KPop Demon Hunters – I thought it was a pretty good show. The Marvelous María Beatriz, not so much. Not so much for the story – it seems melanged from lots of mythical/fantasy stories, much like the ones the MMB watched from Korea and Japan: demons, good and evil (but also some human doubt thrown in that, when resolved, lifts the human to a higher plane of being, of self-knowledge).

There is nothing musty about the storytelling, but, also, the storytelling is as rigorously curated as a social media algorithm to ensure that there are no off-ramps for viewers to sidle down to think their own thoughts and make their own judgments.

It is a visual representation of the math workings of an algorithm calibrated to entice, allure, seduce, enliven on a specific schedule tied to achieving specific effects to keep the eyeballs in place (and, if there were options during the show, to sell stuff tied to what’s happening on the screen – juices, soda pop, broken glass, demon slayer swords and so on).

This is, of course, something of inside joke in the movie, given the power that social media in the movie has over the people using it: the simple shoulder shrug of the Saja Boys suddenly has hundreds of Tik-Tok videos of people imitating it in less time than it takes sunlight to reach the earth. The Honmoon, a protective shield against the demons that the demon hunters are fated to create and renew until it becomes so protective that the demons are defeated, is an analog to the digital Honmoon binding one to the other.

The goal is a world sealed off from the darkness of the demons, that is, a world curated and bright, moved along by a musical score crafted as a playlist of perpetual ear-worms and defined by a psychology that acknowledges trauma but suggests that it can be overcome by will power, self-assertion and self-admiration.

None of that class nonsense, or systemic prejudice, or income inequality, or moral relativism, or crushing grief. No adulthood, really.

This doesn’t take away from the bounciness of it, just as watching BTS or the legions that have followed them or Psy’s Gangnam Style doesn’t mean a betrayal of the good fight for justice and equality. Fun is fun – should be taken when it can be taken.

But there shouldn’t be a mistake in seeing KPop for the product that it is. It is perfectly pitched for its audience – it’s a touch frightening how well the creators know their audience, but, then on the other hand, we shouldn’t be surprised that they do, given the data mountains that everyone has so blithely handed over to the profit-makers, available to anyone with the right bot for mining them.

KPop is not subtle or overly clever or ashamed to steal from Studio Ghibli (the blue lion and the three-eyed pigeon with the hat), but its power is not so much in its own form but the node it occupies on the tangled Honmoon that is today’s virtual world, where in large and small ways, human experience is being flattened and re-shaped for sale without pause or mercy. KPop does its work with flash and glitter, but there isn’t a real moment in it – not that that bothers anyone watching it, who want the verve in the vein and the confected triumph of killing imaginary demons to momentarily displace the ordinary ugliness around them since it doesn’t seem that much of what they do in their real lives makes a dent in the dirt.

•

A few other ideas have popped up while thinking about this, having to do with voices and algorithms. A Dec. 3 article in the New York Times by Sam Kriss, titled “Why Does A.I. Write Like…That?” explores the weird stylistics of AI writing and why it produces what it produces.

The source, of course, is in this variation of AI technology: large language models. LLMs are a probability technology that uses statistics to refine its imitation of human language. They “work” (whatever that means) by guessing a sequence of words based on an omnivorous ingestion of language materials that they compare against a prompt. LLMs don’t produce writing; they output best guesses, some of which, in the AI parlance, are hallucinations – that is, the models just make shit up in order to fulfill the prompt.

Despite claims that the LLMs have harvested all the language there is to harvest, that is not true. They have ingested what they’ve been digitally fed, but they have not touched most of what is out there to touch (not just in English but in all languages), like most of the books and manuscripts in Harvard College Library, Library of Congress, the imperial archives of the Spanish empire in Seville, and so on.

So, when an LLM outputs a response to a prompt, it is really drawing upon a very limited library, which is further limited by being made up of the most readily accessible materials, materials laced through with social bias, clichéd writing, human stereotypes, and so on.

So, if AI’s algorithm continues to spit out em dashes and triplet phrases and the formula “it’s not just X, it’s also Y” and the overuse of “delve” and “tapestry,” that’s because, based on the statistical correlations it has found in its material, these are considered what is “good” writing and “good writing” should be delivered to the person inputting the prompt. It can only generate what its templates permit it to generate.

LLMs shouldn’t be faulted too hard for what they do because, after all, they don’t think, they don’t experience the world, they don’t introspect – they just try to play what Alan Turing called “the imitation game” as best they can.

The LLMs called human beings, however, should not be shown such sympathy. Humans don’t have the massive library of, say, ChatGPT, but they do have skills in craft (and craftiness), imagination (and lying), self-awareness (and hatred of the other), so that when they create the daft mutterings of a script, they do so with intention if not always with insight to correct their foolish work.

I am prompted to say this by our recent viewing of Landman, another product of the Taylor Sheridan multiverse. (I haven’t been able to bring myself to watch Yellowstone or any of its progeny, though I did like the script for Hell and High Water and the movie Wind River.)

Set in the oil fields of Midland, Texas, the “patch,” as it’s called, is a man’s world (and packed into that word “man” is all of the testosterone algorithms in the LLM of Sheridan’s brain: rough and tumble, gritty gentleness, codes of honor).

Of course, there are women in the man’s world, and from Sheridan’s LLM, they are either shrews and empty-headed sex dolls or ball-busters hiding a broken soul. I am not sure why they are even in Landman’s story mix since the workings of the oil industry (primarily run by men) are fascinating enough without having to have obligatory female characters who are not allowed to add any weight to the story and are just objects either coveted by or smothered by (or both) by the manly men of the Taylorverse.

The point of all of these words up to this point? Several, actually. Whenever the MMB goes away for a conference or a trip home, I do not stream anything. And I find that I feel what I can only call “respite,” a welcome quietness (more like an absence, like when a weight is shucked off and there is that momentary rebound of the body as the muscles are released and the strain relieved).

I feel like I gain a bit more agency over my own brain rather than letting the entertainment enter me and work the controls. Shutting off the algorithm shuts off the passivity caused by the playing of the algorithm in the brain, and there is a relief and contentment that comes not only from the negative of being released by the algorithm but the positive of thinking once again for myself, testing my thoughts and reorganizing my scripts.

Another point: The human brain can be released from the grip of its LLM, but it takes effort and discipline to do that. It really takes poetry. Recent articles have shown how malicious information can be injected or requested in LLM chatbots by using poetry because poetry, unlike more metered prose, adheres to no special rules of composition, so the LLMs are terrible at predicting sequences because they have no statistics to go by.

The LLMs behind KPop and Landman provide comfort because they do not require any response or interaction – just take it in, and the fewer questions you have, the better it is for keeping the spell unbroken. Which also makes them culturally conservative. The women of KPop may be energized and self-confident, but it still takes a man (Jinu) to make it possible for Rumi to break the spell of Gi-Wa. In Landman, men do real work (and die) while the women either absorb themselves in looks and being liked or fuck with the old-boy networks where they aren’t wanted.

But streaming services aren’t going to change a formula that gets eyeballs on their products and subscription fees into their bank accounts and dividends to shareholders. The fight is very much within the personal brain of the individual person to resist, to cleanse, to seek out what is not considered “in” to keep the brain fresh and free.

This is the fight that Carol is making in Pluribus, where the largest LLM in the world exists: all human beings save for 12 who have, for some reason, been spared absorption into the whole. Carol’s character is rough, foul, stubborn, hard-hearted – but that is the course of action she needs to follow to save the soul. Maybe it’s the course of action we all need to follow in these days of AI slop and the unreliability of reality.