The only reason I can see for this (old, bad) idea to be pushed again is that some people need to create the next hype to keep their own business-model alive.
On the actual subject, if you really want every system to be individually administrated and fully secured, then go ahead and run this model. For a small network, with, say, less than ten computers this may even work. But even there it can be excessively expensive. In actual reality, any network where people think about a perimeter does need that perimeter. It needs to be implemented right, of course. For example, the only network access must be via that trusted network (enforced VPN if you are not on-site) and software must come from that trusted network as well. Also, any user active anywhere must be identified reliably (password _plus_ chipcard, e.g.) and the trusted network must, of course, be divided into zones with effective firewalling between them. Data import must go via secured channels, no just plugging in an USB stick. So not only do you need that perimeter urgently, it is by far not enough. It is just one element.
Now, this is very expensive to run and maintain. I know that. But unless you have no secrets and no IT-based business processes to protect, this is your only chance to avoid a hugely expensive disaster in the long run.
... playing "Oxygen Not Included". After giving my whole colony food poisoning by using contaminated water in the musher, the principle and what is important became quite clear.
Both, clearly. For a non-physical attack, you would have to curse the device or, say, conjure a fire-elemental to scorch it (which would then be a physical attack by the fire-elemental, but a non-physical by you).
This neatly shows the terminology is bullshit and merely an attempt to make irrelevant and obvious research sound important.
Oh, yes! And I know personally, that *gasp* LINUX is used in federal agencies and banks! They failed to make that source code secret and it is apparently completely open! I was able to just _download_ it!
In other news, the stupidity-level of your posting is staggering.
Every large-enough customer can get access to source-code of closed software. This is completely standard and there is nothing nefarious going on here. This only endangers anything US if the US messed up their own review.
It is you that is talking nonsense. What you do is called "Faulty Generalization" https://en.wikipedia.org/wiki/... It makes you statement completely irrelevant.
Then you know they are just directly and shamelessly lying to you anyways. Scum stays scum, even (or often specifically) when they go into government jobs.
While it is not clear how dangerous Spectre is on AMD and ARM, it seems to be a lot harder to exploit there and it already is hard on Intel.
What really surprises me though, is that Intel is putting out patches this badly done. It is like they are now mocking the customers that bought their fast, but inferior in every other aspect products. It is also like they are actually technologically incompetent now and do not even understand their own products.
Utilitarian viewpoints usually fall short. This one did too. As a result of his actions a lot of research was later not conducted when it actually could have done with reasonable cost/benefit to all involved.
Well, somewhat. Sure, this needs to be done carefully. But if you are so careful that you are not doing it at all, then somebody with far less morals will do it. The Chinese are by far not the worst on the planet in that regards. And they are actually trying to help people. Not so long ago, western medicine did kill most of their patients with serious issues, but those they saved made it still worthwhile. Think surgeries with 20% survival rate, but 0% if not attempted. Biological existence sucks and it is high time that we get better control of it.
Mengele experimented on healthy people that did not give consent and the experiments were not intended to help them at all. That is a tiny weeny bit different, but it takes two brain-cells to rub together to see that.
The point is that they will find out. Sure, the morality of this is problematic, but just categorically denying anything in that direction is nit much better. The history of medical advances is littered with corpses, but these are fewer corpses than if those advances had not happened.
And it is high time that the medical profession gets its head out of its collective backside and start to tackle cancer successfully. Their performance so far in that field is a disgrace.
Yes, their assumptions and predictions are baffling. Just like is somebody paid for their results and they created some fantasy narrative that arrives at that result, come to think of it.
You really have no clue what you are talking about. The name "neural nets" comes from an expectation that was never fulfilled. Actual experts call them "statistical classifiers". And the "deep" part refers to the layering, not the power of what they can do. Even deep learning needs pre-classified data unless all you are interested in is the clusterings. In actual applications you will also need to know what the clusters are and that cannot be done by any machine on the planet, because it requires actual intelligence. Deep learning just makes the part to design the network cheaper and that is it. It does not perform better than a designed network.
You misunderstand what "deep neural networks" are. They are in fact worse than flat networks, but flat networks need a human to structure them and that human need to understand the target problem. The only thing deep networks do is reduce the human part (the one needing actual intelligence) of training, at the price of worse performance both with regards to CPU cycles and accuracy.
Hence deep neural networks are not neural networks on steroids. They are a dumber, cheaper variant.
That "Go" thing was basically a stunt and entirely unfair. Also, "deep learning" is not any smarter than traditional learning and has nothing to do with human intelligence. Deep learning is what you do when you do not have a model of the target space that you could use to design the network. Results are worse than for designed networks, but they are a lot cheaper to obtain. Deep learning has no connection to what humans do when they think. Of course, most humans think rarely and hence much of what they do can be done by non-intelligent automation. And that is the problem we are facing.
And no, general AI is not even on the very distant horizon. We have absolutely nothing in machines that can produce even a dim glimmer if insight or understanding. And we do not even have a credible theory how it could be done. Maybe we will have that glimmer in 50 years, but certainly not before and we may still well never get it. The fact is that the closer we look at what humans do when they think, the more mysterious it gets. Humans can do things where it now looks very likely that the brain does not have enough computing power to do them, even if strong AI is possible. And consciousness? We have no clue at all what that is.
It is not really AI, it is plain dumb automation. But it is a lot cheaper now and a lot faster and accurate at being dumb. That is a killer. It will not replace the 1...10% of a low-skill job that actually cannot be done with "dumb", but that still means most workers in entire classes of jobs will not find work anymore.
Slashdot Mirror
The only reason I can see for this (old, bad) idea to be pushed again is that some people need to create the next hype to keep their own business-model alive.
On the actual subject, if you really want every system to be individually administrated and fully secured, then go ahead and run this model. For a small network, with, say, less than ten computers this may even work. But even there it can be excessively expensive. In actual reality, any network where people think about a perimeter does need that perimeter. It needs to be implemented right, of course. For example, the only network access must be via that trusted network (enforced VPN if you are not on-site) and software must come from that trusted network as well. Also, any user active anywhere must be identified reliably (password _plus_ chipcard, e.g.) and the trusted network must, of course, be divided into zones with effective firewalling between them. Data import must go via secured channels, no just plugging in an USB stick. So not only do you need that perimeter urgently, it is by far not enough. It is just one element.
Now, this is very expensive to run and maintain. I know that. But unless you have no secrets and no IT-based business processes to protect, this is your only chance to avoid a hugely expensive disaster in the long run.
At this time, an ad-blocker must be considered a mandatory security precaution.
Since actual fixes would impact performance, that hope is slim. It will be the least they can get away with calling "a fix".
... playing "Oxygen Not Included". After giving my whole colony food poisoning by using contaminated water in the musher, the principle and what is important became quite clear.
Both, clearly. For a non-physical attack, you would have to curse the device or, say, conjure a fire-elemental to scorch it (which would then be a physical attack by the fire-elemental, but a non-physical by you).
This neatly shows the terminology is bullshit and merely an attempt to make irrelevant and obvious research sound important.
Exactly.
Oh, yes! And I know personally, that *gasp* LINUX is used in federal agencies and banks! They failed to make that source code secret and it is apparently completely open! I was able to just _download_ it!
In other news, the stupidity-level of your posting is staggering.
Indeed. And governments can get access to windows source code as well. It is a good bet that the Russians and the Chinese also have this access.
Every large-enough customer can get access to source-code of closed software. This is completely standard and there is nothing nefarious going on here. This only endangers anything US if the US messed up their own review.
Who writes these demented articles?
It is you that is talking nonsense. What you do is called "Faulty Generalization" https://en.wikipedia.org/wiki/...
It makes you statement completely irrelevant.
Solutions in search of problems. Nothing new here.
Then you know they are just directly and shamelessly lying to you anyways. Scum stays scum, even (or often specifically) when they go into government jobs.
Indeed.
While it is not clear how dangerous Spectre is on AMD and ARM, it seems to be a lot harder to exploit there and it already is hard on Intel.
What really surprises me though, is that Intel is putting out patches this badly done. It is like they are now mocking the customers that bought their fast, but inferior in every other aspect products. It is also like they are actually technologically incompetent now and do not even understand their own products.
Utilitarian viewpoints usually fall short. This one did too. As a result of his actions a lot of research was later not conducted when it actually could have done with reasonable cost/benefit to all involved.
Well, somewhat. Sure, this needs to be done carefully. But if you are so careful that you are not doing it at all, then somebody with far less morals will do it. The Chinese are by far not the worst on the planet in that regards. And they are actually trying to help people. Not so long ago, western medicine did kill most of their patients with serious issues, but those they saved made it still worthwhile. Think surgeries with 20% survival rate, but 0% if not attempted. Biological existence sucks and it is high time that we get better control of it.
Mengele experimented on healthy people that did not give consent and the experiments were not intended to help them at all.
That is a tiny weeny bit different, but it takes two brain-cells to rub together to see that.
The point is that they will find out. Sure, the morality of this is problematic, but just categorically denying anything in that direction is nit much better. The history of medical advances is littered with corpses, but these are fewer corpses than if those advances had not happened.
And it is high time that the medical profession gets its head out of its collective backside and start to tackle cancer successfully. Their performance so far in that field is a disgrace.
Yes, their assumptions and predictions are baffling. Just like is somebody paid for their results and they created some fantasy narrative that arrives at that result, come to think of it.
Well, it is the first time that "paper pushers" will be hit large-scale. But yes, you are not wrong.
It means "we will continue to fuck you with shoddy products". Linus is right on the mark for this one.
You really have no clue what you are talking about. The name "neural nets" comes from an expectation that was never fulfilled. Actual experts call them "statistical classifiers". And the "deep" part refers to the layering, not the power of what they can do. Even deep learning needs pre-classified data unless all you are interested in is the clusterings. In actual applications you will also need to know what the clusters are and that cannot be done by any machine on the planet, because it requires actual intelligence. Deep learning just makes the part to design the network cheaper and that is it. It does not perform better than a designed network.
You misunderstand what "deep neural networks" are. They are in fact worse than flat networks, but flat networks need a human to structure them and that human need to understand the target problem. The only thing deep networks do is reduce the human part (the one needing actual intelligence) of training, at the price of worse performance both with regards to CPU cycles and accuracy.
Hence deep neural networks are not neural networks on steroids. They are a dumber, cheaper variant.
That "Go" thing was basically a stunt and entirely unfair. Also, "deep learning" is not any smarter than traditional learning and has nothing to do with human intelligence. Deep learning is what you do when you do not have a model of the target space that you could use to design the network. Results are worse than for designed networks, but they are a lot cheaper to obtain. Deep learning has no connection to what humans do when they think. Of course, most humans think rarely and hence much of what they do can be done by non-intelligent automation. And that is the problem we are facing.
And no, general AI is not even on the very distant horizon. We have absolutely nothing in machines that can produce even a dim glimmer if insight or understanding. And we do not even have a credible theory how it could be done. Maybe we will have that glimmer in 50 years, but certainly not before and we may still well never get it. The fact is that the closer we look at what humans do when they think, the more mysterious it gets. Humans can do things where it now looks very likely that the brain does not have enough computing power to do them, even if strong AI is possible. And consciousness? We have no clue at all what that is.
It is not really AI, it is plain dumb automation. But it is a lot cheaper now and a lot faster and accurate at being dumb. That is a killer. It will not replace the 1...10% of a low-skill job that actually cannot be done with "dumb", but that still means most workers in entire classes of jobs will not find work anymore.