Don't rent out that container ship yet: CIOs and biz buyers view AI PCs with some caution While vast swathes of the computer industry remain hellbent on inflating the AI PC hype bubble, the response from some analysts and biz consumers is milder as they sit on the side to see how the first generation tech shapes up. Gartner told us recently that it expects AI to this year generate revenue for the infrastructure kit …
OK, so it sat in the corner of the spare room for 6 years totally unused since it was replaced with a decent TV but it took me until quite recently to find a way to get rid of the bloody thing as it was too big to put in my car and take down the tip
Amazing what people will want if you tell them it's free, stuck it in sowshall meeja after someone said it would go and it was picked up the same day. Didn't even have to pay them to take it away!
I thought it was "Quantum computing" that would revolutionize the industry...
no, wait, it was "Blockchain"
hold on, no...that's not it, it was "the Metaverse..
no..."CLOUD" - that was the word!
or was it "5G"?
"Crypto"?
"IoT"?
The industry remains unrevolutionized
but the revolution WILL be televised (or live streamed)
I've just had my windows cleaned. They're nothing to do with Microsoft, these are the real deal, they're the things you look out of. The crew that came manage their work online -- appointments, invoicing etc. -- like other small businesses. But however up to date they are its still a couple of guys with ladders, squeeges, bottles and cloths. AI might assist in the design of better cloths or window glass that never needs cleaning but it has nothing to contribute to their work.
They're probably like most businesses. They have their processes and procedures. Computers can help with some of the work, especially paperwork, but it can't clean windows, fix water heaters, unclog toilets or do the 101 things that businesses do. For most its potentially a useful tool but not central to what business does. For everyday use its probably most useful to marketing (mainly to sell AI to each other.....).
I gave a presentation at the Computers and Writing conference some years ago titled "Boring Information", where I made precisely this point, at length, with a wide range of examples.
The C&W crowd tend to be highly technophiliac in a "gosh we live in an amazing wiz" sort of way, like characters in some Golden Age SF stories — "once again they marveled at the astonishing technology which greeted them every day" sort of thing. At the time, you might still be treated to C&W presentations that boiled down to "my students did blogs!". I was trying to explain that most IT was, while often important, quite mundane; and that included most writing that was done with computers, so maybe we should dial back the breathless enthusiasm a tad and focus on the things that might actually prove to be transformative.1
It's the only academic presentation I manged to squeeze in citations to Essays on Boredom and Modernity, a collection which won my affection right at the title.
1The following year, I proposed a paper for the Conference on College Communication and Composition which explained why soon undergraduate composition students would have access to computer-generated essays essentially indistinguishable from the sort of work most of them currently produced, and pointing out that this would make traditional composition classes pointless. It was turned down. I wonder who was right about that? Oh, it was me.
Nobody wanted those at first, yet within a few years, almost every mobile phone and smartphone had a built-in camera. The same will happen with AI. If you've been rocking an M1 Mac since late 2020, you already have NPUs waiting to be used*.
* Well, they're already used in a few applications.
The same may happen with NPUs. That doesn't mean all PCs will be running "AI" software (that is, DL models, whether they're transformers or diffusion or some other architecture, for purposes of tuning ambiguous inputs into questionable outputs). Mine won't, for example.
And, of course, for every example of an innovation (such as) which became pervasive, there are any number of counterexamples. In this forum we've already seen the 3D TV and "metaverse" boondoggles mentioned. For, say, automobiles, you have features such as airbags, which authorities decided to make pervasive by fiat; or adaptive cruise control, which automakers seem to assume is universally demanded and therefore has become at least very common — but then you also have touchscreens, which manufacturers leapt on but which are now seeing some backlash and may yet recede, like the scum-fouled tide they are. And then there are innovations which for a short time are very popular and then disappear because they're displaced by other technology: fax machines, answering machines, VCRs and DVRs... Or those which are heralded as some sort of technological Second Coming, only to largely vanish after a brief flare, like the Segway.
I thought AI meant "copying things off the internet"
No, it really very much does not.
I am not a fan of "generative AI" for many reasons, but your description wildly misunderstands how transformer models work. The output of a LLM is "copied off the Internet" in approximately the same way you're "copied off Neanderthals".
And an "AI PC" is, in some vague marketing fashion, one that contains one or more NPUs — processors optimized for low-precision FP operations on large matrices, basically — and associated RAM. The point of an AI PC is to use a local model, not just bounce queries off a cloud-hosted service. Again, not at all compelling from my point of view (though I might find other ML tasks for that hardware, depending on what catches my interest); but it does mean something, in a fuzzy way.
I can see a case for AI/Co-pilot with marketing types, it will help them send out even more BS than they normally can, and for analysing big data sets such as JWST data; but who will verify the output of that analysis? Another AI model? Code writing possibly but for the average person, day to day, I don’t, at the moment, see a usage case.
(Mind I’ve not the best forecaster to be honest, I spent quite some time stating that I couldn’t see a usage case for the 1st generation of tablets)
My employer recently bought Copilot subscriptions for everyone in IT to try it out and it's inability to interact with anything outside of it's own text window is really limiting what it can actually do. For example I tried to have Copilot find a picture of a cat and then send it to my cat enjoying coworker with the subject "Look at this cat". Turns out it can't include pictures and it can't send the message either.
In a dark dark world, there is even more need for optimism. So let me try: Modern automobiles, instead of using extremely dangerous touch panels requiring visual confirmation, could provide provide a voice operated interface. Lives saved! Safety increased! (PS, I drive a 1998 car without a touchscreen and would prefer to stay with it until either of dies or I move overseas).
OK, it's not a desktop, so probably off topic. As far as desktops go, the main advantage over 100%-cloud based apps is saving subscription fees and privacy. Both of those are anathema to making a quick buck out for the AI-as-a-service providers. A provider would like to provide a paid for subscription, gain access to monetizable data, AND have the user pay for the intensive computation resources (hardware and electricity).
The challenge is to provide alternatives to the major AI-as-a-service providers that gives users control and privacy. There is obviously a market for it, just as there is a market for servers running Linux. So there IS room for optimism!
> could provide provide a voice operated interface.
The 1980's called ,,,there was a British car that had vocal warnings, I remember the sales rep having one who would deliberately leave the lights on when stopping and opening the door so that he could say "shut up bitch" (it was a female voice)
OK, all input was button and switches but they were halfway there....
Seen it all before, heard it all before in the 1980's. AI hardware didn't happen then, it will not happen now. AI got a recent boost from the Deep Learning idea and collecting training data on the internet. What it can now do (which is impressive) will now plateau until the next new technique in 40 years time.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
