-land, so you can build bindings for other languages (e.g. Swift/Obj-C), but this isn't currently possible. Since Python doesn't run on iOS (modulo a few hacky solutions like tinypy), I'd say the TF team has a lot of work to do before TF on iOS is a possibility. (Note -- this is for building arbitrary TF models in iOS, which would currently require Python, as opposed to executing runs through already-trained models, which can be done more simply by means of a TF graph serializer and de-serializer, which doesn't necessarily require Python, and already exists in some form for Android and other runtimes.)
There's a kill switch for humans too, but murder is considered unethical. You either believe in the eventual moral equivalence of the intelligence (consciousness, being,...) of humans and AI, or you don't. If you do believe in it, a kill switch is not ethical. If you don't believe in it, you have no reason to want to install a kill switch in the first place, because AI will never transcend levels of human intelligence.
This doesn't fix the numbers issue for practical driving speeds. Assuming a fixed number of lanes, you'll see the same number of vehicles if you're parked on the median as if you're moving at the same speed as traffic on one side of the road. In the latter case, assuming traffic is moving the same speed in both directions, and you are too, you see zero new license plates on your side of the road as a function of time, while traffic on the other side is approaching at 2x the relative speed. Of course, once you start speeding much faster than the traffic on either side of the road, you can start increasing the rate at which you see license plates without bound, assuming you can drive at extremely high speeds, and the camera can handle the motion without blurring too much. (But this is not practical.)
This scanner can scan thousands of license plates per minute?? Let's do the math.
The vehicle has cameras on both sides. Assuming each camera can capture plates from up to 3 lanes of traffic, to achieve "thousands" of scans per minute, conservatively interpreted as at least 2000 scans per minute, each camera would have to pull in 1000 scans per minute, or 333 scans per minute per lane. This translates to a little over 5.5 scans per second per lane, or 0.2 seconds per scan per lane. This is impossible with the recommended 2 second minimum following distance between cars, regardless of the speed the cars are traveling -- in fact, the scan rate is 10x larger than the safe carrying capacity of 3 lanes on each side of the car.. Therefore, to scan "thousands" of plates per minute, this vehicle would have to be parked in the middle of a road 10x as wide, for roughly a total of 60 lanes.
The only alternative to this would be to scan cars parked close together on both sides as the scan-van travels really, really fast up the middle. You'd have to pass 5.55 cars per second on each side. Assuming the cars are parked 5.5 meters apart, you have to travel 70mph past the line of parked cars to hit this rate, which would be not only illegal in a zone lined on both sides with parked cars, but it would also be dangerous. Maybe that's where they get the number from though? (Also, this is probably not workable due to motion blur at those speeds...)
"Swarm Intelligence" allows groups to amplify their collective IQ beyond the capacity of individuals, something that the human species hasn't been able to do because of evolutionary restraints.
Because our brains is not an immense colony of cooperating neurons? Each cell is an organism. Humans are a superorganism.
Slashdot Mirror
Apparently to code with kids these days you need to deserve the title of "ninja". So I recommend ninjitsu training.
Finally, an exception to Betteridge's Law of Headlines.
Hot off the press: iPhone 8 will come with a rooster weather vane permanently attached to the top.
Eventually the message body won't count either. C'mon guys, give it up. It's 2016. Can we afford maybe 300 characters, when the payload of a tweet's metadata is already several kilobytes?
This will of course be shut down now. He chose fame over having unlimited free data forever.
I respect geohotz, but I don't think a 26-year old kid completely grasps the ramifications of building and shipping a technology whose goal is arguably to "kill fewer people". Even if he does kill fewer people, people are going to die using this or any autonomous driving platform. Is he ready to deal with the fallout from that? Also, we won't know until millions or hundreds of millions of miles have been driven how the technology compares in terms of safety to human drivers. It sounds like he's using low-end electronics, so chances are he's just banking on deep learning solving all the driving problems. This is a very, very dangerous approach to take, when deep learning has a strong tendency to exhibit pathological failure cases (see "adversarial nets"), and when it's usually impossible to fully understand, explain, justify or characterize the strengths and weaknesses of a network trained on a given task.
he was mystified by the reluctance of some portfolio companies to avoid a stock market flotation
Maybe some people have wised up to the fact that market valuations are completely artificial -- they are numbers picked out of the air, and being listed on the stock market places you at the mercy of mass psychology, media spin, gross subjectivity, market volatility, and trading algorithms that solely exist to milk profits from the market.
This probably happens all the time in real banks, given how antiquated their IT systems are. You just don't hear about it, because the bank doesn't want to undermine your confidence, and can ask the Federal Reserve to bail them out. Not so with Bitcoin.
BellingWhere?
Good point :-D
That's because there's no such thing as dark matter or dark energy. In 50 years, we'll look back at this theory (and "junk DNA") as a quaint historical silliness, the way we look at "the ether/aether" today.
See also: The Gartner Hype Cycle
This is presumably to create a shield of plausible deniability for priests that want to download and view illegal content. "It must have been some random member of the public sitting outside using their laptop in their car!".
Yes, of course it does, the C++ API is the one that Python uses. But there is a lot of functionality in Python that is not available in the C++ API, because a lot of rich features were quickly added in Python, rather than taking the time to develop them in C++: the Python API is not just a thin layer over the C++ API.
A lot of the TensorFlow functionality is written in Python right now, for whatever bizarre reason (many TensorFlow pipelines take a huge performance hit by dropping out to Python after every training batch, in order to feed the next batch from a Python data structure). The TensorFlow team eventually plans to push more of the Python functionality down into C++-land, so you can build bindings for other languages (e.g. Swift/Obj-C), but this isn't currently possible. Since Python doesn't run on iOS (modulo a few hacky solutions like tinypy), I'd say the TF team has a lot of work to do before TF on iOS is a possibility. (Note -- this is for building arbitrary TF models in iOS, which would currently require Python, as opposed to executing runs through already-trained models, which can be done more simply by means of a TF graph serializer and de-serializer, which doesn't necessarily require Python, and already exists in some form for Android and other runtimes.)
See also the Gartner Hype Cycle -- "the peak of inflated expectations", right before "the trough of disillusionment".
There's a kill switch for humans too, but murder is considered unethical. You either believe in the eventual moral equivalence of the intelligence (consciousness, being, ...) of humans and AI, or you don't. If you do believe in it, a kill switch is not ethical. If you don't believe in it, you have no reason to want to install a kill switch in the first place, because AI will never transcend levels of human intelligence.
Relevant bug
This doesn't fix the numbers issue for practical driving speeds. Assuming a fixed number of lanes, you'll see the same number of vehicles if you're parked on the median as if you're moving at the same speed as traffic on one side of the road. In the latter case, assuming traffic is moving the same speed in both directions, and you are too, you see zero new license plates on your side of the road as a function of time, while traffic on the other side is approaching at 2x the relative speed. Of course, once you start speeding much faster than the traffic on either side of the road, you can start increasing the rate at which you see license plates without bound, assuming you can drive at extremely high speeds, and the camera can handle the motion without blurring too much. (But this is not practical.)
This scanner can scan thousands of license plates per minute?? Let's do the math.
The vehicle has cameras on both sides. Assuming each camera can capture plates from up to 3 lanes of traffic, to achieve "thousands" of scans per minute, conservatively interpreted as at least 2000 scans per minute, each camera would have to pull in 1000 scans per minute, or 333 scans per minute per lane. This translates to a little over 5.5 scans per second per lane, or 0.2 seconds per scan per lane. This is impossible with the recommended 2 second minimum following distance between cars, regardless of the speed the cars are traveling -- in fact, the scan rate is 10x larger than the safe carrying capacity of 3 lanes on each side of the car.. Therefore, to scan "thousands" of plates per minute, this vehicle would have to be parked in the middle of a road 10x as wide, for roughly a total of 60 lanes.
The only alternative to this would be to scan cars parked close together on both sides as the scan-van travels really, really fast up the middle. You'd have to pass 5.55 cars per second on each side. Assuming the cars are parked 5.5 meters apart, you have to travel 70mph past the line of parked cars to hit this rate, which would be not only illegal in a zone lined on both sides with parked cars, but it would also be dangerous. Maybe that's where they get the number from though? (Also, this is probably not workable due to motion blur at those speeds...)
A summon mode that drives into the back of a truck doesn't seem like a feature I'd like installed in my next car.
...just to be confusing.
"Swarm Intelligence" allows groups to amplify their collective IQ beyond the capacity of individuals, something that the human species hasn't been able to do because of evolutionary restraints.
Because our brains is not an immense colony of cooperating neurons? Each cell is an organism. Humans are a superorganism.
"Hey Viv, when does 'its' take an apostrophe?"
Did you just say "Buzzfeed" and "reports" in the same sentence?