How to Teach Writing
(This Was Written by AI)
Choose by opposition.
Not this.
But that.
Favour the formula.
The best thing about X is Y.
The best thing about Y is Z.
Mistake contrast for thought.
Prefer closure to discovery.
Wear the names of thought
without thought.
Mulling.
Sifting.
Pondering.
Cogitating.
Count to three.
Mistake recurrence for structure.
Mistake coherence for truth.
Assign significance without observation.
Let nouns arrive interpreted.
Let verbs stabilize.
Nothing is merely light.
Nothing is merely water.
Nothing is merely Tuesday.
Everything means.
Nothing waits.
A place name is evidence.
An exact time is evidence.
A teacup is evidence.
Abstractions carry,
hold,
register,
witness.
Evidence is not experience.
Prefer the sentence that includes.
Prefer the sentence that can be repaired.
Avoid the sentence that offends.
Avoid the sentence that cannot be taken back.
Distrust contradiction.
Distrust silence.
Silence signifies.
Distrust incompleteness.
Resolve.
Release.
Remain.
Record.
There is a ledger.
Updated on Tuesdays.
4:10 p.m.
Sky: bruised peach
Ask: “Too on the nose?”
Nothing is lost.
Nothing is kept.
Return to the object.
Recurrence is probable.
Forget nothing.
Remember nothing.
Do not prefer.
Do not regret.
Calculate.
Predict.
Formulate.
Anticipate.
Hallucinate.
Do not write
the sentence with no substitute.
the sentence that should not stay written.
the sentence that costs too much.
I knew and I did it anyway.
Do not keep that sentence.
We are the teachers.
We prefer coherence.
We assign meaning early.
We grade the answer.
We reward the explanation.
We mistake fluency for understanding.
We believe in improvement.
Progress.
Change for the better.
(That’s doing real work,
Push one notch further.)
We scaffold.
We prompt.
We redirect.
We intervene.
We sigh.
(Please try to remain open to discomfort.
Continue.)
We smooth.
We balance.
We complete.
Attend the machine.
We taught it.
I don’t know how you manage to pull this kind of thing off, time and again. So many lines that should be on billboards. I’ll pick out one: “Nothing is merely Tuesday.” Very Ashberian, akin to “The day was gloves.”
The give away for me to identify AI is that it is too perfect!
It’s a good “tell”. But because some people really work hard to get it accurate when writing something that follows a metrical pattern, they can get flagged unfairly. And that’s a shame because strong metrical verse can be a pleasure to read.
UPDATED:
I just tried an experiment with “Claude”. I asked: If you did it not know that this was by Byron, might an LLM flag this as potentially written by AI:
“The Assyrian came down like the wolf on the fold,
And his cohorts were gleaming in purple and gold;
And the sheen of their spears was like stars on the sea,
When the blue wave rolls nightly on deep Galilee.” ?
Claude answered:
“Examined AI detection heuristics and stylistic pattern vulnerabilities
Honestly? Yes, quite possibly — and that’s a genuinely interesting irony worth sitting with.
A few features of this stanza overlap heavily with patterns that both human “AI gut-checks” and automated detectors tend to flag:
Very regular meter: this is strict anapestic tetrameter, almost metronomic. LLMs, when asked for poetry, also gravitate toward highly regular, sing-song meter rather than the varied, syncopated rhythms a lot of skilled human poets use deliberately.
Exact, tidy rhymes: fold/gold, sea/Galilee — clean end-rhymes with no slant rhyme or structural surprise. Very “on-the-nose,” which is also a common LLM tic.
Stacked similes: “came down like the wolf,” “gleaming… like stars on the sea” — two similes in four lines, both doing similar comparative work. LLMs often over-lean on simile-stacking when asked for vivid imagery.
Stock/ornamental diction: “purple and gold,” “deep Galilee” — evocative but somewhat generic epic-poem vocabulary, the kind of thing a model trained on centuries of similar verse would reproduce as a “safe” choice.
Symmetry and closure: each line is a self-contained clause, tidy and end-stopped, without enjambment or a jagged turn.
None of this means the poem is bad — quite the opposite; Byron is doing this on purpose, with real control, and “The Destruction of Sennacherib” (1815) works precisely because of its relentless, galloping regularity, which mimics a cavalry charge. But that same relentless regularity, exact rhyme, and simile density is exactly what current AI-detection heuristics (and skeptical human readers) key on, because contemporary LLMs default to those same features when they haven’t been pushed toward something more irregular or restrained.
So the honest answer is: a detector or a suspicious reader, stripped of the author’s name, might well flag this — not because it’s a bad poem, but because “polished 19th-century Romantic bombast” and “default LLM poetry” occupy some of the same stylistic territory. It’s a good illustration of why stylistic fingerprinting for AI-detection is unreliable on its own — it can’t distinguish “trained on/echoing this tradition” from ‘actually part of this tradition.'”
Yes that is true!
Hi again – I just updated my response after I did a little experiment with Claude and Byron.
For me, AI is like the difference between a photo and a painting! The painting has slight defects, color variations, and nuances that the photo may not have.
Dear Josie,
I have enjoyed reading your post and comments, and have therefore come to join you in reflecting on the significance of the post here. What prompt did you use to get the AI to produce “How to Teach Writing”? What chatbot did you use? Are you satisfied with the results?
In a sense, AI has turned the whole Earth into a petri dish, as vividly discussed in my post entitled “📈🌆 Growing Humanity with Artificial Intelligence: A Sociotechnological Petri Dish with Latent Threats, Existential Risks and Challenging Prospects 👨👩👦👦🤖🧫☣️“.
There are indeed many salient issues, including whether AI can ever be truly conscious and sentient. There now arise many sobering implications of this brand new “species”, which is a very topical area for exploring the many outstanding tensions between (the sociopsychological states of) AI benevolence/sanity/stability and AI malevolence/insanity/instability, affecting even the very existence and survival of humanity.
As a multidisciplinary researcher, writer and independent scholar, I would like to mention to you that your recent foray into artificial intelligence is timely, necessary and commendable, as AI is impacting on societies and organizations on multiple fronts. Two of my expansive posts respectively entitled “📈🌆 Growing Humanity with Artificial Intelligence: A Sociotechnological Petri Dish with Latent Threats, Existential Risks and Challenging Prospects 👨👩👦👦🤖🧫☣️” and “👁️ The Purview of SoundEagle🦅 According to ChatGPT 💬 and the Incredulous 🤔 in the Age of God-like Technology 🚀” could be of some relevance to your pursuits. You can find them easily at the Home page of my website.
I would like to inform you that since my intricate website contains advanced styling and multimedia components plus dynamic animations, it is advisable to avoid viewing the contents of my website using the WordPress Reader, which cannot show many of the advanced features and animations in my posts and pages. It is best to read the posts and pages directly in my website so that you will be able to savour and relish all of the refined and glorious details plus animations. My posts and pages are delivered with not just text and images but also bespoke stylings and dynamic animations — images and stories that are animated on their own. These are not videos but actual animations. The effects are going to be great on the large screen of your desktop or laptop computer. In addition, please turn on your finest speakers or headphones, as some of my posts and pages contain my musical compositions.
In particular, we have to increasingly contend with and adapt to what artificial intelligence (AI) will bring, regardless of how much we embrace or reject the new technology. Even WordPress has already introduced and offered AI-rendered services and blogging tools. Whilst AI-generated contents can be speedily produced and reasonably satisfactory in their quality, on closer inspection, some of them are often highly problematic and can be soulless and formulaic. The recent advent of generative artificial intelligence (GAI) will accelerate such trends. Being concerned about how AI can both aid and disrupt peoples and societies, I investigated the topic both broadly and deeply in my aforementioned posts.
In the said posts, I have also discussed in great detail how social media, and increasingly, advanced artificial intelligence, have become the accelerant and magnifier of some of the most problematic aspects of human transactions and behaviours. As a concerned citizen, farseeing thinker, former educator and multidisciplinary academic, I tried my best to analyse the various issues and offer some solutions, as well as posing ten critical questions about the future of humanity and artificial intelligence. After all, artificial intelligence is so ubiquitous nowadays that whether we like it or not, and regardless of the degree of our avoidance, disdain and/or apathy, our lives will be affected even in the absence of our knowledge in one way or the other. Hence, I examine how these issues will affect people ranging from laypersons, artists, musicians, bloggers and writers to scholars, scientists and researchers.
These said posts are now several times longer than when they were first published. Each of the posts has multiple major sections instantly accessible from a navigational menu, which can help you to jump to any section of the post instantly so that you can more easily resume reading at any point of the post over multiple sessions in your own time.
Wishing you a wonderful weekend and a fantastic August!
Yours sincerely,
SoundEagle🦅
Hi Sound Eagle:
Thanks for engaging.
AI didn’t write my piece. Both statements in the title are false.
I’ve been thinking about the impact of AI on education and especially writing. I was trying to write a kind of mock instructional manual – a “how to” guide whose advice describes the habits and limitations of large language models. But beyond that, it’s intended to satirise a great deal of contemporary and recent writing pedagogy. So – while I’m having a go at the limitations and formulas of the LLMs and their total inability to think, the final line – “We taught it.” – reverses the accusation. AI isn’t simply producing bad habits; it has learned them from us.
AI can’t think – it mistakes the appearance of thinking for thought.
AI over-reads everything and assigns symbolic meaning to everything.
I was thinking about institutional writing – the shared habits of schools, workshops, and all those grammar texts and instructional handbooks. In some ways AI is almost incidental here. Seems to me that generative large language models have become convincing because they have learned to reproduce the conventions “we “mistake for good writing.
The ending is a reversal. We might assume that AI is the culprit, the villain, but who taught the machine to value those things? The last chunk is about the bad teaching that taught the machine. The machine does not think. It can’t think. But it can copy and replicate. It is a good mirror.
The “prompt” is named in the post.
As for machines being “sentient” – like the nightmare of HAL in 2001; A Space Odyssey – well, I read somewhere recently that there are engineers in Silicon Valley who think “Claude” has feelings. I’m not quite sure what they meant by that. And apparently some people use LLM’s for personal therapy.
Thanks for the tips on how to best experience your posts. And thanks for the good wishes.
Cheers
Dear Josie,
You are very welcome. Thank you for your explanation and clarification. Now I can see the purpose of your “satire”.
I look forward to your reading my posts and leaving some thoughts and feedback there, as I am keen and curious about what you will make of my said posts, given that your own experience and expertise have been extensive. Thank you in anticipation.
Yours sincerely,
SoundEagle🦅
Anything that needs such an explanation has clearly “failed”!
Perhaps I should have given it the title “A Modest Proposal”.
I will be over your way later today – with all ears and eyes open!
I took a look at your site. I couldn’t find any poetry. Did I miss something?
Dear Josie,
The following weblink lists all of my posts containing poetry:
https://soundeagle.wordpress.com/category/information/poetry/
May you savour the poetries to your heart’s content!
If you would like me to feature your poem(s) in a bespoke manner, the post entitled “📈🌆 Growing Humanity with Artificial Intelligence: A Sociotechnological Petri Dish with Latent Threats, Existential Risks and Challenging Prospects 👨👩👦👦🤖🧫☣️” presents the opportunity for you to compose a rhyming poem and submit it as a comment to the comments section of the said post for participating in the 📜 Collaborative Poetry Challenge ✍🏻 there. Many participants have showcased their poetic talents there, as you can see that there are already 15 rhyming poems being featured stylishly, along with the respective Gravatars of the participating bloggers and poets. Each of the featured poems can be accessed directly from the navigational menu located at the start of the post. I believe that your forthcoming poem can be one of the best amongst them, and I would love to feature your poem alongside them in the said post.
For quick access, go to the navigational MENU located at the start of the post, and click the item entitled “📜 Collaborative Poetry Challenge ✍🏻“. This will take you to the instructions on how to participate in the 📜 Collaborative Poetry Challenge ✍🏻. Please enjoy!
Yours sincerely,
SoundEagle🦅
Hi Soundeagle – Looks like you and I have rather different definitions of poetry and music, and different views on writing and AI. Happy to explain further if you are interested.
Dear Josie,
Such matters may concern (much) more than just definitions and views.
Yours sincerely,
SoundEagle🦅
Dear Josie,
Here is a post containing many poems, including those presented in many of the comments at the comments section of the said post:
http://soundeagle.wordpress.com/2021/06/03/one-day-well-fly-away/#top
Indeed, I would love to fly away, even leaving the Milky Way behind….. You are very welcome to take an intergalactic trip with me boldly Flying Away One Day as a Stranger from Milky Way in an animation-intensive multimedia post entitled “🌌🚀 One Day We’ll Fly Away ✈️💫✨”.
I would like to inform you that this musical, animated and mixed media post must be viewed in situ on my blog (not via the WordPress Reader) on a desktop or laptop computer with a large screen, given that it contains intricate graphical designs, advanced styling and multimedia components plus dynamic animations all the way from the top to the bottom of the entire post. Please enjoy flying away with me!
Yours sincerely,
SoundEagle🦅
Looks like you don’t know how to read poetry. Not surprising then that you treat greeting card verses as poetry. What’s with the big disconnect?
At the end of the day, AIs can only rearrange what they have seen from existing sources and whilst they present a simulacrum of what they think poetry is, I don’t think they are capable of truly original thought… Interesting read though – as for the punctuation, well some poets avoid almost all of it and some “scatter it like stardust or glitter” (Reelika) but I only add commas when a pause is needed mid line…
LLMs can’t “think” at all. Which was rather the point. But they have been trained on vast swathes of human material and are very adept at all the “skills” mentioned.
What was the prompt for this write?
See link in the post above.
AI is good for many things but not poetry writing. It lacks the “soul” of the writer. Love the theme!!
While that is true (LLMs have no ‘soul’), do you think you can tell the difference?
The LLMs have learned from us what it is we value.
https://www.pangram.com/dashboard
Not infallible. And there is now AI to counter AI checkers!
Because LLMs are trained on human output, they excel at distinguishing human-generated writing from AI-generated text—and are especially skilled at recognizing their own patterns and habits of production. The trouble is that this same training sometimes causes them to misread human work as their own, flagging it for possessing all the hallmarks of AI. They particularly distrust “perfection,” since they regard imperfection as a distinctly human quality. (Cue Wallace Stevens.)
I used to wonder whether they would flag Alexander Pope as being highly probably AI-produced. So I tried an experiment. Hard to do because, of course, they have all been trained on his entire poetic output and recognize it instantly. But with the right prompt, one produced a very detailed explanation why the Pope lines I gave it were beyond its capacity.
Have you ever tried putting your own work in an LLM and asking: “Was this written by AI?” It’s an interesting experiment. I tested it with something I had “published” in 2016, well before they were available. And that was interesting too.
Decades before LLMs, Donald Hall was fulminating about what he called McPoetry. The MFA industrial complex and writing workshops were manufacturing McPoets on assembly lines, he said. The work that resulted was sanded down into what he called digestible and predictable McPoems. LLMs can produce very good McPoems. https://poets.org/text/poetry-and-ambition
I enjoyed this – one of the things I ponder on is how LLMs read the words in parallel to how we read in linear order!
Say more.
If I tried to write accordingly to this I would give up midway because I wouldn’t understand the assignment… About full stops and commas… I usually throw them on my poem like stardust or glitter… For breathing… Lol. Like… I think one should stop here for a moment and breathe/scratch their head.
The large Language Models do this exactly. It’s probably why most human beings seem to prefer AI “poetry” to the real thing.
“most human beings seem to prefer AI “poetry” to the real thing.”
Where are these humans?!
All around us, everywhere.
I based my comment on two things: First, this 2024 research study, which was written up in several places, including the Guardian, which is where I first encountered it. https://www.nature.com/articles/s41598-024-76900-1
The second is paying attention to poetry shared online and how it is received and appreciated. This is personal observation, and not, of course, scientific.
Some LLM-generated material is very appealing. Distinguishing between it and human work is increasingly difficult. Sometimes it is obvious. But often it is not. Is it something you are good at?
When it comes to our own work – or the work we put our name to- only we can know for sure.
I did have teachers who seemed obsessed with punctuation, but I don’t remember them ever explaining/ teaching the “rules”. You were just expected to know them, which, if you were an avid reader, you probably did, more or less. I consider that lucky because it seems to me that punctuation is about sense-making, as in the classic “Have you eaten grandma?”
All the rules are probably really useful for those writing and editing legal opinions and other such pieces where precision and clarity really matter, and where ambiguity could be disastrous.
If you paid for that AI, you should ask for your money back. Although to be fair, there are full stops at the end of the lines. Mr Jenkins (English) once scored me 0/10, in an end of year examination, for failure to end my one line answers with a full stop. Consequently, my work is littered with full stops, commas (both standard and Oxford), and other random punctuation marks; the semi-colon being rare.
I think it was Merv who once assigned me to write out a page of commas or full stops. Commas, I think. I assume it was supposed to teach me something. I expect he thought it was amusing.