Will Google's Tensor Chip Spell Trouble for 5G? (pcmag.com) 140
Google's Pixel 6 phone will be powered by a Tensor processor which PCMag UK believes is "clearly designed to accelerate machine learning and AI." But does it have bigger implications?
Tensor is a signpost, not a destination. Google has never sold huge numbers of Pixel phones and isn't signaling a change in strategy there. Rather, it's saying that it would like Android as a whole to shift toward more on-device processing for AI and ML. That could give a big boost to Google's two core businesses, advertising and data. It could also create problems for the future of 5G...
The more your phone can handle its own ML and AI, the less it needs the cloud. For example, a Tensor-enabled phone could potentially analyze your photos and share data locally with on-device advertising APIs, letting Google proclaim that its cloud services never access your raw data. That would help bridge the gap between the privacy you want and the targeted ads Google needs to survive. But a lot of consumer 5G app ideas assume your phone will offload processing or rendering onto the network.
Phones that are mostly self-sufficient aren't going to need high-bandwidth, low-latency networks or "mobile edge computing." Sure, the rise of more Tensor-like ML-focused chips would take pressure off the carriers' still-shaky 5G builds, but it'll also keep raising the question of why consumers need 5G in the first place. That could reduce carriers' willingness to keep investing in their consumer 5G networks.
Although the first sence of the article's last paragraph adds, "Maybe I'm overstating the issue here..."
The more your phone can handle its own ML and AI, the less it needs the cloud. For example, a Tensor-enabled phone could potentially analyze your photos and share data locally with on-device advertising APIs, letting Google proclaim that its cloud services never access your raw data. That would help bridge the gap between the privacy you want and the targeted ads Google needs to survive. But a lot of consumer 5G app ideas assume your phone will offload processing or rendering onto the network.
Phones that are mostly self-sufficient aren't going to need high-bandwidth, low-latency networks or "mobile edge computing." Sure, the rise of more Tensor-like ML-focused chips would take pressure off the carriers' still-shaky 5G builds, but it'll also keep raising the question of why consumers need 5G in the first place. That could reduce carriers' willingness to keep investing in their consumer 5G networks.
Although the first sence of the article's last paragraph adds, "Maybe I'm overstating the issue here..."
my phone decides what gets shared? (Score:5, Insightful)
I don't want AI on my phone. I want it to do what I tell it, when I tell it and no more. I don't want Lord Google geting a piece of everything, everywhere everytime I do something. (which is exactly why my phone isn't attached to my hip 24x7).
Re: (Score:2, Interesting)
P.S. almost all the content you interact with allows google to identify you and it is highly unlikely your dns isn't tied to that. It's also almost 100% guaranteed your ISP is data sharing with Google. Then there is facebook and amazon. You might want to just unplug and avoid using the Internet. Also avoid cities and public places in general or wear a big sombrero to avoid the countless surveillance cameras using Google, MS, or Amazon powered AI.
Re: (Score:3)
P.S. almost all the content you interact with allows google to identify you and it is highly unlikely your dns isn't tied to that. It's also almost 100% guaranteed your ISP is data sharing with Google. Then there is facebook and amazon. You might want to just unplug and avoid using the Internet. Also avoid cities and public places in general or wear a big sombrero to avoid the countless surveillance cameras using Google, MS, or Amazon powered AI.
This ship is taking on water thru shaft seal so no big deal if I drill a few holes below the water line.
Re: (Score:3)
And then people wonder why ludditism is on the rise. At some point you look at all this and the question you have to ask is "Is this really worth the trouble?"
Re: (Score:2)
This is just paranoia, and also highly illegal in the EU where GDPR rules do not allow this kind of data collection or sharing without explicit and informed opt-in permission.
Show us some evidence and I'll personally file the GDPR complaint, then Google will get hit with a fine for billions of Euros.
Re: my phone decides what gets shared? (Score:3)
Re: my phone decides what gets shared? (Score:2)
Re:my phone decides what gets shared? (Score:4, Informative)
AI made its way onto the major phone operating systems a long time ago, and as soon as they did that, the mobile SoC manufacturers quickly put tensor accelerators on them.
Apple, Qualcomm, nVidia, and now Google all manufacture mobile ARM CPUs with tensor accelerators on them so that your phone can continue doing really neat shit that you don't understand with the "AI on them"
Re: (Score:2)
so that your phone can continue doing really neat shit that you don't understand
Why would I want this "really neat shit"? Seriously.I can't think of a single thing that I use my phone for where AI would be useful. The interface for the phone is too limited in comparison to a laptop or desktop.
So why would I want all of this nifty AI stuff on my primary communications device?
Re: (Score:2)
It categorizes them into your pets, or your friends, or your family, or by person.
It handles the biometrics that unlock your phone.
If you use voice commands, it handles those.
If you have goofy filters on your video calls, it handles those
It's everywhere. If you don't use any of that shit, good for you. But you're not the norm, and it's there whether you want it or not, and has been for a very long time.
The one thing the AI is not doing? Sp
Re: (Score:2)
It makes your pictures look better, synthesizing HDR imagery.
Potentially useful; however, I would like access to the raw data, not just filtered data. What happens when the algorithms are tuned to remove certain items so you can never take a picture of them? Take a video of some police brutality and all you see is a criminal moving in weird ways and suddenly spouting blood.
It categorizes them into your pets, or your friends, or your family, or by person.
I don't want anything categorized for me. What is it with external organizations trying to do internal organization for me? Seems creepy and suspicious.
It handles the biometrics that unlock your phone.
Do not want. Police can use me to unlock my p
Re: (Score:2)
Potentially useful; however, I would like access to the raw data, not just filtered data. What happens when the algorithms are tuned to remove certain items so you can never take a picture of them? Take a video of some police brutality and all you see is a criminal moving in weird ways and suddenly spouting blood.
Well, at the very least- you can turn it off if you don't trust it.
I don't want anything categorized for me. What is it with external organizations trying to do internal organization for me? Seems creepy and suspicious.
I like it.
It made an album with my dog in it. It's neat. I have tens of thousands of photos, and their only organizational information is their timestamp and their GPS tag. I love that I can pull them up based on subject.
As long as it's done on the phone, and without uploading to a server, I think that's great.
Do not want. Police can use me to unlock my phone rather than requiring my cooperation.
Eh.
Fingerprint? Sure.
Face ID? No. Those are advanced enough to require attention, and someone forcing your eyes open isn't gonna
Re: (Score:3)
Did you even read the summary? The whole point is that your phone won't share anything, it will do all the processing locally.
Google already does this for some things, such as speech recognition. It enables faster responses to commands, and real-time subtitles in any media.
Re: (Score:2)
The whole point is that your phone won't share anything, it will do all the processing locally.
Yes, this is a good thing! For all Apple's touting of the iPhone's AI processing capability Siri, their virtual assistant, can't even do the basic phone functionality without being connected to the internet, not even the simple stuff that integrated vehicle systems from 10 years ago could handle like "Call XYZ". All that data has to be sent to Apple's servers (and even externally [theguardian.com]) for processing.
Re:you'd rather it be done in the cloud? (Score:2)
Overstating the issue, indeed (Score:5, Insightful)
There are a ton of other applications for the advantages of 5G: better games, collaboration, AR, and more. Video streaming in particular -- whether for video calls or media consumption -- wants 5G's improvements in spectral efficiency and multi-user support. Smarter phones do not substantially reduce the value proposition of 5G. In fact, eggs computing and the other things that writer worries about work almost exactly the same with LTE....
Re: (Score:2)
Autonomous cars and Tesla Dojo.
Re: (Score:2)
Sure, V2X and IoT are other use cases for new 5G features, but they're not as core to 5G adoption as what phones demand. I'm pretty sure that LTE and LAA provide enough capacity for all the sideband that V2X needs.
Re: (Score:2)
Depends upon how much data [youtu.be] is being sent.
Re: (Score:3)
This is actually my biggest fear with autonomous cars. It's not that they're worse drivers than humans, but it's that once we become reliant on this technology it will be so much easier to pull the plug on someone, bricking a car remotely.
So in the end it all boils down to better DRM for 'them', the right holders. And less property in your hands until you don't own anything any more but can only rent it. That seems
Re: (Score:3, Interesting)
That's exactly it, the large corporations have no objection to Communism at all, they just want to make sure THEY are the central planning committee.
Re: (Score:2)
Re: (Score:2)
Every business I've ever worked for wanted to be (and 1 successfully became) a large corporation.
Ergo, Isn't that all capitalists?
Take your meds before posting, next time.
Re: (Score:2)
Well, a large part of it is that you don't actually own anything, you're just allowed to use it. Many find that to one of the more objectionable parts. Actually go down the list and see how many of the boxes corporate America has attempted to check for themselves.
Re: (Score:2)
Cool-aid drinking moron.
Re: Overstating the issue, indeed (Score:4, Insightful)
Let innovators innovate. New platforms and infrastructure means new ideas and applications.
Just don't waste energy on 'relevant' ads because they never are. No matter how much you analyze me you won't know if I would actually find an ad interesting or not.
Re: (Score:3)
Just don't waste energy on 'relevant' ads because they never are.
Companies spend billions every year analyzing the effectiveness of different advertising strategies. I doubt if you know more than they do.
No matter how much you analyze me you won't know if I would actually find an ad interesting or not.
People who claim they are unaffected by ads are affected by them about as much as everyone else.
Re: Overstating the issue, indeed (Score:2)
Re: (Score:2)
There is oss sudoku on fdroid.
Re: (Score:2)
Just don't waste energy on 'relevant' ads because they never are.
Companies spend billions every year analyzing the effectiveness of different advertising strategies. I doubt if you know more than they do.
I just categorize them into irrelevant or obnoxious and the obnoxious ones tells me to NOT buy that or from that company.
But I run uBlock to stay sane. Printed ads in magazines - well, for them to end up in the obnoxious category just means that they would be either illegal or upsetting PETA long before I find them obnoxious.
TV ads - that's a reason why I don't watch TV much anymore. Several months between each time.
Targeted junk mail paper - that's getting recycled without being read. Even if there are cou
Re: (Score:2)
People who claim they are unaffected by ads are affected by them about as much as everyone else.
I'm affected by ads. They piss me off and I buy something I don't see advertised.
The closest advertising gets to inducing sales in my world is when someone pays to sponsor eBay results. And even then I'm going to sort price low to high and see if I can get exactly the same thing for less.
Re: (Score:2)
Companies spend billions every year analyzing the effectiveness of different advertising strategies. I doubt if you know more than they do.
It's not about knowing more: it's about what you do with that knowledge, and the fact that there are THREE parties involved, not two.
The companies "spending billions" do not, in fact, know any more than we do, because we and they BOTH know that targeted ads DO NOT WORK. The difference is, we aren't trying to SELL those targeted ads to the marketards that buy them. Google / FB / etc are.
I can put that in an even simpler form, if it helps: if you're trying to sell a food product, would you market it as "Taste
Re: (Score:2)
Quick, think of a soda. You lose your argument if your first three thoughts are Coke, Pepsi, or Mountain Dew.
Re: (Score:2)
Pommac.
Re: (Score:2)
You mean drug addict videos...
https://www.youtube.com/watch?... [youtube.com]
How about neither? (Score:2)
Cloud nor AI.
I want my phone to make phone calls, wake me and maybe send and receive text messages. Everything else should be entirely optional. If you can't produce that, I find me some other phone maker who can.
Re: (Score:3, Insightful)
I want my phone to make phone calls, wake me and maybe send and receive text messages.
You are not the target market, so your opinion is irrelevant. Go buy a $20 flip phone.
Re: (Score:2)
You know, I'd really pay some good money for a phone that comes without any un-uninstallable crapware, "features" you can disable in hardware and that allows me to take control over what it can and what it cannot do.
I am fairly sure that I'm not the only one who'd pay good money for that, without caring whether it has the latest and greatest hardware because my focus is on something different.
Re: (Score:2)
The AI in your phone will be able to do that, you'll be optional.
5G will be fine, until it's not enough (Score:2)
Nonsense (Score:5, Insightful)
People use their phones mostly for content consumption - artificial "Intelligence" and machine learning have nothing to do with that.
Google's main interest in those fields likely has more to do with siphoning up even more information about you, based on what you look at on your phone - information it can then sell to its customers. But they're going to tell you it has to do with analyzing your photos for your benefit, rather than for the benefit of its customers.
Re: (Score:3)
I know it's not your fault, the article is really bad. But still... I'm disappointed in the nerds here today.
Your rant about Google comes from a good place, but phones have had tensor accelerators for years now.
Unless you've got a flip phone, whatever phone you're using right now has an "AI Accelerator" so that it can run ML models quickly and with a low power cost. And
Re: (Score:2)
Unless you've got a flip phone, whatever phone you're using right now has an "AI Accelerator" so that it can run ML models quickly and with a low power cost. And it's doing it all the time.
Heaven forfend your phone not be able to spy on you when you only have the cellular bandwidth to send texts
Re: (Score:2)
And if you're that paranoid, just go buy one. For the rest of us, we like having the mobile computers on us capable of more tasks than "sending texts"
Re: (Score:2)
Heaven forfend your phone not be able to spy on you when you only have the cellular bandwidth to send texts
You have to go pretty far back to only be able to send texts.
Uh no. All I have to do is go to work, where I don't have any signal to speak of.
Re: (Score:2)
Seriously, why the fuck are you even commenting? Obviously none of this is about fuckers living on Walden Pond.
Re: (Score:2)
Wow, you can send texts with no signal? Neat trick.
I can occasionally send texts with almost no signal, like all cellphone users. You can get texts out when you can't successfully make a voice call, in many cases.
Seriously, why the fuck are you even commenting? Obviously none of this is about fuckers living on Walden Pond.
Well, clearly the market does not give two shits about me, if that's what you mean. There are fewer devices worth buying in my eyes every year.
Re: (Score:2)
I can occasionally send texts with almost no signal, like all cellphone users. You can get texts out when you can't successfully make a voice call, in many cases.
Allow me to quote you: "Uh no. All I have to do is go to work, where I don't have any signal to speak of."
No signal to speak of is not almost no signal.
But that's not really relevant.
As you so eloquently put it:
Well, clearly the market does not give two shits about me, if that's what you mean. There are fewer devices worth buying in my eyes every year.
Yup. I'm sorry the market doesn't care about you. Society as a whole has embraced the idea of having little mobile computers strapped to us. It had started doing so even before social media exploded and made those computers into fucking parasites that latch to our face and drain our brains.
The
Re: (Score:2)
People use their phones mostly for content consumption - artificial "Intelligence" and machine learning have nothing to do with that.
Google's main interest in those fields likely has more to do with siphoning up even more information about you, based on what you look at on your phone - information it can then sell to its customers. But they're going to tell you it has to do with analyzing your photos for your benefit, rather than for the benefit of its customers.
Exactly, Google sell you their phones in order to suck up data from you. Any AI in the phone will be mainly used for analysing your behaviour even more thoroughly and make the data more efficient to be sent back to mothership.
An efficient enough AI could analyse everything the camera receives, without waiting for you to take any photo or even switch to the camera app. The AI could be constantly listening to the mic and send everything you said, converted to text, back to Google mothership.
The real question... (Score:2)
Re: The real question... (Score:2)
Also ask the same question about Apple.
There's no answer aside from them being big.
Re: (Score:3)
What industry are you talking about? Apple primarily influences the cell phone industry through competition. When they come up with a feature before an Android device maker the Android phones come up with a competing feature if it's popular.
Google is much more like how Microsoft is with PCs. Despite (primarily) just making the software, they often dictate to OEMs what they will manufacture (and what they won't).
That's nothing compared to Google's power in advertising. All ad agencies must bend the knee to G
Re: The real question... (Score:2)
Same reason Microsoft has so much sway over the PC industry.
Re: (Score:2)
And fast food has so much sway over the waistline industry.
Re:The real question... (Score:5, Insightful)
The only news here, is that Google is going to be shipping a phone with Google silicon in it, as opposed to using other peoples. The AI shit is just some dumbfuck "journalist" googling the word Tensor and thinking the fact that the new Google SoC is called "Tensor" means that it has magical AI abilities, when in reality, it's just Yet Another ARM CPU With A Tensor Accelerator
Re: (Score:2)
It is different from the Qualcomm processor in that it's not a wretched piece of crap.
Seriously.
Re: (Score:2)
It's got a Qualcomm 865+ in it.
In benchmarks, it seems to perform somewhere between my iPad Pro 12.9 inch gen 2 (mp6j2ll/a, A10X, 2017), and my iPad Pro 11 inch gen 1 (mu16ll/a, A12X, 2018)
And honestly, the tablet is so much nicer than either of those in terms of out-of-the-box functionality, stylus, screen, and sound, that I haven't really wanted more from the processor. But then again, I can't tell the difference between my 2 iPad Pros either in that department.
Burning CPU j
Re: (Score:2)
The entire conversation was about neural network processing. Seriously.
It's a wretched piece of crap.
Re: (Score:2)
I'll limit my commentary to NN processing.
The QC Hexagon is capable of 15 TOPS, or about 4 more TOPS than the next highest contender on the market- the M1's NeuralEngine.
We appreciate your dumbfuck opinion though, really.
Re: (Score:2)
You only get that at 8 bit integer however.
The state of the art for 8 bit integer DNNs is, well, still something of a work in progress. You can take an FP32 model and just throw away bits, infer on FP16 hardware and you'll have about the same performance (if you see significant changes you've over fitted anyway). Going to 8 bit integer is a crap shoot, with emphasis on crap.
It's not just lost precision, it's lost range moving from essentially a logarithmic to linear scale. Your network might work, or it mig
Re: (Score:2)
You only get that at 8 bit integer however.
Eyeroll.
That's generally how TOPS is used. When they want to brag about their FP16/32 performance, they use the much more specific FLOPS.
The state of the art for 8 bit integer DNNs is, well, still something of a work in progress. You can take an FP32 model and just throw away bits, infer on FP16 hardware and you'll have about the same performance (if you see significant changes you've over fitted anyway). Going to 8 bit integer is a crap shoot, with emphasis on crap.
Work in progress? What the fuck are you talking about.
INT8 quantization is used all over the place in mobile. That's because tensor accelerators can pump out INT8 like fucking mad.
If you use an FP16/32 model in CoreML right now, you get GPU performance, and it's *far* below 11 TOPS.
I'm no NN expert or anything, but I've been playing with it since I got my Jetson in 201
Re: (Score:2)
Mate you're taking out of your arse.
Quantization is still a research topic. Some models, especially oversized ones solving easy problems can quantize well. If you want state of the art performance, you need luck or an expert working possibly for months. But please, tell me about all the easy problems you're running...
And ML engineered prefer core ML because it accelerates the ever living fuck out of FP16.
Oh and samsung bless their perverse little hearts refer to TOPS as 4 bit because they went all
Re: (Score:2)
Mate you're taking out of your arse.
Uh, right. Says the guy who's doing "deep learning and editing 4K video on my laptop"
You, you're real serious training or doing inference... on a laptop.
Quantization is still a research topic. Some models, especially oversized ones solving easy problems can quantize well. If you want state of the art performance, you need luck or an expert working possibly for months. But please, tell me about all the easy problems you're running...
What kind of stupid fucking statement is this? All of ML is still "a research topic"
Quantization is built into every ML library out there, with handy Python bindings for dumbfucks like you too busy trying to sound cool to learn a real language.
And ML engineered prefer core ML because it accelerates the ever living fuck out of FP16.
Everliving fuck?
Christ.
Ya, those Apple ML engineers. Big group of motherfuckers right there.
Accelerates the ev
5G Whether you like it or not (Score:4, Informative)
Re: (Score:2)
Re: (Score:2)
Patents (Score:3)
5G is patented up the wazoo, that seems like the biggest threat to it. Known and unknown patents.
And so (Score:2)
Interesting. Ten years ago, 3D video cards were a goner as high speed connections would turn computers and phones and consoles into dumb terminals streaming 3D game rendering as with movies.
Now chips are so powerful it goes back the other way, where it never really left.
Re: (Score:2)
Did GeForce Now never get off the ground?
The Article Is Absurd (Score:3)
The wireless hucksters, err carriers, have been promising that we will get true broadband via 5G and we won't need no stinkin' fiber. That was oversold of course, but it has a kernel of truth - with 5G you can start using wireless to get decent Internet service without having to be a hot spot to which you have access. It has nothing to do with Android phones and running AI in data centers, that is at most an edge case.
In Google's wettest wet dreams maybe... (Score:5, Interesting)
Google sells ones of millions of Pixel phones over their whole lifetime. They're not in the top 5 smartphone manufacturers and I doubt they even crack the top 10. Samsung and Xiaomi sell more of a single model than Google sells of all their Pixel models.
Google's Pixel phones get attention in the tech press because it's Google but they are in no way the market leader for Android phones. Carriers don't give a shit what Google puts in their phones.
That out of the way, who in the shit thought 5G's main use was going to be offloading processing of phones? Unless you've suffered some severe brain damage it's been plainly obvious for a decade and a half smartphones have only been getting more powerful over time. AI coprocessors (read GPGPUs with half and quarter precision floats) are in the same vein of improvements as better CPUs and GPUs. The only thing 5G has done is bring the cellular networking on par with WiFi in terms of performance. This means the network intensive things that used to require WiFi can get more mobile. But no one not suffering from brain damaged assumed that meant phones would somehow get less powerful.
So great article written by someone that apparently woke up from a twenty year coma and doesn't know much of anything about smartphones. The world needs ditch diggers too I guess.
Re: (Score:2)
The fact that the Google Tensor SoC has an AI accelerator isn't new or abnormal.
Qualcomm and Apple both do it as well, and have for years, so I have no fucking idea what the author is even going on about.
Tensor accelerators aren't a new thing in the mobile space. They do the ML work of a CPU+GPU in a fraction of the time and power.
Re: (Score:2)
Thank you. I wasn't sure if anyone else noticed that all this complaining is about a higher performance CPU being put into a phone.
Although the first sence of the article's last... (Score:3)
That makes no sentence at all.
Need for speed (Score:2)
As VR gets more and more into the picture so will bandwidth requirements. Also, gaming requires the low latency 5G brings. There's also the fact that nobody buys Pixel phones.
Anyway, when Apple releases the 8K per-eye VR headset in 2023, there will be a huge need for bandwidth. Especially because a large number of VR social and education apps will start coming online. A lecture hall in VR will take a large amount of bandwidth.
Re: (Score:2)
Anyway, when Apple releases the 8K per-eye VR headset in 2023, there will be a huge need for bandwidth. Especially because a large number of VR social and education apps will start coming online. A lecture hall in VR will take a large amount of bandwidth.
Online games never use much bandwidth when playing. It's way less than video streaming. Neither VR or display resolution meaningfully alters bandwidth consumption.
Reality and the telecoms (Score:4, Informative)
Reality and the telecoms are the real trouble for 5G. They can't seem to deploy enough towers to provide proper 4G coverage and we're to believe they'll somehow magically deploy the far greater number required for 5G? If they can't get fiber to your neighborhood, they won't be able to deploy 5G either. In testing, they weren't even able to properly cover an entire baseball stadium.
Meanwhile, they still can't get over their 20th century data caps, so all 5G does is let you burn your allocation in the first 30 seconds of the month.
Re: (Score:2)
They can't seem to deploy enough towers to provide proper 4G coverage
Of course they can. They just don't want to, because it lowers profit margins.
The first fatal flaw in this grand idea (Score:2)
The battery will be dead within minutes.
Smart phones already kill batteries quickly with todays use case.
Add some heavy AI processes and poof, you're battery is dead.
Give the world a battery that can run a smart phone, laptop or a car all day (18hr minimum and recyclable) and we have a new world.
Until then the New Green Deal is a utopian dream.
I think we will have nuclear fusion first.
We will need it to charge the batteries for the New Green Deal anyway.
Re: The first fatal flaw in this grand idea (Score:2)
Presumably that's the point of having a separate tensor chip to do the heavy lifting rather than do it all in software which would kill the battery as you say.
Don't Worry (Score:2)
As long as Android still lags while scrolling, it'll never be taken seriously.
Just the same old cycle (Score:4, Insightful)
Central computing -> device computing -> central computing -> ... forever. This has been going on for at least 50 years.
And every time the move is called a "revolution" and expected to make everything better. Stupid.
Re: (Score:2)
expected to make everything better
It kind of does. When you make a break through in some area it can fix what was previously a bottle neck. Then you make minor improvements on this until something else becomes a bottle neck. Then you make a break through in another area. And so forth...
Re: (Score:2)
For part of the application spectrum, yes. But too many people that understand absolutely nothing ("managers") jump on the current trend expecting things it cannot deliver. Every time. Learning nothing. Making the same mistakes over and over again.
The sane way would be to recognize some things are better done client-side, some things better on server-side and occasionally a small part of the things you do on computers shifts from one side to the other.
Wow (Score:3)
So this AI chip is going to be so good that instead of people downloading torrents it will author and render all of the new TV shows and movies people want to watch?
Market must decide, legislators must legislate (Score:2)
Let me preface by stating I neither support nor object this.
But the ultimate choice must remain to the user to use or not the device, and to install and agree or not to the ToS and EULA of the apps that leverage ML for directed ads. There are inumerous benefits to users from opting in to such technology, such as limiting data when and where data is more expensive (5G), or the simple fact app makers make their service cheaper by not having to process stuff on the cloud, or the very obvious fact of having cap
Re: (Score:2, Insightful)
Why not 'just a phone'? (Score:2)
I regard the ownership of of OS by Google and Apple as a serious potential problem too, means that they can get anything they 'want' without depending on 'apps' (which I hate
No, it won't. (Score:2)
Re: No, it won't. (Score:2)
Re: No, it won't. (Score:2)
What will happen is that the cloud will get more powerful AI chips, some new feature will come along that won't run decently even on the new pixel phones, and we're back to sending data to the cloud.
Bet zero effect on 5G.
Re: (Score:2)
Maybe? I'm not convinced.
The tech for training DNNs is improving too. While people are creating colossal models, the state of the art for shrinking them (distillation) is improving massively at the same time as phones can successfully run increasingly large models. Those combined vastly increase what you can do on the phone.
It doesn't take much before sending one 5MiB model to the phone is a lot cheaper than sending all the data back to the cloud. You'd by surprised what you can do on a phone these days wit
Re: (Score:2)
Well, what you say is true for the reality of 5G, like before but faster and more devices.
The fantasy of 5G however had people acting as if it would usher in a categorically different computing paradigm. That somehow when and *only* when 5G deploys all sorts of random sensors would have dedicated cellular modems and do vaguely great and novel things. The whole point was that we'd have absurdly high bandwidth and it's magically past the threshold for dumb devices to stream all their data to cloud providers,
Google has no shame (Score:2)
Is there any existing use of machine learning algorithms on your phone that is privacy-conscious and does NOT benefit the big data hoovers, either established big tech or startups with shiny demos and to-be-filled-databases? Besides image/video filters.
Stupid question (Score:2)
A smartphone's storage capacity would need to increase at least a thousand fold to match the capacity of the cloud-based solutions.
Good (Score:2)
Spell Trouble for 5G
Now can we please just switch back to AMPS?
what bullshit.. (Score:2)
We never needed 5G in the US (Score:2)
Re: (Score:2)
They're useful.
They do the work of the CPU+GPU while running an ML model using less power and less time.
Arguing that ML has no use on a phone is just pure dumbfuckery, as it's literally used all over the fucking place for features- face recognition, picture subject scanning, hell, even HDR image synthesis from the cameras these days.
ML is fucking cool. Doing it faster and with less power is more cooler. Apple has had great success with it, as wel
Re: (Score:2)
Quantum computing will always happen in the cloud (unless you can cool down your phone to -270C), hence 5G is saved?
No way, it needs a blockchain back end with an augmented reality GUI, with biometric authentication all offered as a service managed by shadow IT. Are you nuts?