Microsoft's vision for the future of work is you trusting Redmond to get AI right If the future of work is a choice and "not a predetermined destiny" – as Microsoft puts it in a recent report – it would be nice to know why Redmond is so intent on shoving its version of that future down our throats. Microsoft's document is called "New Future of Work" and emerged last month with the suggestion we should not …
I hope so. Crypto, NFTs and now AI feel like they're trying to recreate the dotcom bubble of the late '90s; I'm a bit disappointed that the Register seems much less sardonic about the matter this time round and if anything seems to be just another cheerleader. A lot of the problem in all cases is the name, which is (probably deliberately) confusing, vague or misleading; I guess at least some people routinely refer to "AI" as LLM which is a lot more accurate as it isn't what most people would actually think of as AI. I've had several conversations with non-tech people who think it really is intelligent and perhaps even sentient and they seem disappointed that, while ChatGPT may make it seem that way, there's no actual thought nor understanding going on at all and it's no more AI than decades-old ELIZA, it just has larger models to work from (okay, simplistic, but more true than calling it "intelligent").
Part of my cynicism is that it's now 40 years since I first heard of AI being within reach and recalling the endless stream of disappointment that teenage me first experienced when she read that Lisp was the language for AI (really!), and since then the likes of Prolog and intelligent systems (gf did those at college a few years later and seemed to feel likewise; not quite the same disappointment, but that they weren't what they were colloquially described as being) and so on.
Not that it doesn't have its uses, though it's no surprise that the techbros seem to be most interested in finding a way around copyright law. Of course any loopholes will be very quickly closed off as soon as the plebs look like they may be able to do the same.
I guess at least some people routinely refer to "AI" as LLM which is a lot more accurate
Well, no, not really. Most people who don't outright reject the (now largely meaningless) term "AI" include diffusion image-generating models in it, and those are not at all LLMs; diffusion and transformers are very much Not the Same Thing. And we now have multimodal models, which again aren't LLMs because they're not restricted to language. On the other side, many people lump things like autonomous-driving models and other sorts of non-transformer convolutional deep ANN stacks under "AI", and those aren't at all LLMs either.
The problem is we don't have an umbrella term for all of these models except the terrible "AI" and the nearly as atrocious "deep learning", and while the latter isn't as dated as the former, it's still several years old and therefore no longer cool.
Colleagues who have worked with it express frustration that the models are very much biased towards US norms especially around regulatory and financial matters. A lot of work is required to create meta-prompts that rebaseline the models to use international or local non-US standards.
Since Microsoft doesn't deploy rebaselined models into its overseas data-centres, it means that every customer has to do the bootstrapping work for their tenant.
That's what happens when you shove the entire internet into the training set and push the go button. These programs are not looking through the data to find out which things apply to your country, they're just guessing, and if there's more about the US in their training data, it's going to show up when it randomly looks for answers unless you've crafted your prompts to keep reinforcing your country name. Even then, it's not guaranteed to get things right, just more likely to. I'm hoping that people will eventually recognize that this cannot answer specific detail questions when those questions get past simple (I.E. whenever a simple search wouldn't turn up the answer).
Siri often gets things wrong.
Several times a year, Siri will give out my phone number to random businesses. And never any business in the city where I live. It is annoying to hear "Are you Motel 6?" or, "Are you Wal-Mart?" It is never the same business and has never been a local number. I don't know who to talk to stop it. Apple support is just as useless as Microsoft support. When it first happened, someone helpfully said to me: "Siri is not always right." Ever since then, when I get these calls asking for random businesses, I make it a point to tell them I know you have an iPhone.
You can disable it entirely so it can't be activated without enabling it again. I'm not exactly sure what "nuke it completely" entails. Do you want Siri to no longer appear in the Settings so it's impossible to turn back on?
Their point appears more general than that, though, since Siri is a frontend to a set of databases that are usually available in other places. Siri's contact information for businesses, for example, is the same ones you can see in Apple Maps as well. The problem is not the voice interface, but the incorrect data it occasionally returns.
Oh, yes.
Me, driving last week in northern Oregon, about 5 miles away from my local Wilco farm store, but wanting directions because I'm coming in via a different route to normal: "Hey Siri, get me directions to the nearest Wilco farm store"
Siri: "Here are directions to the Wilco store in Sonora, California. Your estimated time of arrival is tomorrow at midnight."
Sigh.
In the immortal lyrics of Eric Idle, "Pray that there's intelligent life somewhere up in space, 'cos there's bugger all down here on Earth."
You don't have to be a miscreant to use AI to publish garbage -- it seems from numerous reports to do that spontaneously quite often, even in benevolent hands.
But the greatest hazard we face is decline of human capacities to recognise garbage as garbage as reliance on the unreliable becomes automatic. This is akin to the decline in capacities in mental arithmetic when pocket calculators became commonplace.
OK, there was no absolute merit in doing mental arithmetic where a calculator could deliver the answer more quickly and easily, but what got widely lost was the ability to validate by mental estimation whether the calculator was delivering the right answer -- that was assumed even where it was not so.
So most of us become less and less capable as we depend more and more on the machines instead of doing our own thinking, relying on a super-set of supposedly brilliant minds that create successive generations of the tools that make us dumb. That's a huge commercial win for the few, but disasterous for the many.
"But the greatest hazard we face is decline of human capacities to recognise garbage as garbage as reliance on the unreliable becomes automatic."
AFAIKS social media have already made great strides in achieving this before LLMs came into the picture.
I was vaguely impressed by ChatGPT generated code given the relatively simple scenarios I tried out, but it's all moot when my employer says it is verboten for any real world activity. Which I had basically assumed based on my understanding from coverage here and elsewhere, but I finally found the policy confirmation this week. Ain't no data leakage gonna be allowed round here.
Developers developers developers developers...
Back then, despite everything else, Microsoft did pay attention to the people who wrote the software that made their platform useful. The older documentation was - and still is - pretty good.
The newer stuff... not so much.
"The Windows giant concludes its report with the admonition that "the future of work is a choice, not a predetermined destiny" – and that we should be thinking about AI’s potential to shape that future."
Err, right. Coming from a company which does its absolute best to ensure that all its customers have to work in the way Microsoft wants (which generally involves maximum use of subscription services) this really is complete bollocks.
Not much of a pilot.
Went to a conference that showcased Copilot but as always its Microsofts marketing wank video. All looked fancy and goid. You can do this and that with it in Excel.
So today, standing with the friend at work where she had to do lots of shit to a massive excel doc I said lets try Copilot on it.
First it would only work when the file was on Onedrive. Then it said "I can't do anything until you put all that shit in a table". So she did. Then asked it the simple question she needed it to do, less difficult than the fancy bollocks on the marketing video. It thought about it and said "No, got no fucking idea how to do that"
What a loud of shit. All it was asked was look in one column. Any time a specific word appears create in a new column flagged it.
We gave up.
I still don't understand why they haven't been hit with anti-trust again with Edge as they are doing similar to what got them in hot water with IE back in the day.
Forcing it on everyone. Misleading text when you try to change default brower. Misleading when you trying to search for another browser. If you use Outlook the forcing of Edge being the browser that will open when clicking a link in an email and finally, making it a pain in the arse changing the default apps. Having to choose each one and another bullshit misleading warning when you do that.
Because, even with all that annoying crap, people aren't adopting Edge. The market share figures typically have it down near the bottom. While that doesn't technically prevent this from being ruled an anticompetitive action, it's unlikely to be brought up as one because it's clearly failing and thus it is hard to establish harm to anyone.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
