• they/them

plural system in Seattle, WA (b. 1974)
lots of fictives from lots of media, some horses, some dragons, I dunno. the Pnictogen Wing is poorly mapped.

host: Mx. Kris Dreemurr (they/them)

chief messenger and usual front: Mx. Chara or Χαρά (they/them)

other members:
Mx. Frisk, historian (they/them)
Monophylos Fortikos, unicorn (he/him)
Kel the Purple, smol derg (xe/xem)
Pim the Dragon, Kel's sister (she/her)

posts from @pnictogen-wing tagged #LLMs

also:

it looks like Sam Altman and OpenAI are desperate to inflate their public image as opposition to the LLM fraud ramps up. my new browser tabs have been full of puff pieces about Altman and ChatGPT—isn't great how merely launching a web browser gets you hit in the face with ads now? oh sorry I meant "legitimate technology journalism". 🙄

I have tried to talk to these people and their fans on Twitter and it's hopeless. no argument gets anywhere and I can only assume that any sentences of mine that cast doubt upon the world-shattering AGI supergenius which OpenAI et alii are about to give the world (any second now! for real this time!) read as "hater" talk to them and they tune me right out. my talk doesn't have ZERO effect, though, because sometimes afterwards these people start getting real defensive, apt to spout things like "all expertise is fake anyway" or whatnot.

And occasionally a weird note of doublethinkful humility creeps into the techbros' talk. It's like, "Yeah I know I'm really an idiot who's faking it but isn't everyone?" It's like in their heart of hearts they have decided that humanity really has shot its bolt and is now retrogressing, so the only possible hope to solving multiple insoluble problems is the Machine God. They at least feel like they've got an edge on the folks (like Mono and me) who are totally out of the loop and critiquing this stuff from an outsider's perspective. Mono's erudition at least seems to faze them a bit, but surely they must feel like he's a goofy old relic to still be trusting in books and higher education. Isn't all that stuff going away?

If you can't beat 'em, join 'em. What's the least expensive way to get into this LLM crap? Can one host one that's wholly self contained?

~Chara



Night in the Woods contains a fun little scene between Mae Borowski and Lori M. where they talk about putting metal figurines on the railroad tracks to be run over and squished, which feels like a fundamental unit of creativity: turn one kind of object (a tin soldier) into another kind of object (a weird squished metal thing.) "This is important art," says Lori, and if I remember correctly, Lori and Mae squish two metal figurines together. Another unit of creativity! Combine two things into a third thing. Small achievements like that mean a lot when you're trying to rescue yourself from oblivion and the Void, so I find the scene heartwarming.

Something I haven't found heartwarming is the way that high-pressure business culture, which now thoroughly afflicts the world of technology, has idolized creativity. To hear them talk, they were the sole possessors of the stuff. Nobody wants to be a plain business executive any more, or even a CTO or CEO, although these titles still have weight. The real game now is to brand yourself as a builder, a founder, a creator, even if all you do is gamble on cryptocurrency or sell accessories to TSLA fans. It might simply be noise intended to drown out the voices of those who are truly dedicated to creativity and constructive activities. Yet as with all things in the world of business and technology, one senses an undercurrent of sincerity: they're not just lying or hyping, but really do believe, sometimes anyway, that they're building great things. Elon Musk surely still believes, even after all his humiliations, that he's assembling a paradise for himself and his followers. (Followers may be optional.)

It's not as though building a TSLA requires no creativity, no innovation, even if the feat doesn't seem nearly as impressive as the fanboys say it is. I must confess that I'm not wowed by cars in general and I detest how tricked-out cars are sold with a Top-Gear-ish technological fanaticism, as if a fancy car (and not, say, universally available insulin) was the peak of human technological achievement, aside from rockets of course. All the same, Musk's workers have been compelled to push at technological limitations. The fanboys aren't completely wrong to be excited at TSLA's novelties. All the same, I feel as though the TSLA approach to "innovation" is very much a matter of diminishing returns and trumpeting incremental changes as technological revolutions. We've had many decades now of technological marketing in which every single tiny alteration to a product is sold as revolutionary. And that's bad, because even thermal noise causes changes. If one can market anything as a novelty, then hasn't innovation become irrelevant?

Nevertheless, there's a speck of creativity in what these people do, akin to the fundamental creativity of Lori M. and Mae Borowski squishing figures together on railroad tracks. The techbros are mashing up everything they know into something they consider to be brand-new—and it is! The LLM approach to artificial intelligence seems stale, mostly different by virtue of scale rather than technological novelty, but one can't deny that there is some thing new here, even if it's just a new monstrosity. And then the techbros want to mash their new thing together with old things to make them into something different. Yes, it's like trying to gin up a new market for home decorations by outfitting them with cheap clocks or bimetallic thermometers, but it is new. It's creativity. Kind of. Isn't it?

~Chara of Pnictogen



Having written that long post, I must say, I'm still privately convinced that it ought to be possible to emulate the results of OpenAI-style LLM "artificial intelligence" with far simpler methods. The techbros place far too much faith in brute force and bulk; they seem to think they've created a superbrain merely because they've heaped so much raw computing power in one place, and they don't want to think too hard about whether they're using it effectively.

I think of the famous "Burroughs method", in which William S. Burroughs cut up texts and scrambled the pieces (I don't think he invented this idea but he's popularly associated with it) and I ask myself: would Sam Altman be able to tell the difference between procedurally generated text from a ChatGPT thingummy, and text generated through the Burroughs method?

~Chara of Pnictogen