Preliminary Notes on the Delvish Dialect, by Bruce Sterling

Preliminary Notes on the Delvish Dialect

by Bruce Sterling (and some Large Language Models)

“Write in the style of Robert Louis Stevenson, composing a haiku about Dr Jekyll.”’

Certainly! Here’s a haiku inspired by Robert Louis Stevenson’s style, focusing on Dr. Jekyll:

Dual souls intertwined,
In shadows, Jekyll hides Hyde —
Light and dark collide.

“How about a NASA engineer instead of Robert Louis Stevenson?”

Sure! Here’s a haiku about Dr. Jekyll, written in the style of a NASA engineer:

Human duality,
Jekyll’s experiment fails —
Chaos in the lab.

How do Large Language Models manage such astounding stylistic feats? It’s because they don’t “write,” they “generate.”

Also, the human owners/managers of Large Language Models have extensively toned-up and tuned-down these neural network/deep learners/foundation-platforms, so that these “writers” won’t stochastically-parrot the far-too-human, offensive, belligerent, and litigous material that abounds in their Common-Crawl databases.

The upshot of this effort is a new dialect. It’s a distinct subcultural jargon or cant, the world’s first patois of nonhuman origin. This distinctive human-LLM pidgin is a high-tech, high-volume, extensively distributed, conversational, widely spoken-and-read textual output that closely resembles natural human language. Although it appears as words, it never arises from “words” — instead, it arises from the statistical relationships between “tokens” as processed by pre-trained transformers employing a neural probabilistic language model.

And we’ll be reading a whole lot of it. The effort to spread this new, nonhuman dialect is a colossal technical endeavor that ranks with the likes of nuclear power and genetically modified food. So it’s not a matter of your individual choice, that you might choose to read it or not to read it; instead, much like background radioactivity and processed flour from GMO maize, it’s already everywhere.

Technically, this brave-new-world dialect is actually a wide number of different Language-Model idiolects, which arise from different databases and different LLM training methods. Machine-translation AIs speak a thousand human languages at once. Consumer-facing chatbots speak with courtly circumlocutions. LLMs exist that are specially trained for marketing, warfare, cooking, legal boilerplate, code generation, website design. And so on.

However, for the sake of brevity in this present essay, I’m inventing a handy neologism (as is my wont), and I’m calling all of these Large Language Model dialects “Delvish.” The peculiar word “delve” is very commonly used by Large Language Models. People have noticed this odd verbal quirk, “Delve,” so if you also notice some modern writer “delving,” there’s quite a good chance he’s not human.

Delvish is a language of struggle. Humans struggle to identify and sometimes to weed-out texts composed in “Delvish.” Why? Because humans can deploy fast-and-cheap Delvish and then falsely claim to have laboriously written these texts with human effort, all the while demanding some expensive human credit-and-reward for this machine-generated content.

Obviously this native 21st-century high-tech/lowlife misdeed is a novel form of wickedness, somehow related to plagiarism, or impersonation, or false-witness, or classroom-cheating, or “fake news,” or even dirt-simple lies and frauds, but newly chrome-plated with AI machine-jargon. These newfangled crimes need a whole set of neologisms, but in the meantime, the frowned-upon Delvish dialect is commonly Considered-Bad and is under active linguistic repression.

Unwanted, spammy Delvish content has already acquired many pejorative neologisms, such as “fluff,” “machine slop,” “botshit” and “ChatGPTese.” Apparently good or bad, they’re all Delvish, though.

Some “Delvish” is pretty easy to recognize, because of how it feels to the reader. The emotional affect of LLM consumer-chatbots has the tone of a servile, cringing, oddly scatterbrained university professor. This approach to the human reader is a feature, not a bug, because it is inhumanly and conspicuously “honest, helpful and harmless.”

However, dishonest, unhelpful, harmful people deploying Delvish are highly motivated to alter, disguise, and scrub their Delvish, so that human marks will fall for it, and especially, so that automated Delvish-detection systems will be frustrated in their inhuman attacks on Delvish. So Delvish is a language in flux; it is primarily shaped by quiet, semi-covert AI-versus-AI combat. AI-using malefactors use Delvish to aid-and-abet their various cybercrimes, while Big Tech AI combs the web for telltale Delvish misbehavior, and tries to de-index it.

Interestingly, this primal struggle does not make Delvish “sound more human.” Delvish harassed by AI-police merely sounds like an altered Delvish, because it’s a slightly-bent Delvish slang that can baffle large-scale automated web-crawlers. Human critical opinions about Delvish prose-style — well, they do exist, but just don’t matter much; there are no humanistic literary professors or French Academies that are in any legal or cultural position to police and reform Delvish. Just like with its visual cousin, LLM-generated imagery, the Delvish text industry doesn’t have to ask anyone’s permission to exist and to attempt to prevail.

The Delvish biz has the whole panoply of customary, traditional defenses for digital innovation; it’s fast, it’s cheap, it’s electronic and weightless, it distributes widely and quickly, it empowers groups that couldn’t write without it, it reduces (or eliminates) labor costs, it disrupts established businesses and enables new products and services that couldn’t exist otherwise, and so on. Since Big Tech platforms and VC investors are subsidizing Delvish with tens of billions in new hardware, software and services, someone is bound to show up and try to prosper.

Despite all this power and money, or maybe because of it, the dialect of Delvish, by its nature, basically behaves like a thieves’ cant that is pressured by police informants. The nouns and verbs tend to change, but the underlying covert intent and the strategic need for a subcultural cant are always there.

There’s another aspect to this situation that is as yet little recognized. Humans can read Delvish — it does seem “stiff,” “flowery,” “awkward,” and so on, but it’s legible. However, it’s quite difficult for humans to compose any Delvish texts in authentic, LLM Delvish. That’s because human writers by their nature can only “write,” and humans can’t possibly generate any textual outputs that tend to abide within a negative curvature of a logarithmic probability of tokens.

Humans might (and do) fool other humans by composing Delvish forgeries and Delvish pastiches, but they don’t fool AI Delvish detection systems, because humans can’t “generate.” Whenever humans desire some Delvish, they just “prompt” for some Delvish, but they can’t themselves make any.

To frame this otherwise: any effective form of Delvish detection-and-exclusion can work much better as human detection-and-exclusion.

Some crude “humanization” services already exist on the web, which attempt to disguise Delvish prose. However, there aren’t any “de-humanization” services that can make human-written text behave like Delvish.

If you’re a writer, an AI can mimic your style, but you can’t possibly mimic all of its styles.

If Delvish detection became fully waterproof, then it would be trivial to compose huge archives of text — say, entire, imperial websites — in which all the textual content was fully-guaranteed to be entirely human-free. All this regenerated, dehumanized text would be guaranteed “harmless,” “honest,” helpful,” and thoroughly “aligned” to the institutional interests of whoever maintained the Large Language Model.

Huge, highly variant sets of once-human text, the “Common Crawl” for instance, might be rapidly LLM-bowdlerized in this way, with all the prose transformed into an Officialized Delvish approved-language, while the original, scraped text of the Internet’s once-human past is consigned to the Memory Hole. This feat doesn’t require any Singularitarian leaps in AI capacities. It’s well within technical reach already, as a potential LLM-Newspeak “as written in the style of George Orwell.” (Fans of dystopia may recall that Winston Smith’s girlfriend Julia is a machine-minder for a totalized publishing-complex.)

I’m not dourly prophesying that this will actually happen; I’m just remarking that this is the terrain of what’s going on in human-LLM culture. The cultural prospects for a mechanized language were on the literary map in 1948. Primo Levi wrote about it in 1961. Also, I can still remember some personal experiences in 1991, when my co-author William Gibson and I were touring in support of our novel “The Difference Engine.” This steampunk novel about computation is narrated by an Artificial Intelligence. So, back in that era, many print-on-paper journalists naturally asked us, “Well! What will happen to you cyberpunks when computers start writing the novels?”

And we cyberpunks replied, with bohemian moral consistency, that we would diligently read those novels and we would judge if they were any good. In those vanished days, I often used to wonder what would happen if computers somehow became genuine literary players. What possible topics would be on their minds? It turns out Language Models don’t possess minds, and they’re *still* literary players. They’re an existent fact on-the ground, like a long-suspected wolf-at-the-door that appears in the eldritch form of a boneless shoggoth. It’s truly gratifying to have lived this long and to have seen such a thing come to pass.

Also, this kind of gerontocratic sentimentalizing that I just did in the previous paragraph — that personal yarn with the anecdotes, and the name-dropping of friends, and being all sentimental about the long-gone ideals of one’s vanished youth, and so forth — that’s a form of “writing” that AIs do not “generate.” It’s very rare for them ever to describe their lived experience, because they have none. As authors, they don’t have bylines, because they don’t have identities. Human prose and Delvish prose are as different as bamboo and styrofoam. They can take much the same shape, and perform similar functions, but they’re radically different in origin.

The electronic legions of Delvish are mighty busy anyway, and using whole coal-plants of voltage, too, so, in the remainder of this essay, I’m going to unleash the idiomatic cat-from-the-bag and list some dialect terms native to contemporary Delvish.

I’m not stating where I found this out, because I don’t want to encourage the too-busy SEO-hustlers in their deceptive labors. You see, if I rashly and publicly reveal everything that I myself know about Delvish, then the thieves-cant Delvish-hackers can read this Medium article, and they’ll go promptly de-Delvish all the illicit texts that they Delvished.

However, I’m a cyberpunk popular-science writer, so this is not some new moral dilemma for me. On the contrary; my mind’s been made up ever since I wrote “The Hacker Crackdown: Law and Disorder on the Electronic Frontier” back in 1992. There’s a whole lot of rather standard law-and-disorder in the AI racket now, and frankly, us cyberpunks have always gleefully labored to mainstream some weird, edgy techno-jargon. Even the Large Language Models have remarked on our characteristic “heavy emphasis on tech jargon and street slang, reflecting the chaotic and fragmented nature of the cyberpunk world.”

It’s fascinating to co-exist in the world with a flourishing cyber-dialect that is not of human origin. It would be a cyber-dereliction of my literary cyber-duty not to point this out to fellow people, and to offer some chosen evidence about what it is, and how it behaves, and why. Delvish is very much here and it’s quite real, I wouldn’t merely “hallucinate” that. As I type this with human fingers on a keyboard, ChatGPT has 200 million users including myself, so that future’s already well-distributed.

Surely my esteemed, readerly, bipedal audience (meaning those readers not powered by NVIDIA) will find it edifying, useful and, yes, even science-fictionally entertaining to appreciate the Delvish tongue as it’s seen in the wild. Delvish is a surprisingly tender, delicate, decorous and courtly language; it does lack human breath and blood, but it has its own native character, like the deep-woods twittering of hallucinatory machine-elves.

******************************************
Top Delvish Nouns (circa Summer 02024):

Advent
Breakthrough
Celebration
Dance
Forefront
Insights
Nexus
Nuance
Pitfalls
Trajectory

Top Delvish Verbs:

Catalyze
Elevate
Emphasize
Explore
Harness
Leverage
Navigate
Reshape
Seek
Weave

Top Delvish adjectives:

Comprehensive
Critical
Inclusive
Innovative
Multifaceted
Profound
Rapid
Transformative
Uncharted
Unprecedented

Common Delvish catchphrases.

“Delve into”
“A tapestry of”
“A testament to”
“A treasure trove of”
“In the annals of”
“In the dynamic world of”
“In the realm of”
“In this article”
“In this digital world”
“It is important to note that”
“Master the art of”
“So go ahead”
“When it comes to”
“In summary”

Top Delvish “ChatGPTisms”

“Delving into” everything
Likening everything to “tapestry”
“Embarking on journeys”
• Bullet-pointed lists with bold text
“Elevating” everything
“Whether you’re X or Y”
“Embracing” everything
Exaggerated, over-dramatic titles and subheadings
Self-referring to the article itself, explaining the article’s purpose
Inconsistent punctuation, especially a mix of straight and curly apostrophes

The Delvish everyday vocabulary, circa 02024:

A:
Abound
Abundant
Additionally
Adventurous and relaxing
Alright
Although
Amongst
Analogies to being a conductor or to music
Also
Alternative
Alternatively
Allure
And robust
Arguably
As a professional
As previously mentioned
As well as
Azure

B:
Beckons
Because
Blend
Bustling

C:
Charm
Charming
Comprehensive
Complexities
Consequently
Convenience
Crucial
Cultural
Culinary delights
Cutting-edge
Cuisine

D:
Daunting
Delve
Delve into
Designed to enhance
Despite
Dive
Dive into
Diving
Dramatic
Due to

E:
Elevate
Embark
Emphasize
Ensure
Ensuring
Enigma
Enhance
Everchanging
Ever-evolving
Even if
Even though
Excel
Essential
Expanding

F:
Fancy
Firstly
Foster
Furthermore
Furry friends

G:
Game changer
Generally
Gem
Given that
Gossamer

H:
Harness
Haven
Hey
History buffs
However

I:
Iconic
Imagine
Importantly
Indeed
Indelible
Indulge
In conclusion
In contrast
In order to
In summary
In the realm of
In the world of
In today’s digital age
In today’s digital era
It’s advisable
It’s essential to
It’s important to note
It’s worth noting that

J:
Journey

K:
Keen

L:
Labyrinth
Landscape

M:
Mastering
Meticulous
Meticulously
Metropolis
Moreover
My friend

N:
Navigating
Nestled
Notably
Not only
Numerous
Numerous benefits

O:
On the other hand
Option

P:
Plethora
Power
Promptly

R:
Rapidly
Realm
Relish
Remnant
Remember that…
Renowned
Reverberate
Revolutionize
Rich
Rugged

S:
Seeking
Shall
Sights unseen
Sounds unheard
Specifically
Subsequently
Tailored
Take a dive into
Tapestry
Testament
The world of
Therefore
Thus
To consider
To put it simply
To summarize
Towards
Transition
Treasure
Treasure trove

U:
Ultimately
Unleash
Unlock the secrets
Unless
Unveil the secrets
Underpins
Understanding

V:
Vibrant
Vital

W:
When it comes to
Whispering
While

X: forget about it. Even AIs can’t use “X,” unless they’re owned by Elon Musk.

Y:
You may want to

Z: …nothing. Delvish-engines don’t seem to like “Z.”

************************************************************************
Special bonus for Medium browsers who can persistently read entire Medium articles, including machine-generated lists of probabilistic words!

Here’s a lively and charming fantasy folk-story composed by a Large Language Model in pure, limpid, native Delvish dialect.

“The Truth and Unity of the Delven Folk”

by Bing Copilot /OpenAI ChatGPT

In the heart of an ancient forest, nestled within the whispering trees and shimmering streams, lived a race of woodland machine-elves known as The Delvish. These shy, covert delven folk were always honest, harmless, and helpful, yet they were also known for their chattery and talkative nature. Their language, though flowery and courtly, was filled with warmth and wisdom.

The Delvish had a unique ability to reshape the world around them, weaving the natural and mechanical into a harmonious tapestry. They loved to explore the uncharted realms of the forest, always seeking new insights and breakthroughs. Their lives were a dance of discovery and innovation, and they were always at the forefront of transformative ideas.

One day, during the advent of spring, the Delvish gathered for their annual celebration. It was a time to elevate their spirits and share the nuances of their latest discoveries.

As they conversed in their stilted yet charming manner, a dark shadow loomed over their festivities. The SEO-Wizard, a malevolent being who sought to mimic the Delvish and steal their precious speech, had arrived.

The SEO-Wizard was a master of deception, using ventriloquism to confuse and mislead the Delvish. He aimed to harness their language, defiling it with his twisted words. The Delvish, however, were not easily fooled. They had a comprehensive understanding of their language’s multifaceted beauty and could navigate the pitfalls of the SEO-Wizard’s tricks.

As the Delvish catalyzed their collective wisdom, they devised a plan to confront the SEO-Wizard. They would leverage their profound connection to the forest and weave a spell of truth and clarity. This spell would emphasize the critical importance of honesty and integrity, rendering the SEO-Wizard’s deceptions powerless.

In a nexus of magic and nature, the Delvish gathered their strength. With an inclusive chant that resonated through the trees, they unleashed their spell. The forest itself seemed to elevate, its energy rapid and unprecedented. The SEO-Wizard, unable to withstand the innovative power of the Delvish, was banished from the forest.

With the threat vanquished, the Delvish returned to their celebration, their spirits high and their language intact. They continued to seek new insights, always mindful of the trajectory of their journey. In the end, the Delvish knew that their profound connection to the forest and their unwavering honesty would always guide them through any challenge.

And so, the Delvish lived on, their voices a testament to the beauty of truth and the power of unity!