Would you pay more for hardware with AI capabilities?

[bug]

I haven't. I've seen some crazy shit advice posted from asking for advice from generative AI. The last one I read was several people asked what was the best way to pass a kidney stone quickly and the advice was to drink plenty of urine.

Google AI Makes Yet Another Blunder, Suggests User To Drink Urine To Pass Kidney Stones Quickly

Earlier this year, Google's Gemini AI image generation tool faced significant criticism after producing historically inaccurate and offensive images, sparking a global outcry.

news.abplive.com

Oh, it does. Hallucination is a real problem. But that doesn't mean it can't be useful.

 

[Caring1]

Never used Ai and don't want to start now and contribute to the robot overlords in the future.

 

[Event Horizon]

I do prefer local AI over relying on the cloud but my answer is still no. I treat it like ARGB: avoid it if I can and tolerate it if I have no choice.

 

[AusWolf]

You mean you've never used ChatGPT?

No, I haven't. What for when I have the good old Google search as my home page? Relying on an AI-generated answer instead of looking up the proper info in a proper source is lazy and careless.

Also, I thought ChatGPT ran on a server, not your home PC.

 

[..0]

the ai shit can go to hell.

 

[bug]

No, I haven't. What for when I have the good old Google search as my home page? Relying on an AI-generated answer instead of looking up the proper info in a proper source is lazy and careless.

Also, I thought ChatGPT ran on a server, not your home PC.

You asked for an example where AI makes your life better. And while ChatGPT doesn't rid you of proper web searches, it can shorten the time it takes you to google the roght things.

Also, bonus points far calling something you never used "lazy and careless".

 

[iameatingjam]

I don't think so. Its just not relevant for me. The only way is if games start taking advantage of that hardware.

 

[JWNoctis]

Considering that what is currently called (and hyped as) AI mostly consisted of relatively uncomplicated calculation over huge amount of data, that existing SIMD extensions of mostly any architecture do the same calculation just fine, to the point that all of them - even a RPi 4 - I've seen so far are memory capacity or bandwidth bound, I -think- dedicated AI coprocessor is mostly a marketing gimmick right now, except maybe for reasons of power efficiency.

That's assuming there is a good general application for them running local, a big, and particularly curly, question mark.

Right now they are fun to experiment (read: fool around) with, and the hallucinations are funnier, sure.

 

[lilhasselhoffer]

You asked for an example where AI makes your life better. And while ChatGPT doesn't rid you of proper web searches, it can shorten the time it takes you to google the roght things.

Also, bonus points far calling something you never used "lazy and careless".

So....let me get this straight. You acknowledge hallucination....and that would be a problem for any search. IE, I could ask an AI based on current LLM to create a synopsis of the Greek and Mandarin war of 1812...and it could spit me out 5 pages of "researched" data about that fiction. You then say that LLMs could be used to improve your search function. It's a bit faster after you spend huge amounts of data to build the model and train it...so I save a few milliseconds on my once weekly google searches for potentially getting back garbage....yay...


Are you aware of the simple idea of consistent messaging...because this isn't a 200 page thread 20 pages apart...this is the third page on a 3 page thread where you literally tell me the function of AI is a problem, then that the benefit lies with said problem. I'm scratching my head to try and follow through here...but you really don't think about the whole picture whenever you talk, do you? If I'm missing something help me....because I'm looking at this as someone who understands puzzle pieces....but cannot put them together as part of a complete puzzle.


Let me also offer you some extras.
It is lazy to resort to an AI to do research.
It is careless to trust an AI to deliver information which is factual.
It will take an AI radically different from our current LLMs, which can fit into modern hardware, to make either of these issues non-deal breakers for anyone to be able to trust.

 

[8086]

I do not want it in my machine or for any machine I own. But instead I want a dumb client as we have now that can connect and disconnect form an AI cloud server at my will and for my needs.

 

[bug]

@lilhasselhoffer Yes, it's an imperfect tool. And yet, it can be useful.

I've used it to cut the time needed to put some presentations together by more than half. I did know the subject, so I could tell where it got things wrong. In those cases I would turn to google and figure things out on my own. And was the free tier of ChatGPT and Gemini (yes, I pit them against each other), the paid-for models are even better.

 

[Psychoholic]

I go out of my way to disable Ai features on anything I own.
That said, if there is/was something about Ai hardware that made it more powerful or efficient in tasks other than Ai, i'd be all fo that.

....I'd still disable Ai features in software

 

[bug]

I do not want it in my machine or for any machine I own. But instead I want a dumb client as we have now that can connect and disconnect form an AI cloud server at my will and for my needs.

It can be useful on local, too.

Like if you were writing a scientific paper, had a bunch of paid-for articles around that you wouldn't upload to the cloud. You can have the local AI scan those docs and then you can query for whatever.

(Doesn't mean that you want AI locally or even that everyone wants it. Just that it can be preferable, under some circumstances.)

 

[lilhasselhoffer]

@lilhasselhoffer Yes, it's an imperfect tool. And yet, it can be useful.

I've used it to cut the time needed to put some presentations together by more than half. I did know the subject, so I could tell where it got things wrong. In those cases I would turn to google and figure things out on my own. And was the free tier of ChatGPT and Gemini (yes, I pit them against each other), the paid-for models are even better.

Did you maybe misread the thread...or the responses? I say this because you agree with me...then want to insist that you're willing to pay extra for a known broken tool.

I offer you this. I go into Home Depot, and buy a hardened steel shovel. It's specially hardened to resist damages, so if you hit a rock it'll shatter. You are telling me you would buy that tool, and then go digging around in the dirt...because you think you know there aren't any rocks. Are you insane, or do you just have too much money?



Let me close here with a reality check. "Anecdotally I sped up busy work with AI, that I then had to check for accuracy" is not a good reason for AI. It's a fantastic reason that the I in AI is something that humanity should be proud of...because many centuries ago people discovered the same trick you did with AI. They called it apprenticeships, or internships, or any one of a various amount of other things where a skilled person delegates simple tasks to a less skilled person, who then has the check them over. Congrats...you've managed to pay for an intern, who can also just start to hallucinate data. It's the wrong tool, applied by the wrong people, in an effort to remove low skilled labor. Yay.

 

[bug]

Did you maybe misread the thread...or the responses? I say this because you agree with me...then want to insist that you're willing to pay extra for a known broken tool.

Maybe this is the misunderstanding. I have already said I wouldn't pay for what AI offers today.

My later replies were to @AusWolf who asked for proof "AI that makes my life 0.001% better". To that, I say it does.

 

[AusWolf]

You asked for an example where AI makes your life better. And while ChatGPT doesn't rid you of proper web searches, it can shorten the time it takes you to google the roght things.

Also, bonus points far calling something you never used "lazy and careless".

Typing a question into ChatGPT takes the same amount of time as typing it into Google, or even asking it here. The difference is that Google gives me multiple results which makes it easier to fact-check. ChatGPT gives me only one answer which may or may not be correct. Being fed inaccurate data doesn't make my life better. Easier, probably, but not better. If it takes me a bit more time to look up proper sources to find the proper information, then so be it. Not being misinformed is worth the extra time and effort.

TL,DW: ChatGPT can only do an astrophysics exam with the results of an average student - which is logical, considering that it works with average data from the internet, and not the most accurate one.

I haven't used ChatGPT because I've done my research on it and concluded that it's not useful to me.

Did you maybe misread the thread...or the responses? I say this because you agree with me...then want to insist that you're willing to pay extra for a known broken tool.

I offer you this. I go into Home Depot, and buy a hardened steel shovel. It's specially hardened to resist damages, so if you hit a rock it'll shatter. You are telling me you would buy that tool, and then go digging around in the dirt...because you think you know there aren't any rocks. Are you insane, or do you just have too much money?

I'm gonna use another example. If I go to Home Depot and ask the attendant which tool he recommends, he'll give me the best one to his knowledge and experience as an employee working for Home Depot, dealing with customers like me every day. If I ask ChatGPT the same question, it will give me an answer that an average human on the internet would. Which answer has a higher chance of being the one I'm looking for?

 

[lilhasselhoffer]

Maybe this is the misunderstanding. I have already said I wouldn't pay for what AI offers today.

My later replies were to @AusWolf who asked for proof "AI that makes my life 0.001% better". To that, I say it does.

I'll admit to my crazy then. As I'm reading this proving that AI makes things better justifies its costs...which is where I'm going to choose to die. Fair point, and thank you for the clarification and opportunity to respond.

Typing a question into ChatGPT takes the same amount of time as typing it into Google, or even asking it here. The difference is that Google gives me multiple results which makes it easier to fact-check. ChatGPT gives me only one answer which may or may not be correct. Being fed inaccurate data doesn't make my life better. Easier, probably, but not better. If it takes me a bit more time to look up proper sources to find the proper information, then so be it. Not being misinformed is worth the extra time and effort.

TL,DW: ChatGPT can only do an astrophysics exam with the results of an average student - which is logical, considering that it works with average data from the internet, and not the most accurate one.

I haven't used ChatGPT because I've done my research on it and concluded that it's not useful to me.


I'm gonna use another example. If I go to Home Depot and ask the attendant which tool he recommends, he'll give me the best one to his knowledge and experience as an employee working for Home Depot, dealing with customers like me every day. If I ask ChatGPT the same question, it will give me an answer that an average human on the internet would. Which answer has a higher chance of being the one I'm looking for?

I....have to laugh. Most people at Home Depot know less than the average fast food employee....because that's literally how they're competing for labor.

I'd trust the average person more than I'd trust the average Home Depot employee...because after seeing the qualifications I believe their average is lower than human averages. Not to offend, when I worked fast food my response to you would be that I knew less about what I served than the average person too. Proximity to a thing does not impact knowledge...but I still love that Subway has 2 types of bread that basically look like a bread stick until they get defrosted and proof.


-I know this is a tangent, so let me get back on topic. Do you think that this is a concern for companies who might try to build AI into their self serve kiosks? I'd put money on the assumption that if they could get a reasonably human voice to recommend things they'd love a GPU+AI combo to try to replace people.-

 

[AusWolf]

I....have to laugh. Most people at Home Depot know less than the average fast food employee....because that's literally how they're competing for labor.

I'd trust the average person more than I'd trust the average Home Depot employee...because after seeing the qualifications I believe their average is lower than human averages. Not to offend, when I worked fast food my response to you would be that I knew less about what I served than the average person too. Proximity to a thing does not impact knowledge...but I still love that Subway has 2 types of bread that basically look like a bread stick until they get defrosted and proof.

Fair enough, I used a bad example. But still, it's better to know what several people recommend, and selecting the answer that best suits your needs, instead of just the average.

Edit: Here's a graphical representation of what I mean:

 

[crlogic]

I think everyone else here has missed the real question. I believe Wizard is asking if we'd pay more for a device with an NPU.

I think the word AI has made everyone think of LLMs and generating images whereas NPUs like NVIDIAs tensor cores or Apples Neural Engine enable a lot of great features I use daily

Or potentially not so great features like Windows Recall that was just announced...

Have I got this right @W1zzard ?

The only thing those cores are really used for are to assist with offline voice recognition and to speed up face-unlock. I might be ignorant of other uses but those are the only headline features that get any coverage/attention.

Apples Neural Engine NPU is responsible for a whole loooooad of stuff. The things you mentioned as well as (and more),

Computer Vision for face, object, landmarks and nature recognition in the photos app so it can categorize faces, allow you to search for key words, or learn more about a plant or animal you took a photo off. Object detection in the photos app where you can pluck the subject out of the image. And allow you to search for all of this stuff from Spotlight

Live Text which provides OCR in Camera and Photos, which allows you to copy handwriting or text or follow links, phone numbers or addresses
I often use that were text isn't selectable in an app or website, but if I take a screenshot I can.

Video effects like background blurring, centre stage and noise reduction

People occlusion and motion tracking in augmented reality, hugely important on something like Vision Pro

Photographic Styles in the Camera app, Memories curation and stylistic effects in Photos, personalized recommendations like wallpaper suggestions, VoiceOver image captioning, finding image duplicates in Photos, etc. etc etc

This might have been an interesting question five years ago.

Today it's a joke. Apple put ML cores in their phones with the iPhone X/iPhone 8 series.

Back in 2017. So no, I didn't bother submitting a vote in this poll. But it might have been fun before the pandemic.

Many TPU discussions are several years behind the times. And a lot of AI functions are already here without any fanfare. You ever get a recent fraud alert from your credit card issuer or bank? That's AI in IRL action.

You use an M2 Pro Mac, so you're using Apples "hardware with AI capabilities" everyday and may not even know it. And that's not a bad thing, it means its working efficiently

This is all stuff that could be brute forced on a CPU, but it's slow and ineffecient. Just like trying to render an image without GPU acceleration or enabling ray tracing without RT cores

 

[AusWolf]

I think everyone else here has missed the real question. I believe Wizard is asking if we'd pay more for a device with an NPU.

I think the word AI has made everyone think of LLMs and generating images whereas NPUs like NVIDIAs tensor cores or Apples Neural Engine enable a lot of great features I use daily

Or potentially not so great features like Windows Recall that was just announced...

Have I got this right @W1zzard ?

I agree, that's the question here - and my answer is no. The only ever interesting (but still not very useful) home application of AI that I've seen is Nvidia Canvas.

Apples Neural Engine NPU is responsible for a whole loooooad of stuff. The things you mentioned as well as (and more),

Computer Vision for face, object, landmarks and nature recognition in the photos app so it can categorize faces, allow you to search for key words, or learn more about a plant or animal you took a photo off. Object detection in the photos app where you can pluck the subject out of the image. And allow you to search for all of this stuff from Spotlight

That sounds useful if you can't be bothered to organise your photos into folders, which I think you should.

Video effects like background blurring, centre stage and noise reduction

People occlusion and motion tracking in augmented reality, hugely important on something like Vision Pro

Photographic Styles in the Camera app, Memories curation and stylistic effects in Photos, personalized recommendations like wallpaper suggestions, VoiceOver image captioning, finding image duplicates in Photos, etc. etc etc

This stuff has existed way before AI.

 

[crlogic]

The only ever interesting (but still not very useful) home application of AI that I've seen is Nvidia Canvas

How about DLSS?

That sounds useful if you can't be bothered to organise your photos into folders, which I think you should.

I find it most useful for automatically creating albums for individuals it recognizes. And even asking if you want to share photos you’ve taken with the subject. Or if I want to find all the photos of cars I’ve taken, as an example. These are features Google Photos has had for a long time, but it wasn’t local and it took time for Google to process all that info on their servers

This stuff has existed way before AI.

You’re right about that, just like my Google Photos example. But..

This is all stuff that could be brute forced on a CPU, but it's slow and ineffecient. Just like trying to render an image without GPU acceleration or enabling ray tracing without RT cores

As well as all this processing happening locally, instantly and (in the case of video effects) at much higher quality, all with no hit to system performance or battery life. That’s a huge benefit

I wouldn’t own a device these days without an NPU

 

[AusWolf]

How about DLSS?

How about FSR? You're free to bring up the image quality argument, but I don't think there's that much of a difference, personally. In short, I don't need DLSS. The fact that I only ever rely on upscaling if it's absolutely impossible to get decent frame rates without it is a different matter altogether.

I find it most useful for automatically creating albums for individuals it recognizes. And even asking if you want to share photos you’ve taken with the subject. Or if I want to find all the photos of cars I’ve taken, as an example. These are features Google Photos has had for a long time, but it wasn’t local and it took time for Google to process all that info on their servers

Fair enough. As for me, I have my photos stored on multiple drivers in separate folders arranged by theme, so I don't need albums. I don't store them in the cloud, either.

 

[crlogic]

How about FSR? You're free to bring up the image quality argument, but I don't think there's that much of a difference, personally. In short, I don't need DLSS. The fact that I only ever rely on upscaling if it's absolutely impossible to get decent frame rates without it is a different matter altogether.


Fair enough. As for me, I have my photos stored on multiple drivers in separate folders arranged by theme, so I don't need albums. I don't store them in the cloud, either.

Your arguments are fair enough as well. I’ll admit too that most of those use cases I mentioned are most useful on a mobile device.

There’s not much an NPU can offer in the desktop or laptop so far (even on Mac), but everyone won’t have a choice but to be paying for one pretty soon. It’s only Intel Desktop AMDs GPUs left without one and we already know they’re next. And I bet there will be a new iteration of FSR that’s going to run on it. Considering FSR is already so good compared to DLSS, it may even surpass it!

 

[AusWolf]

Your arguments are fair enough as well. I’ll admit too that most of those use cases I mentioned are most useful on a mobile device.

There’s not much an NPU can offer in the desktop or laptop so far (even on Mac), but everyone won’t have a choice but to be paying for one pretty soon. It’s only Intel Desktop AMDs GPUs left without one and we already know they’re next. And I bet there will be a new iteration of FSR that’s going to run on it. Considering FSR is already so good compared to DLSS, it may even surpass it!

RDNA 3 has AI units, as far as I know, although their use case is non-existent.

If they give me a functionality that I actually find useful, I'll have it. Until then, it's only marketing gimmick in my eyes.

 

[bug]

Typing a question into ChatGPT takes the same amount of time as typing it into Google, or even asking it here. The difference is that Google gives me multiple results which makes it easier to fact-check. ChatGPT gives me only one answer which may or may not be correct. Being fed inaccurate data doesn't make my life better. Easier, probably, but not better. If it takes me a bit more time to look up proper sources to find the proper information, then so be it. Not being misinformed is worth the extra time and effort.

TL,DW: ChatGPT can only do an astrophysics exam with the results of an average student - which is logical, considering that it works with average data from the internet, and not the most accurate one.

I haven't used ChatGPT because I've done my research on it and concluded that it's not useful to me.


I'm gonna use another example. If I go to Home Depot and ask the attendant which tool he recommends, he'll give me the best one to his knowledge and experience as an employee working for Home Depot, dealing with customers like me every day. If I ask ChatGPT the same question, it will give me an answer that an average human on the internet would. Which answer has a higher chance of being the one I'm looking for?

It's like you keep explaining how Achilles will never be able to catch up with the turtle, while Achilles looks back at the turtle from the finish line already.