Literally everything that you said in that post is wrong. In terms of being very similar to Java, here are a few differences:
C# has signed and unsigned primitive types (Java has only signed)
C# has pass-by-value struct types.
C# has asynchronous dispatch / rendezvous in the language.
C# has type-preserving generics (Java's are type erasing).
C# has type-safe callback functions ('delegates'), Java requires you to use anonymous classes for the same functionality (both now use lambdas where possible).
Java methods are always virtual, C# methods are non-virtual unless otherwise specified.
As to the var thing, this is simple type inference. It is roughly analogous to C++11's auto. It's nothing to do with Visual Basic, it's more important for things like LINQ, where you end up with complex types that depend on the value of an expression and you don't want to force the programmer to write them explicitly. In these cases, var lets you bind variables to values of anonymous types. The code is still strongly statically typed.
Are you assuming that the.NET Core version numbers correspond to some other project's version numbers (e.g..NET Framework of C#)? Because if so, that would explain why your posts seem to be incomprehensible nonsense..NET Core supports the most recent versions of C# and F#. They support a lot of third-party code that was designed for Windows and on Windows can be used instead of.NET Framework (and Visual Studio pushes you in that direction now).
Battle Chess ran on my 8086 with 640KB of RAM. Given the scenario outlined by the grandparent, I'd bring it along to play against AlphaGo. Sure, AlphaGo would segfault on startup having exhausted the memory and forfeit, but even when playing as white I'd get one valid move before it died.
DeepMind's claim is that their deep neural network design can beat something programmed by a human programmer. If they'd added a similar database to their code, it would not have supported their claim. If they could have won with Stockfish keeping theirs, then they'd have done so.
Why do you think that? In the USA, in 2016, 42% of workers made $15/hour or less (quick search, didn't find more recent statistics). Are all of those people busy studying so they can get a better job? If they do all study, who does the jobs that they're doing now? Do you expect to sustain a population where 42% are aged 16-21?
It's quite understandable. The goal of the Play Store is to make it possible to install things that use Google Play Services and provide personal information on users to Google. Similarly, the goal of Chrome is to make it easy to load web pages that depend on Google Ads and Google Analytics, to send personal information on users to Google. Both are doing their job. Tightening restrictions on Chrome extensions makes it harder for things that prevent Google from data harvesting. Weakening restrictions in the Play Store makes it easier for things that data harvest on behalf of Google. They're entirely self consistent activities.
That's not news because the fact that Google and Apple have complete control over their respective app stores is well known. Apple's control over iOS is tighter than Google's control over Android in some ways because at least there are alternative ways of getting pre-built applications for Android (for example, Amazon's app store or F-Droid), whereas the only ways of installing apps on an iOS device other than from Apple's store are to build them from source (on a Mac) or to have a corporate account that lets you install internal apps on a limited number of devices.
The more troubling thing is the number of Android apps that depend on Google Play Services and so won't work on an Android phone without some proprietary Google code running with elevated privileges, even if you could get them from another source.
That depends a lot on where you are. Depending on local geography and prevailing wind patterns, the difficulty of predicting the weather varies hugely. I used to live somewhere that was usually at the intersection of three large weather systems, two coming from the sea and one from land. The actual weather depended on the interaction of the three and so it was pretty common for the forecast for the current day to be wildly inaccurate (as in, predicting sleet in the afternoon on days that turned out to have clear skies and warmish temperatures, or vice versa). Now I live somewhere where most weather systems roll straight over us. Most of the time, you can predict the weather by clanging at a satellite map - assume clouds will follow their current paths all day and you're pretty accurate, do a little bit of curve fitting and you're very accurate.
My biggest complaint about weather forecasts is that they never report their error margins. The weather is a chaotic system, but that's fairly well understood. The Met Office in the UK runs several different models and then picks one result. If all of the models predict the same thing, it's pretty likely. If all of them are predicting different things, then it would be nice to have that information presented so I can see if 'sunny today' means 'we're pretty sure it will be sunny today' or if it means 'it's either going to be sunny or piss it down with rain, depending on what happens when these two fronts collide. Slightly more likely to be sunny, but don't bet on it...'
I didn't make it up to San Francisco on my last trip to the Bay Area, but everywhere between San Jose and Mountain View that I visited had people riding them on sidewalks and on the road ignoring traffic signals. Parking them seemed to be the smallest problem: people are riding them dangerously without any kind of insurance, who suffers when they hit a car or a pedestrian?
That's true, but it's worth noting that the ACME protocol that Let's Encrypt supports is open and can be deployed by other providers. The idea was always that other providers would be able to deploy it. There's nothing stopping the government from running a server for the protocol that issues certificates to anything on a.gov domain. I think at least a couple of CAs now support ACME for deploying subdomain certificates once you've validated that you own a particular domain.
I've seen it in a bunch of companies. Not at a company-wide level, but if your division doesn't spend its budget this quarter then next year it will get a smaller one, so you have a strong incentive to spend it all. What companies have you worked for where not spending your budget in one quarter gives you more the next quarter?
Last time I looked, AMD was responsible for around 20% of total CPU shipments. Even if they aren't suffering from any supply-chain problems, even gaining 1% of the market would require that they increase production by 5%, getting a majority of the market would require that they increase their production by 150%. That's not normally very easy.
How about an 89% share? An 88% share? An 87% share? At what level is it reasonable, and how exactly do you determine that?
It varies from market to market. The test is whether you can behave as if you had 100%. For example, if you put up your prices, do you lose market share? There are a number of tests that regulators perform to determine whether your market share is distorting the market. In a market with no barriers to entry and high visibility for new entrants, even a 99% market share would not be likely to need regulating. A market with strong network effects and high barriers to entry may start to see distortion as soon as one player is bigger than the second-largest player.
You're assuming that the Chinese data collection incorporates high security. When you're talking about spying, it's not a question of who gets your data, it's who gets your data first. I probably don't have anything to worry about from any government knowing the contents of my devices, but I might be more worried about various organised crime syndicates knowing things that can be easily inferred from these devices (e.g. when my house is going to be unoccupied). I'd trust the NSA or GCHQ to keep the data that they've stolen about me secret more than I would trust their Russian or Chinese counterparts.
As an individual, there's little that China can do to me - I've been to China once and I'm not inclined to repeat that. But as a corporate user, a Chinese company having access to all of my work-related emails and access to any work-related files on my device could have a significant impact on my employer's ability to remain competitive (production is much cheaper than R&D), which would impact their ability to pay me.
British companies used to be warned not to discuss work on Air France flights because the French intelligence agencies would put microphones in the seats and pass on any commercially-sensitive information they picked up to French companies. China still has a very close relationship between corporations and intelligence. The NSA is far less likely to pass on anything that they learn to US-based competitors.
It's not nearly as closely tied as it used to be. Modern flash uses multi-level cells, so each cell stores 2 or more bits, but is also very unreliable so a lot of cells are used for error correcting codes. The amount of redundancy that you need for error correction dominates the useable capacity of modern flash. Cheap flash storage typically uses the same chips as more expensive flash storage but doesn't reserve as much space for error correction. The probability of an error is the same, the probability of a non-correctible error is a lot higher.
I don't mind the lack of a mouse so much - it's a small-screen device and the touchscreen is usually better in this form factor. It does bug me that the 'Pro' keyboard has an emoji button but no escape key though.
I suspect that the argument for not supporting mice is that they don't want people doing crappy ports of desktop apps that require a mouse and then getting complaints from users that they don't work well with touchscreens. Apple's decision to ship one-button mouse worked out well for them in the era of resistive touchscreens, where a UI designed for a single-button mouse worked well but using Windows on the same devices was painful (you typically had a button in a floating window that you had to tap to make the next mouse click a right button).
And no, asking slashdot to bend over backwards because iPads are doing the wrong thing
You mean the wrong thing where they parse a page that sets its encoding to UTF-8 in both meta tags and HTTP headers, send a response with HTTP headers indicating that it contains UTF-8 data, and are surprised when the server then incorrectly handles UTF-8 data?
Slashdot advertises UTF-8 support in meta tags in this page and in the HTTP headers. When the iOS device submits the form, it also sets the content type to UTF-8. Slashdot then interprets this as some random 8-bit codepage. If Slashdot advertised ASCII support, then iOS would use ASCII for form submission. Advertising that you accept UTF-8 and then breaking on any non-ASCII characters is just plain broken.
Soylent News' fork of Slashcode fixed these issues ages ago.
There are all sorts of interesting things here. In the UK, a lot of supermarkets sell milk as a loss leader: many people think about the price of milk as a proxy for the price of everything, so if milk is cheaper in your shop then they assume that everything is cheaper. They also often have very short-term supply contracts with farms and will buy a week in advance and shift suppliers to keep their costs low. Farms often find that no one will buy their milk in a given week for above the production cost, but they sell it for below cost because they have no marginal cost from producing milk and their fixed costs are the same whether they sell or don't (if they sell, they go out of business eventually, if they don't sell then they go out of business soon). This then means that there isn't a contraction in supply (at least, in the short term), so other farms can't raise their prices. The supermarkets form an oligopsony that keeps the price very low.
As you say, if you can produce yoghurt some of your milk, you can keep it for a few weeks and watch the prices fluctuate, selling it only when the price goes above the cost of production. I'd expect that to keep the price of milk lower, but it may be that in the Netherlands there either isn't the same oligopsony of supermarkets, there isn't the trend of using milk as a loss leader, or there so many farms are betting on yoghurt that there's a supply shortage for milk. It may even be that yoghurt is filling the same loss-leader niche as milk in the UK.
The tens of thousands of people that voted and didn't see the stickers on their pages?
You're misunderstanding the attack. Person A votes. Person A puts an 'I voted' sticker on their account. Person B has not voted. Person B does not put a sticker. Person C, who votes for the same party as Person A goes to Person A's Facebook page, but does not see an 'I've voted' sticker. Person D, who votes the same way as Person B, goes to Person B's Facebook page and does see one even though B didn't post it. There's now an increased probability that Person D will vote and person C won't. If you pick the people in marginal constituencies, there's a good chance you can swing the election. If Person C notices the sticker is missing, then Facebook just blames it on delays in updates propagating between the system. If Person D notices a spurious sticker, then Facebook blames it on a glitch putting stickers on accounts that shouldn't have had them.
That's assuming that someone would even bother to check - when was the last time you saw a sticker on someone's account and checked with them that it was actually something that they'd put there? When was the last time you checked that other people could see all of the stickers on your account?
This kind of manipulation is hard to spot, but has a measurable impact. Most of the time, Facebook uses it to convince people to buy certain products, it just seems a little bit more sinister when those products are political parties.
Then there'd be a huge backlash against FB and it'd never, ever happen again.
Maybe. If people noticed. Remind me, how much backlash has there been against Facebook for the last half dozen times they were caught testing psychological manipulation techniques?
I think your "except two things" are nearly impossible for MS to fix though... Are you saying you would be happy with another key combo sending SIGINT in a terminal (for example) or you would only be happy with an Apple-key style situation across the OS? Or are you saying that when the time comes it's going to be a lump situation?:-)
I hadn't thought of that, but having windows-key-is-control in the terminal would fix that annoyance for me. Oh, and having the WSL terminal support ^T for SIGINFO (the one of the big things I miss going from FreeBSD to Linux).
They've done some really nice work on the terminal subsystem. If you run a Konsole or the GNOME terminal (or your favourite X11 terminal emulator) in WSL, you can then run cmd.exe or powershell.exe and run Windows commands. It doesn't quite work properly, because the traditional Windows terminal handles editing, history, and autocompletion (the shell doesn't), so all of these are broken, but hopefully they'll be fixed at some point soon. At the very least, I can stay in konsole and just pop into cmd briefly to run Windows builds.
How would you notice it? Would you notice if your friends didn't display 'I voted' stickers on Facebook? Or if they showed up a day late, after the election? That kind of thing could be easily explained as a propagation glitch and would be really hard for anyone to determine was directed at a specific group. Remember, they'd only be targeting people with a specific affiliation in selected areas where their decision to vote or not to vote would impact the results. That's a fairly small proportion of the electorate. And you wouldn't do it to everyone in that group, only to enough to affect the outcome. Unless you get everyone to record when they've seen the badges, it's basically impossible to detect.
Don't fool yourself into thinking that that face detection is anything more than a GPGPU program.
Nope, on OMAP SoCs (and quite a few others) it's dedicated hardware in a separate fixed-function logic block.
Ditto for any "AI", which is just a bunch of matrix multiplications of signals with state matrices, and is almost, but not quite, entirely unlike real natural fleshy neural networks.
Except that AI workloads typically have much higher locality of reference than GPU ones, so accelerators have totally different memory controller designs (memory + caches, rather than streaming).
Besides: Said face detection is a perfect example of exactly this useless-gimmick-adding behavior I criticized in my original comment. So thank you for that.:)
The face detection block is usually used with the camera to auto-focus on the part of the image that contains a face, rather than the background, even if the face isn't in the centre of the picture. I think most people that use their phone's camera find that a useful feature.
Slashdot Mirror
Literally everything that you said in that post is wrong. In terms of being very similar to Java, here are a few differences:
As to the var thing, this is simple type inference. It is roughly analogous to C++11's auto. It's nothing to do with Visual Basic, it's more important for things like LINQ, where you end up with complex types that depend on the value of an expression and you don't want to force the programmer to write them explicitly. In these cases, var lets you bind variables to values of anonymous types. The code is still strongly statically typed.
Are you assuming that the .NET Core version numbers correspond to some other project's version numbers (e.g. .NET Framework of C#)? Because if so, that would explain why your posts seem to be incomprehensible nonsense. .NET Core supports the most recent versions of C# and F#. They support a lot of third-party code that was designed for Windows and on Windows can be used instead of .NET Framework (and Visual Studio pushes you in that direction now).
Battle Chess ran on my 8086 with 640KB of RAM. Given the scenario outlined by the grandparent, I'd bring it along to play against AlphaGo. Sure, AlphaGo would segfault on startup having exhausted the memory and forfeit, but even when playing as white I'd get one valid move before it died.
DeepMind's claim is that their deep neural network design can beat something programmed by a human programmer. If they'd added a similar database to their code, it would not have supported their claim. If they could have won with Stockfish keeping theirs, then they'd have done so.
Minimum wage jobs are not meant to live on.
Why do you think that? In the USA, in 2016, 42% of workers made $15/hour or less (quick search, didn't find more recent statistics). Are all of those people busy studying so they can get a better job? If they do all study, who does the jobs that they're doing now? Do you expect to sustain a population where 42% are aged 16-21?
It's quite understandable. The goal of the Play Store is to make it possible to install things that use Google Play Services and provide personal information on users to Google. Similarly, the goal of Chrome is to make it easy to load web pages that depend on Google Ads and Google Analytics, to send personal information on users to Google. Both are doing their job. Tightening restrictions on Chrome extensions makes it harder for things that prevent Google from data harvesting. Weakening restrictions in the Play Store makes it easier for things that data harvest on behalf of Google. They're entirely self consistent activities.
That's not news because the fact that Google and Apple have complete control over their respective app stores is well known. Apple's control over iOS is tighter than Google's control over Android in some ways because at least there are alternative ways of getting pre-built applications for Android (for example, Amazon's app store or F-Droid), whereas the only ways of installing apps on an iOS device other than from Apple's store are to build them from source (on a Mac) or to have a corporate account that lets you install internal apps on a limited number of devices.
The more troubling thing is the number of Android apps that depend on Google Play Services and so won't work on an Android phone without some proprietary Google code running with elevated privileges, even if you could get them from another source.
That depends a lot on where you are. Depending on local geography and prevailing wind patterns, the difficulty of predicting the weather varies hugely. I used to live somewhere that was usually at the intersection of three large weather systems, two coming from the sea and one from land. The actual weather depended on the interaction of the three and so it was pretty common for the forecast for the current day to be wildly inaccurate (as in, predicting sleet in the afternoon on days that turned out to have clear skies and warmish temperatures, or vice versa). Now I live somewhere where most weather systems roll straight over us. Most of the time, you can predict the weather by clanging at a satellite map - assume clouds will follow their current paths all day and you're pretty accurate, do a little bit of curve fitting and you're very accurate.
My biggest complaint about weather forecasts is that they never report their error margins. The weather is a chaotic system, but that's fairly well understood. The Met Office in the UK runs several different models and then picks one result. If all of the models predict the same thing, it's pretty likely. If all of them are predicting different things, then it would be nice to have that information presented so I can see if 'sunny today' means 'we're pretty sure it will be sunny today' or if it means 'it's either going to be sunny or piss it down with rain, depending on what happens when these two fronts collide. Slightly more likely to be sunny, but don't bet on it...'
I didn't make it up to San Francisco on my last trip to the Bay Area, but everywhere between San Jose and Mountain View that I visited had people riding them on sidewalks and on the road ignoring traffic signals. Parking them seemed to be the smallest problem: people are riding them dangerously without any kind of insurance, who suffers when they hit a car or a pedestrian?
That's true, but it's worth noting that the ACME protocol that Let's Encrypt supports is open and can be deployed by other providers. The idea was always that other providers would be able to deploy it. There's nothing stopping the government from running a server for the protocol that issues certificates to anything on a .gov domain. I think at least a couple of CAs now support ACME for deploying subdomain certificates once you've validated that you own a particular domain.
I've seen it in a bunch of companies. Not at a company-wide level, but if your division doesn't spend its budget this quarter then next year it will get a smaller one, so you have a strong incentive to spend it all. What companies have you worked for where not spending your budget in one quarter gives you more the next quarter?
Last time I looked, AMD was responsible for around 20% of total CPU shipments. Even if they aren't suffering from any supply-chain problems, even gaining 1% of the market would require that they increase production by 5%, getting a majority of the market would require that they increase their production by 150%. That's not normally very easy.
How about an 89% share? An 88% share? An 87% share? At what level is it reasonable, and how exactly do you determine that?
It varies from market to market. The test is whether you can behave as if you had 100%. For example, if you put up your prices, do you lose market share? There are a number of tests that regulators perform to determine whether your market share is distorting the market. In a market with no barriers to entry and high visibility for new entrants, even a 99% market share would not be likely to need regulating. A market with strong network effects and high barriers to entry may start to see distortion as soon as one player is bigger than the second-largest player.
You're assuming that the Chinese data collection incorporates high security. When you're talking about spying, it's not a question of who gets your data, it's who gets your data first. I probably don't have anything to worry about from any government knowing the contents of my devices, but I might be more worried about various organised crime syndicates knowing things that can be easily inferred from these devices (e.g. when my house is going to be unoccupied). I'd trust the NSA or GCHQ to keep the data that they've stolen about me secret more than I would trust their Russian or Chinese counterparts.
As an individual, there's little that China can do to me - I've been to China once and I'm not inclined to repeat that. But as a corporate user, a Chinese company having access to all of my work-related emails and access to any work-related files on my device could have a significant impact on my employer's ability to remain competitive (production is much cheaper than R&D), which would impact their ability to pay me.
British companies used to be warned not to discuss work on Air France flights because the French intelligence agencies would put microphones in the seats and pass on any commercially-sensitive information they picked up to French companies. China still has a very close relationship between corporations and intelligence. The NSA is far less likely to pass on anything that they learn to US-based competitors.
The Tesco own-brand ones are called 'Cheese Food Slices'. I'm quite surprised that they're allowed to get away with calling them food...
It's not nearly as closely tied as it used to be. Modern flash uses multi-level cells, so each cell stores 2 or more bits, but is also very unreliable so a lot of cells are used for error correcting codes. The amount of redundancy that you need for error correction dominates the useable capacity of modern flash. Cheap flash storage typically uses the same chips as more expensive flash storage but doesn't reserve as much space for error correction. The probability of an error is the same, the probability of a non-correctible error is a lot higher.
I don't mind the lack of a mouse so much - it's a small-screen device and the touchscreen is usually better in this form factor. It does bug me that the 'Pro' keyboard has an emoji button but no escape key though.
I suspect that the argument for not supporting mice is that they don't want people doing crappy ports of desktop apps that require a mouse and then getting complaints from users that they don't work well with touchscreens. Apple's decision to ship one-button mouse worked out well for them in the era of resistive touchscreens, where a UI designed for a single-button mouse worked well but using Windows on the same devices was painful (you typically had a button in a floating window that you had to tap to make the next mouse click a right button).
And no, asking slashdot to bend over backwards because iPads are doing the wrong thing
You mean the wrong thing where they parse a page that sets its encoding to UTF-8 in both meta tags and HTTP headers, send a response with HTTP headers indicating that it contains UTF-8 data, and are surprised when the server then incorrectly handles UTF-8 data?
Slashdot advertises UTF-8 support in meta tags in this page and in the HTTP headers. When the iOS device submits the form, it also sets the content type to UTF-8. Slashdot then interprets this as some random 8-bit codepage. If Slashdot advertised ASCII support, then iOS would use ASCII for form submission. Advertising that you accept UTF-8 and then breaking on any non-ASCII characters is just plain broken.
Soylent News' fork of Slashcode fixed these issues ages ago.
There are all sorts of interesting things here. In the UK, a lot of supermarkets sell milk as a loss leader: many people think about the price of milk as a proxy for the price of everything, so if milk is cheaper in your shop then they assume that everything is cheaper. They also often have very short-term supply contracts with farms and will buy a week in advance and shift suppliers to keep their costs low. Farms often find that no one will buy their milk in a given week for above the production cost, but they sell it for below cost because they have no marginal cost from producing milk and their fixed costs are the same whether they sell or don't (if they sell, they go out of business eventually, if they don't sell then they go out of business soon). This then means that there isn't a contraction in supply (at least, in the short term), so other farms can't raise their prices. The supermarkets form an oligopsony that keeps the price very low.
As you say, if you can produce yoghurt some of your milk, you can keep it for a few weeks and watch the prices fluctuate, selling it only when the price goes above the cost of production. I'd expect that to keep the price of milk lower, but it may be that in the Netherlands there either isn't the same oligopsony of supermarkets, there isn't the trend of using milk as a loss leader, or there so many farms are betting on yoghurt that there's a supply shortage for milk. It may even be that yoghurt is filling the same loss-leader niche as milk in the UK.
The tens of thousands of people that voted and didn't see the stickers on their pages?
You're misunderstanding the attack. Person A votes. Person A puts an 'I voted' sticker on their account. Person B has not voted. Person B does not put a sticker. Person C, who votes for the same party as Person A goes to Person A's Facebook page, but does not see an 'I've voted' sticker. Person D, who votes the same way as Person B, goes to Person B's Facebook page and does see one even though B didn't post it. There's now an increased probability that Person D will vote and person C won't. If you pick the people in marginal constituencies, there's a good chance you can swing the election. If Person C notices the sticker is missing, then Facebook just blames it on delays in updates propagating between the system. If Person D notices a spurious sticker, then Facebook blames it on a glitch putting stickers on accounts that shouldn't have had them.
That's assuming that someone would even bother to check - when was the last time you saw a sticker on someone's account and checked with them that it was actually something that they'd put there? When was the last time you checked that other people could see all of the stickers on your account?
This kind of manipulation is hard to spot, but has a measurable impact. Most of the time, Facebook uses it to convince people to buy certain products, it just seems a little bit more sinister when those products are political parties.
Then there'd be a huge backlash against FB and it'd never, ever happen again.
Maybe. If people noticed. Remind me, how much backlash has there been against Facebook for the last half dozen times they were caught testing psychological manipulation techniques?
I think your "except two things" are nearly impossible for MS to fix though... Are you saying you would be happy with another key combo sending SIGINT in a terminal (for example) or you would only be happy with an Apple-key style situation across the OS? Or are you saying that when the time comes it's going to be a lump situation? :-)
I hadn't thought of that, but having windows-key-is-control in the terminal would fix that annoyance for me. Oh, and having the WSL terminal support ^T for SIGINFO (the one of the big things I miss going from FreeBSD to Linux).
They've done some really nice work on the terminal subsystem. If you run a Konsole or the GNOME terminal (or your favourite X11 terminal emulator) in WSL, you can then run cmd.exe or powershell.exe and run Windows commands. It doesn't quite work properly, because the traditional Windows terminal handles editing, history, and autocompletion (the shell doesn't), so all of these are broken, but hopefully they'll be fixed at some point soon. At the very least, I can stay in konsole and just pop into cmd briefly to run Windows builds.
How would you notice it? Would you notice if your friends didn't display 'I voted' stickers on Facebook? Or if they showed up a day late, after the election? That kind of thing could be easily explained as a propagation glitch and would be really hard for anyone to determine was directed at a specific group. Remember, they'd only be targeting people with a specific affiliation in selected areas where their decision to vote or not to vote would impact the results. That's a fairly small proportion of the electorate. And you wouldn't do it to everyone in that group, only to enough to affect the outcome. Unless you get everyone to record when they've seen the badges, it's basically impossible to detect.
You're assuming that most people use their computers to get real work done. Particularly for home computers, this is not really true.
Don't fool yourself into thinking that that face detection is anything more than a GPGPU program.
Nope, on OMAP SoCs (and quite a few others) it's dedicated hardware in a separate fixed-function logic block.
Ditto for any "AI", which is just a bunch of matrix multiplications of signals with state matrices, and is almost, but not quite, entirely unlike real natural fleshy neural networks.
Except that AI workloads typically have much higher locality of reference than GPU ones, so accelerators have totally different memory controller designs (memory + caches, rather than streaming).
Besides: Said face detection is a perfect example of exactly this useless-gimmick-adding behavior I criticized in my original comment. So thank you for that. :)
The face detection block is usually used with the camera to auto-focus on the part of the image that contains a face, rather than the background, even if the face isn't in the centre of the picture. I think most people that use their phone's camera find that a useful feature.