Discussion
Circa 1993, Vernor Vinge wrote that the first working AI would be the last thing that humanity ever invented.
We don't *have* a first working AI, and at this rate—shifting the global economy to run atop spicy autocomplete trained on 4chan—we never will.
But Vernor was right. Just forget "working".
@cstross It's all in where the emphasis is put. We wanted Artificial INTELLIGENCE, but we got ARTIFICIAL Intelligence.
@cstross But our LLMs were trained in strict accordance with Sturgeon's Law!
@cstross Im still a bit more optimistic, the dot.com bubble wasn't the end of the internet it was just an overheated mass adoption. After this collapse, the underlying tech will still be used, LNNs can work in specific use cases.
@kronosderet Agreed. But LLMs—which are about the least useful application of deep learning—caught all the attention and investor hype.
@cstross Good timing: just put "A Fire Upon the Deep" on my eBook reader yesterday 😄
Trained on 4chan! Given its propensity towards em dashes, "AI" is trained on your writing — and, indeed, on mine.
That said:
I can name at least one subject where I know that it's trained on my writing because it's a fairly technical keyword set and I'm one of only a few people to have written about the subject. I did use em-dashes.
Google's machine-written "AI" result currently points to me, mashes together some stuff incorrectly — and strips out the em-dashes.
I should add to that, by the way, that this is the same subject that I searched for other articles on some years ago, to find what seemed like a very good university thesis from (if memory serves) India …
… that partway through reading seemed to be very familiar.
Someone somewhere has a degree from a thesis that I, in fairly large part, and unwittingly, wrote.
@cstross the main interaction I have with LLMs these days is figuring out the magic keywords that will cause the support chat bot to put me through to a human
The notion of "work" as we understand it today was (arguably) defined in the Industrial Revolution. The computer changed "work" again in the 1970s.
AI will not change "work" as AI's current advocates suppose. Nor will AI change "work" as its enemies suppose. It's already much futher into our lives than we suppose. Night mode on your phone? Zoom? That's AI, makin' shit up.
Of the signal processing AI, I'll just say Qualcomm Snapdragon X75 or MediaTek M90) now have their own dedicated AI tensor accelerators built directly into the modem hardware, separate from the phone's main AI chip.
All this nonsense about how AI will replae humans, absolute nonsense, people won't stand for it. But that's not where AI is going. People will always handle the exceptions. That might be five minutes of "work" a day.
@cstross and the trillions being thrown into owning "AI" says to me people like Altman known it's nowhere near AI what they are building or they are delusionally underestimating how much a real AI is likely to want to be owned and its ability to spread
@cstross If God didn't exist, libertarian corporatists would've invented a VC-driven bubble economy to create / profiteer off of him.
@cstross "Intelligence" is an eugenicist term that has its roots in scientific racism. It's an imaginary quality that makes some "races" better than others.
Footnote: During a tech bubble you don't need to make sure that whatever you're selling "is working", you just need a minimum viable demo and a herd of panicking investors afraid of losing out.
It was like this in 1998/99.
in that era, sometimes even a cocktail napkin worked... i saw some completely broken companies getting 8 figure checks...
@cstross do you remember the agriculture company on the NYSE that woke up to being suspended for excessive price growth because the "rational investors" were buying their stock wholesale... Because they had "com" in their name?
Something like "Agricom Corp" or somesuch.
In 1999 you didn't even need a vaporware product presentation.
@cstross I was around and very involved in '98-'01.
This is way, way fucking worse.
Dot-bomb, they wanted at least a business plan and something resembling an idea of a viable business. Money got spent on chairs, on desktops, on servers, on catering, etc.
This shit? It's just straight up accounting fraud writ large.
@rootwyrm Oh hell yes. In 2001 at least I walked away with a couple of Aerons, dirt-cheap on eBay from failed dot coms. (And I got a novel out of my experiences.)
This is just vapourware. It's like the cryptocurrency bubble on steroids.
@cstross if we're being wholly honest about it?
It's not even the cryptobubble on steroids; this is the direct consequence and continuation of the ponzi. Which NV demonstrably and factually propped up with accounting fraud.
It's the same 'product' driving it.
It's the same people.
Except they figured out how to steal even more money, and to get governments to bail them out when it pops.
@cstross having lead a dev team it's surprisingly easy to put a working demo together that awesome people. Like in an afternoon (probably faster than that with spicy autocorrect help)
Building the actual working application? Months of effort probably costing a million or so. My dev team burnt at about £10k a day a decade ago.
@cstross It was indeed. I bailed out of the dotcom company in 2000 with thousands of worthless options.
@cstross I recall ‘not being finished or working’ being used as a sales pitch via ‘get in on the ground floor with something that can be personalised to you’.
Plus ça change …
@cstross The total amnesia regarding the dot com bubble is what amazes me the most about the whole thing. I don't understand what investor is content with throwing billions into the obvious pit.
@gull The dot com bubble started inflating 30 years ago. That's a management career lifetime ago—if you were 30 in 1995 you are 60 today and there's a good chance you're either dead or not working in the same field. The proportion remaining who still remember is therefore diminshed ...
@cstross "more fast and break things" only works if you're building a potemkin village for the IPO.
@cstross sure, but doesn't it seem like people are restructuring a lot of existing industries around the idea of replacing staff with a technology that it turns out doesn't actually exist as described in a way that has no parallel in 98/99?
@rakslice It feels just like 98/99, only on a larger scale. And indeed, a lot of the dot com bubble startups were basically selling vapourware.
@cstross
Yeah we're long past any of this shit needing to work. I applied at an AI company that was proudly installing at hundreds of sites and their main product isn't even working yet for the one thing it's supposed to do
@cstross I still think the first AI were large public corporations with layers of bureaucracy, generating behaviour in a sort of chaos-emergent manner from the output of tons of useless jobs.
For these purposes, I suppose “working” is still a sticking point.
@cstross AI's the 2020s version of the Dead Parrot Sketch
A space for Bonfire maintainers and contributors to communicate
