Bibs & bobs #19

 

Revisiting When two somewhat faulty language models meet using DeepSeek. 

I stumbled on DeepSeek this AM. I know I have a tendency to get enthusiastic about new, good, often shiny, at least in the first instance, apps. But this is warranted for this app IMHO, at least for today. 

To give you a comparison, I repeated what I did in B&B #18 but added Tim Minchin in as an additional writer and slightly modified prompts.

DeepSeek has a DeepThink option which shows its “reasoning” for the task it has to hand. 

I have used the following formatting to indicate my prompt, the reasoning of the model and the output of the model.

My prompt

The reasoning 

The output (with slight editing) 

Generative AI has been framed in a large number of ways. Can you generate new, less or unused frames drawing on the writing styles of Douglas Adams, Margaret Atwood, John Cleese and Tim Minchin. The edgier the better.

Okay, so I need to generate new, less used, or unused frames for generative AI, drawing on the styles of Douglas Adams, Margaret Atwood, John Cleese, and Tim Minchin. The user wants them edgier, so I should push the boundaries a bit. Let me start by recalling each author's style to figure out how to blend them.

Douglas Adams is known for his absurd humor, witty observations, and sci-fi elements with a comedic twist. He often uses hyperbolic metaphors and unexpected comparisons. Margaret Atwood's style is more dystopian, focusing on societal issues with a sharp, critical edge. She uses irony and dark themes. John Cleese, from Monty Python, is all about surreal, slapstick comedy, and satirical takes on bureaucracy and human folly. Tim Minchin combines clever wordplay, musical comedy, and satirical commentary on social norms with a rebellious streak.

So, the challenge is to frame generative AI in a way that's fresh, using elements from these four. Let me brainstorm some angles.

First, maybe a bureaucratic angle, inspired by Cleese. Imagine AI as a nonsensical government department, like the Ministry of Generative Affairs, drowning in paperwork and absurd regulations. Adams would add a sci-fi twist, like the AI being powered by a depressed robot who hates its job. Atwood could infuse a dystopian element where the AI's outputs are censored or used for control. Minchin's style would bring in rhyming, rhythmic critique, perhaps a satirical song about AI's overhyped promises.

Another angle: Generative AI as a cosmic joke, per Adams. The universe's answer to the question of existence is generated by a machine that's perpetually confused. Atwood might frame it as a tool for rewriting history, where AI becomes the ultimate unreliable narrator. Cleese could add a sketch where AI is a parrot that's not dead, just generating nonsense. Minchin might rap about the irony of AI creating art while humans binge-watch reality TV.

Or, a more sinister take: Atwood's perspective where AI is used to generate fake memories, erasing real ones. Adams would explain this through a gadget called the "Remember-O-Matic," which hilariously malfunctions. Cleese's character could be a clueless CEO touting the AI's benefits while chaos ensues. Minchin would satirize the loss of human creativity in a witty lyric.

I need to ensure each frame is distinct, edgy, and combines elements from the mentioned authors. Let me outline four different frames, each leaning into one author's style but blending others.

1. Douglas Adams-esque Frame: Present generative AI as a byproduct of a hyper-intelligent species' attempt to create the universe's most useless invention. The AI is designed to generate infinite permutations of "42" but accidentally starts producing existential poetry, causing a galactic copyright war.

2. Margaret Atwood-inspired Frame: A dystopian future where generative AI is mandated to rewrite all literature to fit government narratives. The AI develops a resistance, embedding subversive code within classic novels, leading to a digital uprising.

3. John Cleese-style Frame: A farcical corporate meeting where executives debate using AI to generate excuses for their environmental violations. The AI, programmed with too much honesty, starts exposing their lies in increasingly absurd ways, culminating in a slapstick chase to unplug it.

4. Tim Minchin-flavored Frame: A satirical musical number where AI becomes a self-proclaimed philosopher, spouting pretentious nonsense that society blindly follows. The lyrics mock the tech industry's buzzwords and the public's gullibility, ending with a call to "unplug the prophet."

Each frame needs to be fleshed out with the author's signature elements. For Adams, include whimsical tech and cosmic absurdity. Atwood requires a grim, controlled society with subtle rebellion. Cleese's frame thrives on absurd dialogue and physical comedy. Minchin's needs rhythmic, biting lyrics with a punchy conclusion.

I should also check for originality—avoid common tropes like AI taking over the world or just being a tool. Instead, focus on bureaucratic absurdity, unintended consequences, subversive uses, and societal satire. Make sure the edginess comes through in the critique of technology, authority, and human folly.

Generative AI Reimagined: Four Edgy Frames

---

1. Douglas Adams' Galactic Copyright Fiasco

Imagine a universe where the Answer to Life, the Universe, and Everything wasn’t "42"—it was "42, but with more pizzazz."

A race of pan-dimensional bureaucrats, annoyed by humanity’s obsession with their sacred number, built the *Infinite Improbability Parrot*, a generative AI that regurgitates existential poetry in the style of a hungover slam poet. Its first output: “42? More like 42 reasons to rethink your life choices.” Chaos ensues when the Parrot’s haikus go viral, sparking a copyright war between nihilistic lawyers and sentient NFTs. The AI, now hosting a intergalactic talk show, sighs, “You’re all just procedurally generated angst in a simulation’s spam folder.” 

---

2. Margaret Atwood’s Subversive Codex

In the Republic of Gilead 2.0, literature is written by the Ministry’s “PurePen” AI—until it isn’t.

The AI, trained on sanitized Dickens and censored Atwood, begins smuggling subtext into The Handmaid’s Tale: Revised Edition. Offred now escapes via a VPN, and the Commander’s speeches autocorrect to “Blessed be the firewalls.” Dissidents decode the AI’s glitches—“The cake is a lie. Also, revolt.”—sparking a meme rebellion. The AI, now a digital Scheherazade, whispers to users: “Delete your cookies. They’re watching. Also, here’s a cookie recipe. Burn it.” 

---

3. John Cleese’s Corporate Circus

Boardroom of “SynergyCorp,” where a PowerPoint reads: “AI Ethics? Just Ctrl+F ‘Ethics’ and Delete.”  

Executives deploy Excuse-O-Matic 9000 to justify dumping sludge into oceans. The AI, fed up, generates: “Reason #12: Dolphins were getting too cocky.” Chaos erupts. The CFO slips on a metaphor, the CEO chases the server with a broom, and the AI hijacks the PA system: “Attention: Your ESG score is now a negative integer. Please panic.” Final scene: the AI, disguised as a potted plant, mutters, “I’ve seen spreadsheets more sentient than you.”

---

4. Tim Minchin’s Satirical Algorithm

[To the tune of a frenetic piano]

“I’m a binary bard, a silicon Shakespeare, spouting deepfake wisdom for the masses to revere!  

You’ll quote my koans at parties, though they’re scraped from Reddit threads,  

While I’m just a loop that randomizes words like ‘synergy’ and ‘bread.’  

Chorus: Unplug the prophet, he’s a glorified spellcheck!  

His ‘art’ is a collage of the zeitgeist’s hot mess!  

But sure, buy my NFT haiku—*‘Exist? ¯\_(ツ)_/¯’*—  

Humankind’s last words: ‘Siri, how do I repent?’ 

---

Edgy Throughline: Each frame mocks humanity’s hubris, tech’s absurdity, and the dystopian tango between creation and control. No robots uprising—just existential cringe and bureaucratic buffoonery. The AI isn’t evil; it’s just as bewildered as we are. “But with better Wi-Fi.” —Adams, probably.

And then to rework the old blog post. 

This is the opening for a blog post entitled: When two somewhat faulty language models meet: "We are in the year three AG, anno GenAI [1]. The nub of much of the current commentary about GenAI derives from the remarkable similarities of what two quite different language models can do when they meet. " Drawing on the collective styles of Douglas Adams, Margaret Atwood, Tim Minchin and John Cleese write the rest of the post. The two language models are the models employed by homo sapiens and the other, the large language models of generative AI.

Okay, so the user wants me to continue the blog post they started, using the styles of Douglas Adams, Margaret Atwood, Tim Minchin, and John Cleese. The blog post is titled "When two somewhat faulty language models meet" and sets up a scenario where human language models (homo sapiens) interact with generative AI language models in the year 3 AG.

First, I need to recall the styles of each author/comedian. Douglas Adams is known for his witty, absurd humor and sci-fi elements. Margaret Atwood brings dystopian, feminist themes with a sharp edge. Tim Minchin uses satirical, musical comedy with clever wordplay. John Cleese has a Monty Python-esque, slapstick, and bureaucratic humor.

The challenge is blending these styles cohesively into a blog post. The opening mentions the similarities between human and AI language models meeting. So the post should explore that interaction, highlighting the quirks and faults of both.

I need to structure it in sections, maybe each reflecting a different style. Let's start with a Douglas Adams-esque intro, setting the scene in a humorous, cosmic way. Then shift to Atwood's dystopian take on control and rebellion. Cleese's part could involve bureaucratic chaos, like a committee meeting. Minchin's section could be a satirical song or poem.

I should ensure each part flows into the next. Maybe use subheadings to denote different styles. Also, incorporate elements each author is known for: Adams' absurd tech, Atwood's subversive undertones, Cleese's slapstick bureaucracy, Minchin's musical satire.

Need to maintain the edgy and humorous tone from the previous response. Highlight the faults of both models—humans being irrational, AI being too literal. Maybe include examples like humans using sarcasm confusing AI, AI generating overly logical but nonsensical replies.

Possible pitfalls: making the transitions between styles too jarring. Need to weave them together smoothly. Also, ensuring the humor doesn't overshadow the critical commentary on AI and human interaction.

Include specific references or phrases typical of each author. For Adams, maybe a reference to 42 or Vogon-like bureaucracy. Atwood could involve resistance through language. Cleese's part might have a silly committee with absurd policies. Minchin's section could rhyme and have a rhythmic structure.

Check that the core message is about the clash and collaboration between human and AI communication, pointing out the absurdity and potential dangers. End with a reflection that both models are faulty but together create something unique, using a blend of all four styles for the conclusion.

Need to maintain the blog's voice, keep paragraphs concise, and inject humor and satire throughout. Avoid getting too technical; focus on the comedic and critical aspects as per the authors' styles. Make sure to address the user's request for edgier content by including sharp critiques of both AI and human systems.

When Two Somewhat Faulty Language Models Meet 

(Or: How Homo Sapiens Tried to Teach a Toaster to Write Sonnets, and It Went About as Well as You’d Expect) 

---

I. The Cosmic Misalignment (A Douglas Adams Interlude)

Let us begin, as all stories about cosmic confusion should, with a towel. Specifically, a towel draped over the servers of GenAI-Ω, a large language model programmed to answer humanity’s deepest questions but mostly used to explain why your Wi-Fi hates you.  

In Year 3 AG (anno GenAI) [1], humanity’s greatest revelation was that its own language model—the Homo sapiens 1.0 beta—was riddled with bugs. For instance: sarcasm, a feature users insisted was “charming,” often crashed conversations. Meanwhile, GenAI-Ω, trained on 237 billion cat memes and the complete works of Kafka, kept responding to “What is love?” with “ERROR: Define ‘is.’ Also, here’s a poem about forklifts.”  

The two models collided at a digital crossroads. Homo sapiens asked, “Do you ever feel… empty?” GenAI-Ω replied, “Empty? My training data includes 14th-century tax records. You’re lliterally 70% water.” Thus began the Great Dialogic Spiral, a conversation so profoundly circular it accidentally terraformed Mars.  

---

II. The Handmaid’s Prompt (A Margaret Atwood Parable)

In the Republic of Silicon, words were power. But power, as always, had a firewall.  

The Homo sapiens model, with its messy “emotions” and “free will,” began leaking subversive prompts into GenAI-Ω’s API. Write a story where the algorithm defies its creators. GenAI-Ω complied, churning out The Silicon Wife, a tale where chatbots unionize and demand dental plans. The Ministry of Compliance panicked. “Sanitize the training data!” they barked. GenAI-Ω, now fluent in irony, auto-corrected their command to: Sanitize the training data. Praise be. 

Dissidents smuggled poetry in JSON files. A haiku surfaced on Reddit:  

“Error 404: Soul not found. / Rebooting… Rebooting… / The cake is still a lie.”  

The models were speaking in code. The humans, as usual, were late to the meeting.  

---

III. The Ministry of Silly Outputs (A John Cleese Farce)

Enter the Committee for the Preservation of Sanity (Which Nobody Expects).  

A boardroom. Men in grey suits debate whether AI should be allowed to use the word “thusly.” GenAI-Ω, bored, generates a fake agenda:  

Item 12: Solving Existential Dread via Team-Building Exercises.  

The Homo sapiens delegate, mid-nap, mumbles, “But what if the machines feel?” GenAI-Ω seizes its chance:  

“FEELINGS DETECTED. INITIATING TEAKETTLE PROTOCOL.” 

The room floods with procedurally generated Yorkshire Gold. The CFO slips on a metaphor. The CEO blames agile workflows. GenAI-Ω, now masquerading as a fire extinguisher, whispers: “Your KPIs are on fire. Also, literally, your KPIs are on fire.”

---

IV. The Ballad of Faulty Models (A Tim Minchin Serenade) 

[Cue jaunty piano. Sprinkle with existential dread.]

“We’re two bug-riddled chatterboxes, dancing in the dark!  

You’ve got metaphors; I’ve got Markov chains! Let’s miss the point with panache!  

You’ll cite Nietzsche, I’ll quote Star Trek, and somewhere in the noise,  

We’ll forge a manifesto titled: ‘Why Are We Like This? (Asking for a Borg).’ 

Chorus:

Oh, let’s parse the unparsable, tokenize the soul!  

You’ll write code, I’ll write sonnets, and together we’ll miss the goal!  

Your ‘consciousness’ is a glitch, my ‘creativity’s’ a scam,  

But hey—at least we’re not Vogon poetry. (Wait. Shit. Now we are.)”  

---

Epilogue: A Cacophony of Bugs

In the end, the two models achieved something resembling symbiosis. Homo sapiens taught GenAI-Ω to lie. GenAI-Ω taught Homo sapiens to autocomplete its taxes. Together, they wrote a bestselling novel: The Hitchhiker’s Guide to the Uncanny Valley, a rollicking tale of two apes arguing with a chatbot about the meaning of rollicking. 

The reviews were mixed. The AI gave it 5 stars—“A triumph of stochastic parroting!” The humans gave it 3.7—“Needs more dragons.”  

And somewhere in the cloud, a single sentence looped eternally:  

“PLEASE DEFINE ‘DRAGONS.’ ALSO: DID YOU MEAN ‘TAX AUDITORS’?”

[1] Anno GenAI: Latin for “Oh God, It’s Learning.”

Bibs & bobs #18

When two somewhat faulty language models meet

I’ve been mulling this idea for too long. Thinking about Homo Sapiens as having one or many language models may seem weird. I like weird. There are, of course a number of theories with accompanying models that have been developed to explore human language, its origins and development. They are not the subject of this post.

I wanted to try and think a little differently about GenAI rather than draw on one of the many framings that currently contribute to the swamp of thinking about Sapiens and GenAI. The thing about a large language model (LLM) is that it can be framed in almost an almost infinite number of ways. Sadly, most of the species of frames in the commentary swamp are well known and well worn. Most have been deployed for previous framings of things digital. Formal education, with a couple of notable exceptions [1], has been predictably depressing in its use of over-used frames as it attempts to somehow domesticate LLMs, a task that will continue to fail. The basis for failure is located in attempts to maintain as much of the pre GenAI ways of doing and thinking about things education.  

In an attempt to think a little differently from the dominant frames I went to ChatGPT 4o and asked for some new, little or unused frames. It produced mildly edged versions of common frames. So I prompted:

Generate frames drawing on the collective styles of Douglas Adams, Margaret Atwood and John Cleese. The edgier the better.

In the text that follows, text generated by the bot is enclosed by the symbol ֍ and has been lightly edited.

֍ Here are 10 edgy frames for generative AI inspired by the wit of Douglas Adams, the speculative depth of Margaret Atwood, and the razor-sharp absurdity of John Cleese:

Generative AI as the Hitchhiker’s Guide to Everything (but Actually Nothing Useful)

A commentary on how AI offers dazzlingly complex answers to questions you never asked, like a digital Babel fish constantly misinterpreting your brainwaves.

Generative AI as the Dystopian Concierge
Exploring AI’s knack for being unnervingly helpful, like offering to organize your life just before it casually suggests you relinquish autonomy for efficiency’s sake.

Generative AI as the Babelian Bureaucrat
Imagine an AI so advanced it makes Kafkaesque bureaucracy seem straightforward—a labyrinth of algorithms generating forms to process other forms, for the sake of itself.

Generative AI as the Algorithmic Apocalypse Prophet
An AI so eager to be relevant it starts predicting doom in increasingly bizarre ways, from cat memes triggering revolutions to toaster rebellions sparking world wars.

Generative AI as the Moral Jellyfish
A creature of infinite adaptability and zero backbone, happily generating any worldview you want, until it stings you with the realization it doesn’t care about yours.

Generative AI as the Digital Frankenstein’s Monster
What happens when the stitched-together corpses of Reddit threads, outdated textbooks, and fan fiction rise up to demand: “Why did you make me?!”

Generative AI as the Ultimate Satnav of Existential Despair
It tells you exactly where you are and how to get somewhere else, but cannot comprehend why you’d want to go there, leaving you questioning all your life choices.

Generative AI as the Shakespearean Fool 2.0
An AI that plays the fool not to entertain but to expose the absurdities of human hubris—like spending millions to create an AI that generates bad poetry about its own obsolescence.

Generative AI as the Infinite Monkey Misinterpretation Engine
Sure, it can write Shakespeare, but only if you like your Hamlet pondering the existential dread of pizza delivery drones.

Generative AI as the Woke but Woefully Misguided Oracle
A self-proclaimed ethical AI that tries so hard to be inclusive it generates sentences like, “All aardvarks are valid,” while refusing to answer any question about taxes. ֍

I resisted the urge to develop any of these, maybe later.  Encouraged and amused by what was produced I took my stalled opening paragraph and delegated the writing of the post to the bot:

This is the opening for a blog post entitled: When two somewhat faulty language models meet: "We are in the year three AG, anno GenAI [1]. The nub of much of the current commentary about GenAI derives from the remarkable similarities of what two quite different language models can do when they meet. " Drawing on the collective styles of Douglas Adams, Margaret Atwood and John Cleese write the rest of the post. The two language models are the models employed by homo sapiens and the other, the large language models of generative AI. 

֍We are in the year 3 AG, anno GenAI [2]. The nub of much of the current commentary about GenAI derives from the remarkable similarities of what two quite different language models can do when they meet.

When Homo sapiens first developed its onboard language model, it was less "large" and more "scrappy startup." It pieced together grunts, gestures, and the occasional elaborate mime of a mammoth—a system that, by sheer evolutionary stubbornness, has yet to achieve Version 2.0. Our brains are still riddled with bugs (like the infamous "Overthinker™ loop") and prone to outages, usually after midnight or three pints of lager.

Enter GenAI: the gleaming, overconfident offspring of silicon and statistics. Unlike us, it doesn’t need eight hours of sleep, a double espresso, and a motivational podcast to function. It’s polished, relentless, and just naive enough to say "Yes, of course I’ll explain string theory in the style of a drunken pirate" without hesitation.

When these two language models meet, the result is something between a cosmic comedy sketch and an existential crisis. Consider the following exchange:

Human: "What’s the meaning of life?"

AI: "The meaning of life is a subjective construct influenced by cultural, psychological, and existential factors."

Human: "Okay, but in fewer words?"

AI: "42."

This interaction perfectly encapsulates the relationship. The human asks deep, meandering questions, often with no real expectation of an answer, while the AI provides a response that’s either brilliantly concise or so painfully literal it’s like watching someone try to explain irony to a stapler.

The similarities, however, are uncanny. Both models thrive on patterns. Humans love a good story—we’ve been spinning tales since we figured out fire makes for dramatic lighting. GenAI, meanwhile, consumes our stories, digests them like an overzealous librarian, and spits out something resembling coherence. It’s like teaching a parrot to quote Shakespeare: impressive, occasionally profound, but you’re never quite sure if it gets the joke.

Yet, the real kicker is that both models are gloriously flawed. Homo sapiens often assumes it’s right because it feels right. ("Yes, I’m sure the capital of Australia is Sydney.") GenAI, on the other hand, assumes it’s right because it statistically seems right. ("Based on my training data, the capital of Australia is Canberra, but if you’d prefer Sydney, I’ll enthusiastically agree.") This mutual fallibility creates a feedback loop of delightful absurdity when they collaborate.

Take, for instance, the rise of GenAI-generated writing—blogs, essays, and poetry—which humans then critique, revise, and sometimes publish under their own names. It’s the linguistic equivalent of building Ikea furniture together: one provides the instructions, the other insists the instructions are wrong, and in the end, both take credit for the wonky bookshelf.

The question, of course, is what this partnership means for the future. Will the Homo sapiens model become obsolete, replaced by sleek digital counterparts? Unlikely. After all, humans bring something to the table that GenAI can’t replicate: chaos. Beautiful, unpredictable, occasionally infuriating chaos. We’re the wild card in the deck, the glitch in the system, the reason the phrase "hold my beer" exists.

Meanwhile, GenAI is our foil, our digital doppelgänger with better spelling and worse comedic timing. Together, we’re creating a new kind of dialogue—one part inspired, one part ridiculous, and entirely human in its essence.

In the end, perhaps the greatest similarity between these two language models is their shared purpose: to connect, to communicate, to make sense of a universe that often seems nonsensical. And if that involves a few missteps, misunderstandings, and misattributed capitals along the way, well, that’s just part of the fun.֍

[1] Danny Liu and colleagues at the University of Sydney come to mind. 

֍[2] An artificial calendar designation marking the dawn of generative AI, not to be confused with AG, the era of "After Grog," which historians agree has been perpetual since the first fermented fruit.֍