The human ear has a 120 dB dynamic range. While we're not all listening in anechoic chambers, studies show narrowband signals are detectable as much as 20 dB below a broadband noise floor. Now, very few amps achieve distortion and SNR that can cover this range (indeed, I'm not aware of any commercial ones — only a DAC that exceeds that range, and a few circuits prototyped by a few crazed members at the diyaudio forum). This means that the suggestion that solid state and tube amps that are properly designed sound the same is incorrect. There is no "linear range"; only an approximately linear one, which is still not completely linear with respect to psychoacoustics. I think what you meant to say is that well designed solid state and tube amps sound practically the same to most people in typical listening conditions. Please see my comment here as well: http://entertainment.slashdot....
>"And don't even get me started on the tube mythologies."
What do you mean? In the end, a tube is just a gain device, like a transistor. Given a distortion spec, one can just as much build a tube amp to match it as a solid state one. There are both SS and tube audio amps that achieve distortion levels in the part-per-million level. Multiple gain stages are required in either case to get very low distortion; the real fault in most consumer tube amps is not the use of tubes, but the use of circuits that are too simple — that's the audiophile fetishist fault. Indeed, a single (constant current source loaded) vacuum triode is more linear a voltage gain device than any single transistor. If you use as many tubes as transistors, you can easily match the low distortion levels of the SS design. Tubes have specific benefits including removing thermal memory distortion (modulation of gain device parameters by the temperature changes caused by varying power dissipation). See for example this AES paper:http://www.aes.org/e-lib/brows... With transistor designs, to deal with the issue you need to add more devices to even out the power dissipation at least in the differential pair input and the VAS, and in the case of chip power amps, add compensation for the effect of the output stage thermal dissipation affecting the previous stages. Then there's the issue that transistor gain curve is exponential whereas the tube's is power-of-two, which makes the distortion profiles of a tube and an SS amp that achieve the same THD quite different, with more of the THD in the tube case caused by lower order and even harmonics — the very ones that the human auditory system masks anyway (psychoacoustics what ultimately matters, and there is interesting research and AES papers on more relevant metrics than THD/IM). Tube's problem is simply one of practicality in regards to their size and the need for filament power. Other issues can all be dealt with. For example, in terms of typical speakers, the low impedance has been traditionally solved with transformers, which introduces phase nonlinearities and some hysteresis effects, so they add distortion. But this is unnecessary. One elegant solution is the replacement of the output transformer with a switching impedance converter that operates far above the audio band; see D. Berning's patent (I think it expires soon). While the converter is an active SS state, it has no gain and no distortion in the audio band. Another solution is to directly couple tube output to electrostatic speakers, which have very high impedance. A third solution is to use hybrid circuits with both SS and tube stages. It's possible to get the best of both worlds there. Here's a great hybrid circuit that achieves a few ppm THD for 1500 kV p-p output for electrostatic headphones:http://headwize.com/?page_id=7... Note especially the hybrid third-fourth gain stage. One reason the amp gets such low distortion with only moderate NFB is that the third stage transistor's nonlinearity, in the operating range, is roughly inverse to the final tube stage's nonlinearity.
It will be for the worse. While the level of resistance of a system of inequality (hierarchical civilization) to revolutions that remake the social order has not been increasing monotonically throughout history, on average it has been increasing. In recent times, sufficient corrective feedback mechanisms have been integrated into human society that, in my opinion, successful revolution is impossible. Save for a global catastrophe (whether manmade or natural) that decimates the population, the current trends will continue. The endgame I would bet on is that advanced robotics will make poor people obsolete; what happens to them at that point is going to be something akin to what Marshall Brain wrote about in his story "Manna", minus the happy ending. So, you can't beat them, and chances are against you joining them. Upward mobility is mostly a game of luck no matter how skilled, dedicated, and intelligent you are. However, playing this lottery is the only hope, small as it is, that you, as a representative of the non-elite, have to create a decent life for yourself and your offspring.
The surface of the collapsing star takes an infinite time to cross the event horizon form the point of view of an outside observer? No star which has collapsed has yet turned into a black hole, and no one will at a finite age of the outside universe. The only way to prove the existence of a black hole is to fall through an event horizon. Of course, then you only prove it for yourself, and cannot tell anyone else.
Is that clip supposed to be funny, Chris? My gosh, whatever happened to British humor? From the heights of Monty Python and Benny Hill, it's been a precipitous fall to this.
Cursive exists because it's faster. This is why the letters are joined; it's not for looks. If it was about the latter, they would still be teaching Spencerian script in North America and similar systems elsewhere. There are other benefits: http://www.nytimes.com/roomfor...
Renewables kill more people per amount of energy generated than nuclear: http://nextbigfuture.com/2011/... You'd have to be a misanthrope to invest in that over nuclear.
For all the talk of the dangers of nuclear, it has still caused less deaths per amount of energy generated than any other method that has been used to practically generate electricity: http://nextbigfuture.com/2011/... If you're not ignorant of these facts, then the only remaining reasons to oppose nuclear are either political (Naomi Klein-style anti-capitalist), or you're simply a misanthrope.
The whole issue of waste has been beaten to death. Reprocessing and breeder reactors leave only a little waste that can't be used for energy, and waste transmutation is a proven concept that further reduces any dangerous waste. With these processes, the actual nuclear waste left over is a tiny amount, and glassification trivially takes care of that.
Nuclear kills the least number of humans per amount of energy generated compared to any other energy source used for power generation, including wind/solar/hydro: http://nextbigfuture.com/2011/...
I don't disagree that the body provided inputs can be simulated, but that is non-trivial because the brain-body system forms a very complicated set of feedback loops. My point is not that human-like AI is unachievable, but that most here are underestimating what, and how long, it will take. Regarding your question as to the minimum feedback needed, Damasio goes to some extent to address this; really, look up his latest book in the library (it helps that he's a great writer and it's easy to read). As for making intelligence that is non-human like so you can avoid having to deal with the embodied cognition issue, I discuss this in my post here: http://slashdot.org/comments.p...
Indeed. It should be required reading for AI researchers as well, and I say this as someone with a graduate degree in comp.sci. and more than a passing professional interest in AI.
What could be more dangerous than building AI that's smarter than us but cannot relate to us because its intelligence is drastically different from ours? There's no fool-proof way to actually implement "software" constraints in a general super-human intelligence AI (Asimov's robotics laws are about the most unrealistic thing I've ever read in sci fi), and the safety factor falls even further over generations as you get the AI to design an even smarter AI. Physical constraints? Do you think that when an ultra-intelligent AIs are available, businesses won't connect them to as many control systems as possible to profit from improved efficiencies and replaced labor? And what is more inefficient than some sort of "air gap"?
The best bet is to implement AI that can understand us, and since human cognition is completely intertwined with feelings/emotions at a basic neurological level (see somatic markers, etc.) this requires making human-like AI that can feel the way we feel, so that it can have consciousness that is sufficiently similar to ours. This is a very hard problem and requires simulating a human-like brain _and_ also the body that goes with it (see embodied cognition on wiki). If we build powerful AIs that don't love us, then humanity will be doomed.
This is far more than a philosophical thesis; it's backed up by neuroscience. I highly suggest you read Damasio, who's one of the top neuroscientists in the world. A good overview can be found in his book Self Comes to Mind, the price of it being justified by the selection of paper references in the endnotes alone.
Your simulation is of purely academic interest if it relies the usual gross oversimplification of the activity of a real neuron. It's only two years ago that we've even attempted a simulation of 100 trillion synapses (comparable to a human brain), in a joint IBM and Lawrence Livermore National Laboratory project. That simulation ran on what was in 2012 the top supercomputer in the world, yet the simulation still ran over 1500 times slower than real-time, and, worse, was still using quite simplified neuron models! Having a brain-equivalent information processor that fits in the space of a skull and runs on the brain's approximately 20 W? It won't happen in your life time if you're old enough to be posting on this site.
By the Bekenstein bound (http://en.wikipedia.org/wiki/Bekenstein_bound), human brains can contain only a finite number of distinguishable quantum states. This means that the brain's information processing has the same limits as a nondeterministic linear bounded automaton, which is more restricted than a Turing machine. We can't think anything more than a sufficiently sophisticated computational construction can. The same limits ultimately apply to us that the authors ascribe to their killer robot targets.
Drivers are giving up. Almost every other major North American city has freeways running through or near the city core. Those are part of the infrastructure that invites city centres to become focal points of business, commerce, and finance. Vancouver's residential core cements the city's status as a place where wealth made elsewhere is spent; none is actually generated here. I can only pray that the feds force through the pipeline so that the port can expand, and offset some of this failure. Even much of the high tech industry ran off to Quebec and Ontario, despite high hipster quotient.
And when he gets discouraged by how our buses are never on time, or how the Skytrain (Vancouver's light rail) has its tracks under maintenance far too often during active hours, maybe he can instead make use of recently-reelected-Mayor-Moonbeam's pet project bicycle lanes — I mean, he'd be meaningfully increasing bike ridership statistics and maybe I'll feel a teensy bit better about the traffic disruption the bike lanes have caused downtown and in other communities in a city that is becoming ever more _not_ a commerce, technology, or industrial centre, but a residential playground for the wealthiest in the international real estate market.
Not simply because his post is wrong (that is not considered grounds for moderation by itself), but because it's willful ignorance and grossly insulting to readers. His post is, in essence,
"The article contains this trivially verifiable statement of fact, but I _feel_ it doesn't sound right to me, so I'm compelled to rant about it on Slashdot, while at the same time being far too lazy to spend ten seconds to check it with a search engine — but then again, I just don't respect the audience enough to care if I post total nonsense; all that matters is that I get to express my feelings, and who the hell are Slashdot readers to tell me my feelings are wrong anyway?!"
Slashdot Mirror
The human ear has a 120 dB dynamic range. While we're not all listening in anechoic chambers, studies show narrowband signals are detectable as much as 20 dB below a broadband noise floor. Now, very few amps achieve distortion and SNR that can cover this range (indeed, I'm not aware of any commercial ones — only a DAC that exceeds that range, and a few circuits prototyped by a few crazed members at the diyaudio forum). This means that the suggestion that solid state and tube amps that are properly designed sound the same is incorrect. There is no "linear range"; only an approximately linear one, which is still not completely linear with respect to psychoacoustics. I think what you meant to say is that well designed solid state and tube amps sound practically the same to most people in typical listening conditions. Please see my comment here as well: http://entertainment.slashdot....
>"And don't even get me started on the tube mythologies."
What do you mean? In the end, a tube is just a gain device, like a transistor. Given a distortion spec, one can just as much build a tube amp to match it as a solid state one. There are both SS and tube audio amps that achieve distortion levels in the part-per-million level. Multiple gain stages are required in either case to get very low distortion; the real fault in most consumer tube amps is not the use of tubes, but the use of circuits that are too simple — that's the audiophile fetishist fault. Indeed, a single (constant current source loaded) vacuum triode is more linear a voltage gain device than any single transistor. If you use as many tubes as transistors, you can easily match the low distortion levels of the SS design. Tubes have specific benefits including removing thermal memory distortion (modulation of gain device parameters by the temperature changes caused by varying power dissipation). See for example this AES paper:http://www.aes.org/e-lib/brows... With transistor designs, to deal with the issue you need to add more devices to even out the power dissipation at least in the differential pair input and the VAS, and in the case of chip power amps, add compensation for the effect of the output stage thermal dissipation affecting the previous stages. Then there's the issue that transistor gain curve is exponential whereas the tube's is power-of-two, which makes the distortion profiles of a tube and an SS amp that achieve the same THD quite different, with more of the THD in the tube case caused by lower order and even harmonics — the very ones that the human auditory system masks anyway (psychoacoustics what ultimately matters, and there is interesting research and AES papers on more relevant metrics than THD/IM). Tube's problem is simply one of practicality in regards to their size and the need for filament power. Other issues can all be dealt with. For example, in terms of typical speakers, the low impedance has been traditionally solved with transformers, which introduces phase nonlinearities and some hysteresis effects, so they add distortion. But this is unnecessary. One elegant solution is the replacement of the output transformer with a switching impedance converter that operates far above the audio band; see D. Berning's patent (I think it expires soon). While the converter is an active SS state, it has no gain and no distortion in the audio band. Another solution is to directly couple tube output to electrostatic speakers, which have very high impedance. A third solution is to use hybrid circuits with both SS and tube stages. It's possible to get the best of both worlds there. Here's a great hybrid circuit that achieves a few ppm THD for 1500 kV p-p output for electrostatic headphones:http://headwize.com/?page_id=7... Note especially the hybrid third-fourth gain stage. One reason the amp gets such low distortion with only moderate NFB is that the third stage transistor's nonlinearity, in the operating range, is roughly inverse to the final tube stage's nonlinearity.
It will be for the worse. While the level of resistance of a system of inequality (hierarchical civilization) to revolutions that remake the social order has not been increasing monotonically throughout history, on average it has been increasing. In recent times, sufficient corrective feedback mechanisms have been integrated into human society that, in my opinion, successful revolution is impossible. Save for a global catastrophe (whether manmade or natural) that decimates the population, the current trends will continue. The endgame I would bet on is that advanced robotics will make poor people obsolete; what happens to them at that point is going to be something akin to what Marshall Brain wrote about in his story "Manna", minus the happy ending. So, you can't beat them, and chances are against you joining them. Upward mobility is mostly a game of luck no matter how skilled, dedicated, and intelligent you are. However, playing this lottery is the only hope, small as it is, that you, as a representative of the non-elite, have to create a decent life for yourself and your offspring.
The surface of the collapsing star takes an infinite time to cross the event horizon form the point of view of an outside observer? No star which has collapsed has yet turned into a black hole, and no one will at a finite age of the outside universe. The only way to prove the existence of a black hole is to fall through an event horizon. Of course, then you only prove it for yourself, and cannot tell anyone else.
Is that clip supposed to be funny, Chris? My gosh, whatever happened to British humor? From the heights of Monty Python and Benny Hill, it's been a precipitous fall to this.
Cursive exists because it's faster. This is why the letters are joined; it's not for looks. If it was about the latter, they would still be teaching Spencerian script in North America and similar systems elsewhere. There are other benefits: http://www.nytimes.com/roomfor...
Misanthrope much?
"to breathe"
FTFY
Renewables kill more people per amount of energy generated than nuclear: http://nextbigfuture.com/2011/... You'd have to be a misanthrope to invest in that over nuclear.
BS. All human nuclear activity and accidents have in the end not raised background radiation levels appreciably above what it naturally is.
For all the talk of the dangers of nuclear, it has still caused less deaths per amount of energy generated than any other method that has been used to practically generate electricity: http://nextbigfuture.com/2011/... If you're not ignorant of these facts, then the only remaining reasons to oppose nuclear are either political (Naomi Klein-style anti-capitalist), or you're simply a misanthrope.
The whole issue of waste has been beaten to death. Reprocessing and breeder reactors leave only a little waste that can't be used for energy, and waste transmutation is a proven concept that further reduces any dangerous waste. With these processes, the actual nuclear waste left over is a tiny amount, and glassification trivially takes care of that.
Nuclear kills the least number of humans per amount of energy generated compared to any other energy source used for power generation, including wind/solar/hydro: http://nextbigfuture.com/2011/...
Hydro kills far more people per amount of energy generated than nuclear: http://nextbigfuture.com/2011/...
Solar kills a lot more people per amount of energy generated than nuclear does: http://nextbigfuture.com/2011/...
I don't disagree that the body provided inputs can be simulated, but that is non-trivial because the brain-body system forms a very complicated set of feedback loops. My point is not that human-like AI is unachievable, but that most here are underestimating what, and how long, it will take. Regarding your question as to the minimum feedback needed, Damasio goes to some extent to address this; really, look up his latest book in the library (it helps that he's a great writer and it's easy to read). As for making intelligence that is non-human like so you can avoid having to deal with the embodied cognition issue, I discuss this in my post here: http://slashdot.org/comments.p...
I think it's silly to suppose there will be humans left 500 years from now, let at the 1K bottom rung of your ladder.
Indeed. It should be required reading for AI researchers as well, and I say this as someone with a graduate degree in comp.sci. and more than a passing professional interest in AI.
What could be more dangerous than building AI that's smarter than us but cannot relate to us because its intelligence is drastically different from ours? There's no fool-proof way to actually implement "software" constraints in a general super-human intelligence AI (Asimov's robotics laws are about the most unrealistic thing I've ever read in sci fi), and the safety factor falls even further over generations as you get the AI to design an even smarter AI. Physical constraints? Do you think that when an ultra-intelligent AIs are available, businesses won't connect them to as many control systems as possible to profit from improved efficiencies and replaced labor? And what is more inefficient than some sort of "air gap"?
The best bet is to implement AI that can understand us, and since human cognition is completely intertwined with feelings/emotions at a basic neurological level (see somatic markers, etc.) this requires making human-like AI that can feel the way we feel, so that it can have consciousness that is sufficiently similar to ours. This is a very hard problem and requires simulating a human-like brain _and_ also the body that goes with it (see embodied cognition on wiki). If we build powerful AIs that don't love us, then humanity will be doomed.
300 synapses? Humans have at least 100 trillion.
But our bodies are one of the most essential determinants of the nature of human consciousness: http://en.wikipedia.org/wiki/E...
This is far more than a philosophical thesis; it's backed up by neuroscience. I highly suggest you read Damasio, who's one of the top neuroscientists in the world. A good overview can be found in his book Self Comes to Mind, the price of it being justified by the selection of paper references in the endnotes alone.
Your simulation is of purely academic interest if it relies the usual gross oversimplification of the activity of a real neuron. It's only two years ago that we've even attempted a simulation of 100 trillion synapses (comparable to a human brain), in a joint IBM and Lawrence Livermore National Laboratory project. That simulation ran on what was in 2012 the top supercomputer in the world, yet the simulation still ran over 1500 times slower than real-time, and, worse, was still using quite simplified neuron models! Having a brain-equivalent information processor that fits in the space of a skull and runs on the brain's approximately 20 W? It won't happen in your life time if you're old enough to be posting on this site.
By the Bekenstein bound (http://en.wikipedia.org/wiki/Bekenstein_bound), human brains can contain only a finite number of distinguishable quantum states. This means that the brain's information processing has the same limits as a nondeterministic linear bounded automaton, which is more restricted than a Turing machine. We can't think anything more than a sufficiently sophisticated computational construction can. The same limits ultimately apply to us that the authors ascribe to their killer robot targets.
Drivers are giving up. Almost every other major North American city has freeways running through or near the city core. Those are part of the infrastructure that invites city centres to become focal points of business, commerce, and finance. Vancouver's residential core cements the city's status as a place where wealth made elsewhere is spent; none is actually generated here. I can only pray that the feds force through the pipeline so that the port can expand, and offset some of this failure. Even much of the high tech industry ran off to Quebec and Ontario, despite high hipster quotient.
And when he gets discouraged by how our buses are never on time, or how the Skytrain (Vancouver's light rail) has its tracks under maintenance far too often during active hours, maybe he can instead make use of recently-reelected-Mayor-Moonbeam's pet project bicycle lanes — I mean, he'd be meaningfully increasing bike ridership statistics and maybe I'll feel a teensy bit better about the traffic disruption the bike lanes have caused downtown and in other communities in a city that is becoming ever more _not_ a commerce, technology, or industrial centre, but a residential playground for the wealthiest in the international real estate market.
Not simply because his post is wrong (that is not considered grounds for moderation by itself), but because it's willful ignorance and grossly insulting to readers. His post is, in essence,
"The article contains this trivially verifiable statement of fact, but I _feel_ it doesn't sound right to me, so I'm compelled to rant about it on Slashdot, while at the same time being far too lazy to spend ten seconds to check it with a search engine — but then again, I just don't respect the audience enough to care if I post total nonsense; all that matters is that I get to express my feelings, and who the hell are Slashdot readers to tell me my feelings are wrong anyway?!"