I have a hypothesis, a possible reason why so many people in #tech are irrationally impressed by the #LLM and why they keep seeming to think they're seeing massive improvement in these things even though the LLM trickery is really a technological dead-end: it's solely because #technology people tend to have an extremely narrow range of human knowledge.

This is especially likely to be true as one scales the ladder of social privilege and corporate hierarchy. The more power a person has, the more likely it is they're using that power to make sure they're surrounded only by people who talk the same way about the same limited range of things. You see an extreme example of this from Elon Musk, easily among the stupidest "geniuses" who has ever blighted the Earth, but he's been able to seal himself away from acknowledging his own stupidity by hiding amid a bubble of courtiers and yes-men, people easily impressed by his technobabble.

And so, #LLMs are capable of surprising these dullards—not only that, LLMs are, by their very nature as unintelligent devices, are given a social latitude to be surprising that is not accorded to ordinary human beings.

I suggest in fact that the sloppy enthusiasm one's been seeing from far too many heedless computer-touchers (qq.v. @kumarvibe or @enantiomer, oh boy gotta love the appropriated chemistry term there!) about the brilliance of #ClaudeCode (or whatever other #LLM stochastic trash-generator they've happened to fall in love with) is indirect proof that these tools LACK intelligence.

Why do I say that? It's because a genuinely intelligent being, with genuine intellectual dependence and force of personality, is a challenge. A relationship between true intellectual peers is not a smooth one, nor should it be! A legit thinker wants their thinking challenged. A fallacious thinker—and that, I suspect, is where the usual high #tech booster belongs—crumples up when their words or ideas meet with serious challenge.

And that's where "generative #AI" steps in, supplying people with a stupid assistant, one that doesn't actually think and is therefore never an intellectual challenge, but which does sometimes offer up surprises.

Computer geeks like @enantiomer talk about "discussing" things with #Claude or some other emitter of confabulated #LLM gibberish in a way they'd never talk about with a real-life intelligent person with a real-life intellect, because such a person might make the mistake of telling the computer geek something they don't like and can't easily answer, whereupon the "discussion" would simply turn into a slapfight. But #LLMs do not think; they apply no reasoning to the output they confabulate, which is simply a kind of statistical average drawn from its training materials that happens superficially to resemble intelligent language.

The users of this fraudulently marketed intelligence-free #AI are, I suggest, actually exploiting the fact that these devices are stupid and stochastic in nature. 'Enantiomer' et alii can simply run them over and over again, knowing that they can get answers that satisfy them through a mindless process of repetition. Again...not a behavioral pattern characteristic of genuine intellectual discussion between peers.

So where does the surprise aspect come in? It's simply a matter of what's been stuffed into the #LLM or other "Generative #AI" thingummy. Sam Altman and Elon Musk and all the rest of the #technology barons who are hoping to exploit the LLM craze are relying upon a brute-force method: appropriating or stealing as much human writing, from all conceivable sources, as they can grab and stuff into a "data center" somewhere.

The LLM isn't truly reading this material, merely chopping it up and collecting statistics on word and phrase usage, but simply because there's so much of it—and coming mostly from sources outside the extremely narrow range of human communication with which a tech geek is likely to be familiar—the LLMs are getting trained to emit vocabulary and rhetoric, picked up from this vast mass of material, which the tech geeks may truly never have EVER seen before.

The geeklords are used to rigorous policing of their human boundaries. Tell @pluralistic that he's not nearly so clever as his Internet sycophants say he is, and he'll simply cut you out of his life and stick to socializing with his carefully selective and manicured in-crowd. That's simply behavior as usual for anyone of importance in #technology (or in #business generally). Use unfamiliar words or challenging turns of phrase with such people, as a real person, and you'll probably just get a faceful of insults and slurs in response—or, because it's the Internet, you'll be accused yourself of being a mere mindless LLM bot.

It's such a common accusation, and a telling one, because these same people would be absurdly happy to have a "discussion" with a mindless LLM bot, one they know to be mindless, and therefore no actual challenge, but one that was capable of surprising them because occasionally it'd emit some unfamiliar words or phrases. The LLM offers the #tech geeks a fascinating compromise between rote predictability of response and the unexpected.

Challenge them hard enough on this topic and the computer geeks might pounce on the idea that "intelligent conversation" isn't even a meaningful idea in the first place. Sam Altman infamously tweeted "i am a stochastic parrot, and so r u," attempting to beat back the (true) accusation that #ChatGPT and other #LLM devices are merely assembling their outputs stochastically (i.e. through a randomizing algorithm). Far more recently, Marc Andreessen blurted out in public that he has no long-term memory and regards self-awareness as basically fake and some sort of modern-era fallacy of thought, though of course Andreessen said that EVERYONE is like this, not just himself.

It's sort of like how, under other circumstances, #tech geeks who get themselves into trouble during an argument will start hinting that maybe reality isn't real and how there's no distinction between truth and lies and other fragments of ideology that they hope will serve as mic-drop excuses for talking nonsense. What if EVERYTHING is nonsense?? ever consider that??? Check and mate, you woke mind virus zombie you.

I rather wish there were more attention drawn to this specific point: the techlords who have set themselves up as THE experts on #intelligence and thinking, who are so frequently boasting about their own superior intellects while lamenting (in racist and eugenicist terms) about how so much of the human population are basically like mindless zombies and an impediment to civilization, are also ready to pounce on the idea that "intelligence" is merely an illusion, a sort of handwaving by which we pretend that our own words—or the words emitted by machines—are more than just strings of stochastically generated noise which happen to elicit desired responses from others.

Ultimately, Musk and Altman and Andreessen et alii do not want the concepts of intelligence and consciousness to be systematically examined and studied. It suits their purposes for "intelligence" to remain a deliberately vague, doublethinkful buzzword. The techlords themselves don't want to think too hard—and they don't want anyone round them to be brighter than themselves, certainly not their "artificial intelligence".

@mxchara the most arrogant people on Earth think intelligence (and probably every other fake or real trait that they respect) is just arrogance? Sounds about right! /gen
@mxchara But if intelligence isn't just cut up text strings combined with a random walk, then maybe I haven't understood intelligence...
That means that maybe I am not as smart as I thought...
Nah, I think life ist just a simulation so people are basically just markov models on legs.
@mxchara
Congrats on the new PFP again! (I like it) 
@mxchara *mashing follow button
@mxchara you're right on the money. The industry lionizes dropping out of university before you have to bother with you know, ethics courses, poetry classes, exposure to Black literature etc. And it sure shows.