Adversarial examples raise security and safety concerns when applying DNNs in the real world. For example, adversarially perturbed inputs could mislead the perceptual systems of an autonomous vehicle into misclassifying road signs, with potentially catastrophic consequences....
We are still a long way from finding the optimal defense strategy against these adversarial examples, and we are looking forward to exploring this exciting research area.
Consider what a generative adversarial network (GAN) can do to such an algorithm. Perhaps all an attacker needs to do is paint some rooftops in a manner that creates an off-by-one error on counting streets and then they can misdirect (or even intercept) somebody's mail.
Audacious, a descendant of XMMS (which was a clone of Winamp), works wonderfully. Its "Winamp Classic Interface" looks exactly like Winamp and even (iirc) supports Winamp skins.
That said, I do miss the old (original) Whitecap visualization (one of the very few in which you could really see the music in what was still a visually stunning display), which only works on Winamp on Windows. (...not that Winamp's return would allow me to run this again.)
We need to use stronger language here. Some perpetrators of swatting do not understand the gravity, the fact that people die in some of these incidents.
This isn't a "hoax". This is a potentially fatal attack by proxy.
I applaud the Seattle police for trying to be proactive on this. This service will save lives.
I do not think it is acceptable to use this type of tool for anything resembling data science or statistics, and access to it needs to be very tightly controlled with transparent policies and a justifiable purpose.
I saw a documentary on the NYC surveillance capabilities a few years ago, back when they could search only by clothing color and a few other visual cues easier than race/ethnicity. This is a system for responding to descriptions of suspected terrorists (or criminals?), allowing easy filtering of a massive collection of CCTV cameras (the documentary demonstrated a real-time filtering of CCTV feeds, seeking somebody wearing red. It was eerie). I'd be okay with this if it had a strict access control policy requiring highly detailed audit logs and ideally something resembling a warrant procedure, especially if it were limited to terrorism investigations.
I skipped over this article and only later found it on Twitter. Yes, it's good that MS is making a stand to demand reasonable parental leave of its partners, using it substantial economic power to help this happen... but it's quite significant that they're demanding twelve weeks of paid parental leave for all partners, even landscapers, janitors, and cafeteria workers.
I've been pretty happy with CardDAV support via the CardBook addon, which lets me connect (read/write) to the same contact list as on my smart phone and web mail. CardDAV is an extension of WebDAV and implemented via HTTP rather than LDAP, but it's far more standardized and specialized to contact management.
Perhaps you can connect via CalDAV to a DavMail intermediary that then translates to LDAP. Perhaps your enterprise can maintain a global DavMail server to ease that. See also Bug 86405 comment 86, which extols the virtues of CardDAV.
I've recently read studies that show that sleeping in bright (especially UV-lit) conditions harms sleep efficiency. I wonder how much of this test's variance would be explained by that? I have blackout curtains in my bedroom for a reason. (The Live Science article even eludes to this in the researchers' recommendation "that people make sure they're exposed to light early in the morning, but not at nighttime." No mention of whether this was controlled for in the study, but I doubt it.)
Another similar thought is that of consistent bed times, which are also shown to help sleep efficiency. Night owls are (perhaps) less likely to consistently go to bed at the same time every night while early risers (perhaps) go to sleep --and wake up-- at consistent times. These are guesses though.
Once is happenstance. Twice is coincidence. Three times is enemy action.
That's stated by James Bond the 1964 Goldfinger film. I see no indication of it being named "Fleming's Razor" or that the original author (Ian Fleming) wrote the line, though it has been quite some time since I read that book.
Like others here, I strongly disagree with this on the grounds of anonymity and flamewar prevention.
However, the idea itself has merits if it's kept hidden. Slashdot-internal metrics (unavailable to users) could track users that down-vote too much, especially when such votes disagree with the majority of other votes on the same post. These users could then be penalized in some way, e.g. lowered likelihood of being selected as a moderator, fewer mod points, and/or negative Achievements (which might become a badge of honor among trolls, and that's okay so long as it also negatively affects the user's moderation capabilities).
The reCAPTCHA service does two things. Verifying a user is a human by offering something that's really hard to automate is the one everybody knows about. The other is an effort to crowdsource understanding of images. This started with decoding the words in scanned books that OCR was having difficulty with.
There's your competition (though it's admittedly restricted to modern texts, so historical context and historical characters are beyond its scope... and reCAPTCHA has recently moved on to other forms of image recognition.)
The original article, as saved by the Internet Archive, had a slightly different subtitle:
‘I borrowed from the films of Leni Riefenstahl to show that these US soldiers were like something out of Nazi propaganda. I even put one in an SS uniform. But no one noticed’
(Emphasis added to highlight the text that was removed).
The current version has a note at the bottom saying:
The subheading of this article was amended on 23 January 2018 to remove a reference to US soldiers.
I've been thinking about this for quite a while. Breaking this into genres should be pretty easy, dividing between non-fiction ("documentary" podcasts, audio books, etc, as further classified like books), fiction (non-music "entertainment" pieces, as further classified like movies), and presumably "other" for oddities that can't be classified that way (the fourth group is "music," which they've already got a good handle on).
For music, Pandora currently pays musicians and other experts to manually go in and rate numerous features for music because they're too hard to extract with machines; just like Uber and Lyft, the idea is that eventually this will be automated). For spoken word, this becomes pretty trivial. Cadence, tone, and similar metrics should be easy to get at least rudimentary metrics on, and the more important stuff (content!) should be a matter of state-of-the-art speech-to-text translation and then enter your favorite text parsing machine learning system(s).
Even categorizing into each genre should be really easy since the words can be easily extracted on nearly every title (when speech-to-text fails and there isn't already a transcript provided, perhaps there's room for some paid human transcription... or perhaps that bars entry to the Pandora world, leaving the duty up to the content creators). Additionally, there's metadata on the listener's preferences for speakers' timing, cadence, style, and associations.
I've been thinking about this with respect to short-form standup comedy. Small quips that aren't knit into an overarching theme (the way you get in a larger standup routine), or that are knit into a common enough theme to weave together and create a custom comedy show on the spot. Think about "airline security" or "road rage" for example. A sophisticated ML system (with sufficient source material) designed to tell a story (not just a "playlist") with small standup segments would be worth a pretty penny.
If there's an easy way to get it that MPAA, Amazon Prime, Netflix, and others can actually support (and ideally offer a more reliable service with better UX and more content), then the "need" for these illegal add-ons will diminish radically. Then it's okay to pick off the bigger facilitators if they're still too big for comfort.
MLB.tv does this. I can watch it on my Kodi TV setup by logging into the account that I pay for. It's not supported by MLB, but it still works (most of the time) and MLB has no incentive to shut it down.
At some point, these content providers will realize that their content is actually worth something on its own. They'll be fine releasing free and open source software that can securely log in and stream their content to paying customers without an iota of non-free software on the client system.
Like any kind of basic test, IQ tests aren't terribly abstractable. Therefore, the supposed correlation between this type of games and IQ tests isn't terribly indicative of intelligence.
Therefore: if you like IQ tests, you should really try these games.
This also reminds me of a quote:
I have no idea [what my IQ is]. People who boast about their IQ are losers. -- Stephen Hawking
I think this study may have been too bounded by what we currently know and see as limits.
Perhaps once we better understand cancer, it will be the key to making this all work; a "mild" and controlled cancer might solve the cell regeneration issue, especially if coaxed by nano-scale machines, which present another possibility: replacing and/or supplementing the failing parts of our bodies. Theoretically, a very careful balance of repaired telomeres, nanotech-supplemented systems, and controlled cancers could establish an equilibrium that would buy enough time for us to come up with another form of immortality (mind transference, fully artificial cells, or something more fantastical and yet to be imagined). You can't really disprove that with math since there are too many variables and suppositions.
No, and I therefore cannot comment on how soundly it may or may not break "the formula" but there are exceptions to every rule. This does not appear to be one, however, since it is not a Hollywood film (sorry, I failed to quantify my remark by that characteristic). It's also an adaptation.
Modern movies all fit a very tight formula, which is admittedly very powerful and compelling, but it prevents certain types of creativity from shining. Atop that, movie studios refuse to take risks on new material, instead making adaptations, reboots, remakes, and encores. This further limits what a movie can do.
This is an arena that televisions series have stolen from movies; most episodes are designed to fit that tight formula while advancing a larger arc (better yet, multiple larger arcs!) while a few can break the mold with minimal risk to audience retention (for example, instead of the plot twist being half to three-quarters through, it can be elsewhere, or even a build-up for a larger surprise in the next episode).
Horror movies are rarely heavy in sophistication. They just go in for emotional investment so they can lead you to a series of surprises, some of which will startle you and others that might haunt you. This adapts to that oversimplified formula very very well. Additionally, horror has its own tight formulae, so audiences get what they expect and are only disappointed when there wasn't the anticipated level of startles, eeriness, or innuendo. There's no risk to the hook being problematic since it's pretty much always shown in full force in the movie's trailer.
(Also note that It is a remake (and an adaptation), though Get Out is not.)
Last I heard, Google has all of its internal services exposed to the public internet. This means that when an incident like this happens, anybody can exploit it.
Using a VPN (or equivalent, such as requiring a dynamic SOCKS tunnel through an SSH bastion, a.k.a. a jump host) would at least add one layer of protection beyond this: jump into the dev network (which may or may not be the same as the office network), then connect to internal services (selective use of proxies is made easier with things like FoxyProxy). That way you need access to the network in addition to access to the server within that network.
A subset of that network could be made available (via VLANs, IPTables on specific SSH bastions, or the like) for remote contractors who only need access to certain servers.
We will continue to press our case in court that the Electronic Communications Privacy Act (ECPA) – a law enacted decades before there was such a thing as cloud computing – was never intended to reach within other countries’ borders.
... We challenged the warrant that resulted in this ligation because we believed U.S. search warrants shouldn’t reach over borders to seize the emails of people who live outside the United States and whose emails are stored outside the United States.
This is really important not only for international privacy but also for US business profits from international sources (which is a major reason for Microsoft being on the right side of the issue).
Slashdot Mirror
Consider what a generative adversarial network (GAN) can do to such an algorithm. Perhaps all an attacker needs to do is paint some rooftops in a manner that creates an off-by-one error on counting streets and then they can misdirect (or even intercept) somebody's mail.
Read this article on Camouflaged Graffiti on Road Signs Can Fool Machine Learning Models (from 1y ago) for some examples of how simple stickers or graffiti can fool these systems. It also hit Slashdot, though all I can find is an article on 'Psychadelic' Stickers That Confuse AI Image Recognition.
Audacious, a descendant of XMMS (which was a clone of Winamp), works wonderfully. Its "Winamp Classic Interface" looks exactly like Winamp and even (iirc) supports Winamp skins.
That said, I do miss the old (original) Whitecap visualization (one of the very few in which you could really see the music in what was still a visually stunning display), which only works on Winamp on Windows. (...not that Winamp's return would allow me to run this again.)
We need to use stronger language here. Some perpetrators of swatting do not understand the gravity, the fact that people die in some of these incidents.
This isn't a "hoax". This is a potentially fatal attack by proxy.
I applaud the Seattle police for trying to be proactive on this. This service will save lives.
I do not think it is acceptable to use this type of tool for anything resembling data science or statistics, and access to it needs to be very tightly controlled with transparent policies and a justifiable purpose.
I saw a documentary on the NYC surveillance capabilities a few years ago, back when they could search only by clothing color and a few other visual cues easier than race/ethnicity. This is a system for responding to descriptions of suspected terrorists (or criminals?), allowing easy filtering of a massive collection of CCTV cameras (the documentary demonstrated a real-time filtering of CCTV feeds, seeking somebody wearing red. It was eerie). I'd be okay with this if it had a strict access control policy requiring highly detailed audit logs and ideally something resembling a warrant procedure, especially if it were limited to terrorism investigations.
I skipped over this article and only later found it on Twitter. Yes, it's good that MS is making a stand to demand reasonable parental leave of its partners, using it substantial economic power to help this happen ... but it's quite significant that they're demanding twelve weeks of paid parental leave for all partners, even landscapers, janitors, and cafeteria workers.
I've been pretty happy with CardDAV support via the CardBook addon, which lets me connect (read/write) to the same contact list as on my smart phone and web mail. CardDAV is an extension of WebDAV and implemented via HTTP rather than LDAP, but it's far more standardized and specialized to contact management.
Perhaps you can connect via CalDAV to a DavMail intermediary that then translates to LDAP. Perhaps your enterprise can maintain a global DavMail server to ease that. See also Bug 86405 comment 86, which extols the virtues of CardDAV.
See also the full Cisco Talos post, New VPNFilter malware targets at least 500K networking devices worldwide, which has all of the technical details, including all indicators of compromise (IOCs).
yes, "allude." surely you could have figured that out.
I've recently read studies that show that sleeping in bright (especially UV-lit) conditions harms sleep efficiency. I wonder how much of this test's variance would be explained by that? I have blackout curtains in my bedroom for a reason. (The Live Science article even eludes to this in the researchers' recommendation "that people make sure they're exposed to light early in the morning, but not at nighttime." No mention of whether this was controlled for in the study, but I doubt it.)
Another similar thought is that of consistent bed times, which are also shown to help sleep efficiency. Night owls are (perhaps) less likely to consistently go to bed at the same time every night while early risers (perhaps) go to sleep --and wake up-- at consistent times. These are guesses though.
That's stated by James Bond the 1964 Goldfinger film. I see no indication of it being named "Fleming's Razor" or that the original author (Ian Fleming) wrote the line, though it has been quite some time since I read that book.
Like others here, I strongly disagree with this on the grounds of anonymity and flamewar prevention.
However, the idea itself has merits if it's kept hidden. Slashdot-internal metrics (unavailable to users) could track users that down-vote too much, especially when such votes disagree with the majority of other votes on the same post. These users could then be penalized in some way, e.g. lowered likelihood of being selected as a moderator, fewer mod points, and/or negative Achievements (which might become a badge of honor among trolls, and that's okay so long as it also negatively affects the user's moderation capabilities).
The reCAPTCHA service does two things. Verifying a user is a human by offering something that's really hard to automate is the one everybody knows about. The other is an effort to crowdsource understanding of images. This started with decoding the words in scanned books that OCR was having difficulty with.
There's your competition (though it's admittedly restricted to modern texts, so historical context and historical characters are beyond its scope ... and reCAPTCHA has recently moved on to other forms of image recognition.)
The original article, as saved by the Internet Archive, had a slightly different subtitle:
‘I borrowed from the films of Leni Riefenstahl to show that these US soldiers were like something out of Nazi propaganda. I even put one in an SS uniform. But no one noticed’
(Emphasis added to highlight the text that was removed).
The current version has a note at the bottom saying:
The subheading of this article was amended on 23 January 2018 to remove a reference to US soldiers.
This is the slowest Linux kernel release process, not the slowest kernel itself.
I've been thinking about this for quite a while. Breaking this into genres should be pretty easy, dividing between non-fiction ("documentary" podcasts, audio books, etc, as further classified like books), fiction (non-music "entertainment" pieces, as further classified like movies), and presumably "other" for oddities that can't be classified that way (the fourth group is "music," which they've already got a good handle on).
For music, Pandora currently pays musicians and other experts to manually go in and rate numerous features for music because they're too hard to extract with machines; just like Uber and Lyft, the idea is that eventually this will be automated). For spoken word, this becomes pretty trivial. Cadence, tone, and similar metrics should be easy to get at least rudimentary metrics on, and the more important stuff (content!) should be a matter of state-of-the-art speech-to-text translation and then enter your favorite text parsing machine learning system(s).
Even categorizing into each genre should be really easy since the words can be easily extracted on nearly every title (when speech-to-text fails and there isn't already a transcript provided, perhaps there's room for some paid human transcription ... or perhaps that bars entry to the Pandora world, leaving the duty up to the content creators). Additionally, there's metadata on the listener's preferences for speakers' timing, cadence, style, and associations.
I've been thinking about this with respect to short-form standup comedy. Small quips that aren't knit into an overarching theme (the way you get in a larger standup routine), or that are knit into a common enough theme to weave together and create a custom comedy show on the spot. Think about "airline security" or "road rage" for example. A sophisticated ML system (with sufficient source material) designed to tell a story (not just a "playlist") with small standup segments would be worth a pretty penny.
People just want easy access to content.
If there's an easy way to get it that MPAA, Amazon Prime, Netflix, and others can actually support (and ideally offer a more reliable service with better UX and more content), then the "need" for these illegal add-ons will diminish radically. Then it's okay to pick off the bigger facilitators if they're still too big for comfort.
MLB.tv does this. I can watch it on my Kodi TV setup by logging into the account that I pay for. It's not supported by MLB, but it still works (most of the time) and MLB has no incentive to shut it down.
At some point, these content providers will realize that their content is actually worth something on its own. They'll be fine releasing free and open source software that can securely log in and stream their content to paying customers without an iota of non-free software on the client system.
I've bookmarked GhostText as a potential replacement for It's All Text, though I'm still holding onto FF56 for NoScript, the Debian testing package (it landed in unstable just today), and a few security fixes (like this DOMParser cookie bug). It actually looks better in some regards. Learn more on its GitHub page.
Like any kind of basic test, IQ tests aren't terribly abstractable. Therefore, the supposed correlation between this type of games and IQ tests isn't terribly indicative of intelligence.
Therefore: if you like IQ tests, you should really try these games.
This also reminds me of a quote:
I think this study may have been too bounded by what we currently know and see as limits.
Perhaps once we better understand cancer, it will be the key to making this all work; a "mild" and controlled cancer might solve the cell regeneration issue, especially if coaxed by nano-scale machines, which present another possibility: replacing and/or supplementing the failing parts of our bodies. Theoretically, a very careful balance of repaired telomeres, nanotech-supplemented systems, and controlled cancers could establish an equilibrium that would buy enough time for us to come up with another form of immortality (mind transference, fully artificial cells, or something more fantastical and yet to be imagined). You can't really disprove that with math since there are too many variables and suppositions.
(We are the borg?)
Have you seen game of thrones? They took risks and broke out of the simplified formulas
Game of Thrones is not a Hollywood movie either. See my second paragraph. The "multiple larger arcs" was written with GoT chief in mind.
Have you seen Victoria and Abdul?
No, and I therefore cannot comment on how soundly it may or may not break "the formula" but there are exceptions to every rule. This does not appear to be one, however, since it is not a Hollywood film (sorry, I failed to quantify my remark by that characteristic). It's also an adaptation.
Modern movies all fit a very tight formula, which is admittedly very powerful and compelling, but it prevents certain types of creativity from shining. Atop that, movie studios refuse to take risks on new material, instead making adaptations, reboots, remakes, and encores. This further limits what a movie can do.
This is an arena that televisions series have stolen from movies; most episodes are designed to fit that tight formula while advancing a larger arc (better yet, multiple larger arcs!) while a few can break the mold with minimal risk to audience retention (for example, instead of the plot twist being half to three-quarters through, it can be elsewhere, or even a build-up for a larger surprise in the next episode).
Horror movies are rarely heavy in sophistication. They just go in for emotional investment so they can lead you to a series of surprises, some of which will startle you and others that might haunt you. This adapts to that oversimplified formula very very well. Additionally, horror has its own tight formulae, so audiences get what they expect and are only disappointed when there wasn't the anticipated level of startles, eeriness, or innuendo. There's no risk to the hook being problematic since it's pretty much always shown in full force in the movie's trailer.
(Also note that It is a remake (and an adaptation), though Get Out is not.)
Last I heard, Google has all of its internal services exposed to the public internet. This means that when an incident like this happens, anybody can exploit it.
Using a VPN (or equivalent, such as requiring a dynamic SOCKS tunnel through an SSH bastion, a.k.a. a jump host) would at least add one layer of protection beyond this: jump into the dev network (which may or may not be the same as the office network), then connect to internal services (selective use of proxies is made easier with things like FoxyProxy). That way you need access to the network in addition to access to the server within that network.
A subset of that network could be made available (via VLANs, IPTables on specific SSH bastions, or the like) for remote contractors who only need access to certain servers.
This is really important not only for international privacy but also for US business profits from international sources (which is a major reason for Microsoft being on the right side of the issue).