Replying to your question about what's wrong with X11, I can say that one thing that puzzles me is how a protocol designed form the ground up to allow you to open windows transparently over a network connection can be *so* *much* *slower* than AJAX, which is an unholy (but impressive) melange of techs never designed to do remote apps.
I know that X11 does a lot more than AJAX, but still, couldn't anyone figure out how to split up the client/server workload so that I don't get a multi-second delay rendering every drop-down menu when I work across my 600+ kb/s connection?
What Google cares about is that there is a least one standards-compliant browser out there with fast javascript. Sure Google might have a slight preference for people using Chrome over another browser with fast javascript (like, say, Safari), but what really matters to them is that they are able to deliver web apps that are fast enough to be reasonable competitors to traditional desktop apps.
Chrome is a combination insurance policy/open-source soapbox whose purpose is to make sure that Google apps (and other web apps) will always have a browser to run on.
I'd like to respectfully correct a very common and understandable error in your terminology. I think you mean "topography" when you talk about the peaks and troughs of the ocean floor. "Topology" is a mathematical term describing the connectivity of sets of points: for example the surface of s sphere has one kind of topology while the surface of a donut has a different kind, because continuous transformations that don't break the 2D surface of a sphere can't morph it into a shape with a hole in it. See http://en.wikipedia.org/wiki/Topology for more if you're interested.
That aside, your point is well-taken that subs might tend to congregate in the same areas due to favorable underwater geological features.
First things first: I'm not an alarmist, and I don't think the LHC will blow up the Earth.
That said, I'd like to point out that not nearly all of that 10^20 eV is available to make new particles/black holes in the center of mass frame of the collision. Since all the collision products will have to have a ton of momentum in the direction that the cosmic ray was originally traveling, the available energy for creating new, potentially dangerous particles scales with the square root of the product of the energies (see http://www-bd.fnal.gov/public/relativity.html for a pretty good explanation of where this square root dependency comes from).
In contrast, the LHC will collide two particles in the TeV range head-on, which means the collisions have more of a chance of creating an "exotic" than even a 10^20 eV particle hitting stationary atmosphere.
However, I bet two high-energy cosmic rays each with energy > 10^14eV sometimes collide with *each other*, and that collision would have more available energy than the LHC collisions. The big question is how often does this happen? If collisions like these happen at a slow enough rate, I could imagine that the LHC might put Earth into unexplored territory in terms of numbers of collisions with ~10^14eV of available (i.e. not constrained to producing products with high momentum) energy.
I trust that the physicists have worked out the rates of these head-on, two-cosmic-ray collisions. Otherwise they would have no right saying that cosmic ray history shows that the LHC will be safe. Still, the only defense based on cosmic rays I've heard has been talking about cosmic rays hitting atmosphere, which isn't valid. Does anyone have a good link to a website analyzing the frequency of head-on two-cosmic-ray collisions?
To boil a teacup's worth of water, therefore it takes ~80 kJ.
For this to be twice the energy consumed with one search, that's ~40 kJ per search.
If a search takes Google about 100 ms, that means Google would be using 400 kW while responding to your search. That feels like it's about 3 orders of magnitude too high. It's possible that the original researchers got Calories and kCal confused.
It's not a matter of composing email in 4 ms instead of 20 ms. It's that entirely new functions will drive adoption.
Examples? Apple just added face recognition to iPhoto and image stabilization to iMovie. Lots of folks are starting to do video editing. Dictation (useful when you have carpal tunnel) performance can be improved a bit with more clock cycles.
As long as engineers build faster chips, software and systems geeks will find a way to add new (useful or flashy) features that the public wants.
That's certainly the dominant position in neuroscience these days: that Nature never wastes neural hardware. I think that maybe Nature tries not to waste a combination of metabolic energy, learning time and neural hardware.
Suppose that to perform a task you need X well-trained neurons. Suppose, for a small cost to the animal, you could build 100 X neurons and have most of them silent in exchange for a lower total metabolic cost or in exchange for the ability to learn faster. (This supposition is something I'm working on: that neural hardware redundancy can buy you faster learning and metabolic efficiency, as well as robustness in the face of random cell death.) Are the 99 X neurons that almost never fire really "wasted" even though they're not used? I'd argue that they're the byproduct of an efficient process: one that's efficient at learning and at conserving energy as well as efficient enough with neural hardware.
In any case, it looks like brain regions that are not undergoing a whole lot of learning really do have only 1% of their neurons fire on a regular basis. This has been shown consistently in a bajillion systems using every technique available to modern experimenters; read (1, above) for an enormously exhaustive list.
There's no evidence to support the idea that most cortical neurons fire on a regular basis, which is your position and what most neuroscientists have been assuming all along. Although this seems to be an absolutely reasonable assumption, it's just not how things work in reality.
If a neuron has become essentially inactive, we can still poll it once every (say) 1000 simulation cycles, and simulate it with 1000 times the real chance that it will fire. If it starts firing in such a way that would increase its excitability (see (3)), we can crank up the duty cycle of the simulation for that neuron. The net effect of this trade-off trick might be to introduce noise in the extremely rare spikes coming from mostly inactive neurons (which form the vast majority of all neurons, see (1) above). In my opinion, this extra noise is a small price to pay in exchange for a factor of ~1000 speedup.
PS this trade-off is the project I should be working on now rather than checking out Slashdot. There's a good chance it won't pan out, since nobody's published about it yet.
(3) Bidirectional Modification of Presynaptic Neuronal Excitability Accompanying Spike Timing-Dependent Synaptic Plasticity. Li, Lu, Wu, Duan and Poo. Neuron, Vol. 41, 257-268, January 22, 2004.
You're right unless there's a computational way to take advantage of the fact that most neurons in cortex pretty much never fire (1), and that a small minority of synapses are responsible for nearly all of the excitation in a slab of cortical tissue (2). If not active == not important == not necessary to simulate with a 100% duty cycle (these are big "ifs"), then we could be literally about 3-5 orders of magnitude closer to being able to simulate whole brains than anyone realizes.
(1) How silent is the brain: is there a "dark matter" problem in neuroscience? Shy Shoham, Daniel H. O'Connor, Ronen Segev. J Comp Physiol A (2006)
(2) Highly Nonrandom Features of Synaptic Connectivity in Local Cortical Circuits. Sen Song, Per Jesper Sjostro, Markus Reigl, Sacha Nelson, Dmitri B. Chklovskii. PLOS biology March 2005
Except when fools call 7.04 "7.4," which sounds higher than "7.1," which is what the same fools call "7.10". We should either get rid of two months or adopt a base-12 numbering system (and maybe genetically alter humans to have 6 fingers per hand while we're at it...), all in the name of Ubuntu.
It's much easier to build entrancingly-good virtual worlds which push all the right psychological buttons (by creating the illusion of accomplishing what instincts tell you is good for survival) than to go out and conquer foreign worlds. Even if humans made interstellar space travel their top priority, it would be centuries before we could colonize other star systems. By then we'll probably be able to upload our brains into a virtual world much more rewarding than a remote, desolate rock orbiting around some distant star.
Added bonus: as long as Moore's Law holds, time inside a virtual world (measured in clock cycles, not seconds) will be much less limited than our sun's remaining couple billion years of prime fusion.
I'm a big fan of Firefox, and I use a few extensions, but could you please let me more about this awesome "insert" extension? Does it allow you to insert content into sites (à la Wikipedia), or is it pr0n-related?
Plus you have the potential of diluting the Apple brand name. What would an Apple be? If you have an Apple CPU do you now have an Apple?
Exactly. Apple isn't going to start selling non-Apple hardware. The only scenario where an Apple-AMD merger isn't retarded is if they're sure they'll be able to use all of AMD's chips and not need any extras. Can you imagine a CPU company buying another company's CPUs to build computers? Can you imagine Apple selling MS boxes? Unless you can answer "yes" to both, then if Apple's demand doesn't fit AMD's supply exactly the merged company would be worse off. That exact fit isn't the case, nor is it ever likely to be (both Apple's demand and AMD's supply will vary due to umpteen forces), so in short an Apple-AMD merger just ain't gonna happen.
Excellent point! I think the best way to handle this is to organize people so that if a neighbourhood wanted faster service, they could each make an agreement like "If somebody increases my bandwidth to X, I would pay them Y." When enough people make enough pledges, a small, private ISP company could lay the right tech and claim the bounty. In essence, hardware upgrades should be paid by those who demand it, and as long as the network is available to the public in general the public should provide the backbone bandwidth. Would that work?
So, what you really mean instead of "free" is "paid for through taxation".
The big question is "paid how much?" The three biggest expenses for big ISPs are:
"Last mile" connectivity
Advertising and promotion
Billing & accounting
With community- or government-provided networks, #2 and #3 go away completely, and most of #1 goes away too. Here's why.
Most of the expense of getting residents Internet connectivity comes from connecting up the last few miles to individuals' homes. ISPs could in principal do away with a wired "last mile" with the right wireless technology, but that would make it easy for people to share their connections and passwords with neighbours. Instead, ISPs are charging you a premium so they can keep you from sharing your Internet connection when they insist on using cable or DSL.
We're at the point now where the inefficiency inherent in having to advertise, charge for and segregate Internet service is greater than the inefficiency inherent in public sector projects, which means that financially it's better now to have government- and community-provided Internet connectivity.
Slashdot Mirror
Right on. If your metric is (military power)/(cost) then these planes would have to be *extremely* deadly for them to make more sense than drones.
Dude,
The GP was referring to the Berry paradox.
http://en.wikipedia.org/wiki/Berry_paradox
This one cost me a few hours' sleep two weeks ago; it's that cool.
Replying to your question about what's wrong with X11, I can say that one thing that puzzles me is how a protocol designed form the ground up to allow you to open windows transparently over a network connection can be *so* *much* *slower* than AJAX, which is an unholy (but impressive) melange of techs never designed to do remote apps.
I know that X11 does a lot more than AJAX, but still, couldn't anyone figure out how to split up the client/server workload so that I don't get a multi-second delay rendering every drop-down menu when I work across my 600+ kb/s connection?
is that its future per se doesn't matter.
What Google cares about is that there is a least one standards-compliant browser out there with fast javascript. Sure Google might have a slight preference for people using Chrome over another browser with fast javascript (like, say, Safari), but what really matters to them is that they are able to deliver web apps that are fast enough to be reasonable competitors to traditional desktop apps.
Chrome is a combination insurance policy/open-source soapbox whose purpose is to make sure that Google apps (and other web apps) will always have a browser to run on.
I'd like to respectfully correct a very common and understandable error in your terminology. I think you mean "topography" when you talk about the peaks and troughs of the ocean floor. "Topology" is a mathematical term describing the connectivity of sets of points: for example the surface of s sphere has one kind of topology while the surface of a donut has a different kind, because continuous transformations that don't break the 2D surface of a sphere can't morph it into a shape with a hole in it. See http://en.wikipedia.org/wiki/Topology for more if you're interested.
That aside, your point is well-taken that subs might tend to congregate in the same areas due to favorable underwater geological features.
First things first: I'm not an alarmist, and I don't think the LHC will blow up the Earth.
That said, I'd like to point out that not nearly all of that 10^20 eV is available to make new particles/black holes in the center of mass frame of the collision. Since all the collision products will have to have a ton of momentum in the direction that the cosmic ray was originally traveling, the available energy for creating new, potentially dangerous particles scales with the square root of the product of the energies (see http://www-bd.fnal.gov/public/relativity.html for a pretty good explanation of where this square root dependency comes from).
In contrast, the LHC will collide two particles in the TeV range head-on, which means the collisions have more of a chance of creating an "exotic" than even a 10^20 eV particle hitting stationary atmosphere.
However, I bet two high-energy cosmic rays each with energy > 10^14eV sometimes collide with *each other*, and that collision would have more available energy than the LHC collisions. The big question is how often does this happen? If collisions like these happen at a slow enough rate, I could imagine that the LHC might put Earth into unexplored territory in terms of numbers of collisions with ~10^14eV of available (i.e. not constrained to producing products with high momentum) energy.
I trust that the physicists have worked out the rates of these head-on, two-cosmic-ray collisions. Otherwise they would have no right saying that cosmic ray history shows that the LHC will be safe. Still, the only defense based on cosmic rays I've heard has been talking about cosmic rays hitting atmosphere, which isn't valid. Does anyone have a good link to a website analyzing the frequency of head-on two-cosmic-ray collisions?
You're right. Here's some math:
250g water in a cup of tea.
Specific heat of water = 4186 J/kg/(degree C). (see http://en.wikipedia.org/wiki/Specific_heat_capacity).
80 Celsius degree change from room temperature to boiling.
To boil a teacup's worth of water, therefore it takes ~80 kJ.
For this to be twice the energy consumed with one search, that's ~40 kJ per search.
If a search takes Google about 100 ms, that means Google would be using 400 kW while responding to your search. That feels like it's about 3 orders of magnitude too high. It's possible that the original researchers got Calories and kCal confused.
It's not a matter of composing email in 4 ms instead of 20 ms. It's that entirely new functions will drive adoption.
Examples? Apple just added face recognition to iPhoto and image stabilization to iMovie. Lots of folks are starting to do video editing. Dictation (useful when you have carpal tunnel) performance can be improved a bit with more clock cycles.
As long as engineers build faster chips, software and systems geeks will find a way to add new (useful or flashy) features that the public wants.
That's certainly the dominant position in neuroscience these days: that Nature never wastes neural hardware. I think that maybe Nature tries not to waste a combination of metabolic energy, learning time and neural hardware.
Suppose that to perform a task you need X well-trained neurons. Suppose, for a small cost to the animal, you could build 100 X neurons and have most of them silent in exchange for a lower total metabolic cost or in exchange for the ability to learn faster. (This supposition is something I'm working on: that neural hardware redundancy can buy you faster learning and metabolic efficiency, as well as robustness in the face of random cell death.) Are the 99 X neurons that almost never fire really "wasted" even though they're not used? I'd argue that they're the byproduct of an efficient process: one that's efficient at learning and at conserving energy as well as efficient enough with neural hardware.
In any case, it looks like brain regions that are not undergoing a whole lot of learning really do have only 1% of their neurons fire on a regular basis. This has been shown consistently in a bajillion systems using every technique available to modern experimenters; read (1, above) for an enormously exhaustive list.
There's no evidence to support the idea that most cortical neurons fire on a regular basis, which is your position and what most neuroscientists have been assuming all along. Although this seems to be an absolutely reasonable assumption, it's just not how things work in reality.
If a neuron has become essentially inactive, we can still poll it once every (say) 1000 simulation cycles, and simulate it with 1000 times the real chance that it will fire. If it starts firing in such a way that would increase its excitability (see (3)), we can crank up the duty cycle of the simulation for that neuron. The net effect of this trade-off trick might be to introduce noise in the extremely rare spikes coming from mostly inactive neurons (which form the vast majority of all neurons, see (1) above). In my opinion, this extra noise is a small price to pay in exchange for a factor of ~1000 speedup.
PS this trade-off is the project I should be working on now rather than checking out Slashdot. There's a good chance it won't pan out, since nobody's published about it yet.
(3) Bidirectional Modification of Presynaptic Neuronal Excitability Accompanying Spike Timing-Dependent Synaptic Plasticity. Li, Lu, Wu, Duan and Poo. Neuron, Vol. 41, 257-268, January 22, 2004.
You're right unless there's a computational way to take advantage of the fact that most neurons in cortex pretty much never fire (1), and that a small minority of synapses are responsible for nearly all of the excitation in a slab of cortical tissue (2). If not active == not important == not necessary to simulate with a 100% duty cycle (these are big "ifs"), then we could be literally about 3-5 orders of magnitude closer to being able to simulate whole brains than anyone realizes.
(1) How silent is the brain: is there a "dark matter" problem in neuroscience? Shy Shoham, Daniel H. O'Connor, Ronen Segev. J Comp Physiol A (2006)
(2) Highly Nonrandom Features of Synaptic Connectivity in Local Cortical Circuits. Sen Song, Per Jesper Sjostro, Markus Reigl, Sacha Nelson, Dmitri B. Chklovskii. PLOS biology March 2005
"It's easy to get really messy fuckin' code when you start building something big/complex."
*ducks*
Except when fools call 7.04 "7.4," which sounds higher than "7.1," which is what the same fools call "7.10". We should either get rid of two months or adopt a base-12 numbering system (and maybe genetically alter humans to have 6 fingers per hand while we're at it...), all in the name of Ubuntu.
I certainly need more than a small number of cores to make OpenOffice.org run smoothly.
*ducks*
... but they're playing video games. Seriously.
It's much easier to build entrancingly-good virtual worlds which push all the right psychological buttons (by creating the illusion of accomplishing what instincts tell you is good for survival) than to go out and conquer foreign worlds. Even if humans made interstellar space travel their top priority, it would be centuries before we could colonize other star systems. By then we'll probably be able to upload our brains into a virtual world much more rewarding than a remote, desolate rock orbiting around some distant star.
Added bonus: as long as Moore's Law holds, time inside a virtual world (measured in clock cycles, not seconds) will be much less limited than our sun's remaining couple billion years of prime fusion.
"It's a free country!"
(After shoving classmate into dirt.)
Dear hobo sapiens
I'm a big fan of Firefox, and I use a few extensions, but could you please let me more about this awesome "insert" extension? Does it allow you to insert content into sites (à la Wikipedia), or is it pr0n-related?
LeDopore
Exactly. Apple isn't going to start selling non-Apple hardware. The only scenario where an Apple-AMD merger isn't retarded is if they're sure they'll be able to use all of AMD's chips and not need any extras. Can you imagine a CPU company buying another company's CPUs to build computers? Can you imagine Apple selling MS boxes? Unless you can answer "yes" to both, then if Apple's demand doesn't fit AMD's supply exactly the merged company would be worse off. That exact fit isn't the case, nor is it ever likely to be (both Apple's demand and AMD's supply will vary due to umpteen forces), so in short an Apple-AMD merger just ain't gonna happen.
Sorry - collection costs are what I meant. You wouldn't have to interact with millions of users, just dozens of companies.
Excellent point! I think the best way to handle this is to organize people so that if a neighbourhood wanted faster service, they could each make an agreement like "If somebody increases my bandwidth to X, I would pay them Y." When enough people make enough pledges, a small, private ISP company could lay the right tech and claim the bounty. In essence, hardware upgrades should be paid by those who demand it, and as long as the network is available to the public in general the public should provide the backbone bandwidth. Would that work?
Awesome! Thanks, M. LeFou.
The big question is "paid how much?" The three biggest expenses for big ISPs are:
- "Last mile" connectivity
- Advertising and promotion
- Billing & accounting
With community- or government-provided networks, #2 and #3 go away completely, and most of #1 goes away too. Here's why.Most of the expense of getting residents Internet connectivity comes from connecting up the last few miles to individuals' homes. ISPs could in principal do away with a wired "last mile" with the right wireless technology, but that would make it easy for people to share their connections and passwords with neighbours. Instead, ISPs are charging you a premium so they can keep you from sharing your Internet connection when they insist on using cable or DSL.
We're at the point now where the inefficiency inherent in having to advertise, charge for and segregate Internet service is greater than the inefficiency inherent in public sector projects, which means that financially it's better now to have government- and community-provided Internet connectivity.
Right you are! Methinks I need to re-jig my comment threshold...
Thanks for understanding.