Bibs & bobs #41

 A better read of my pet hates or a semi-interesting, time wasting serendipity machine

I’m interested in uses of LLMs that allow things that were not possible or too time consuming to be carried out pre-LLMs to now be possible with a LLM. 

I was reading one of Jeremy Caplan’s Wonder Tools posts and one of the suggestions in an intriguing list of alt think apps and ideas seemed to fit the bill. The list comes from a conversation between Caplan and A.J. Jacobs. 

The suggestion:

• Seek out surprises A.J. asked Claude for 10 articles he would probably find boring, annoying, or offensive. It surfaced one gem, about ancient Egyptians buried with golden tongues so they could speak in the afterlife.

 I used ChatGPT 5.6 and lazily copied the text:

Find 10 articles I would probably find boring, annoying, or offensive

This is what the machine produced:

The anti-CJ reading list

These are not necessarily bad articles. Some are sensible. That may make them worse: intellectual irritation works best when wrapped in reasonable prose.

1. “AI Is Making University Students Dumber, Melbourne Professor Warns” — The Australian
   Likely offence: 9/10. AI causes “brain-rot”; handwritten examinations arrive as intellectual holy water. You will probably object to treating writing technologies as contaminating substances rather than participants in changing practices. Useful, though, as an unusually pure specimen of restorationist thinking. (The Australian)
2. “How to Stop Students From Cheating With AI” — The Wall Street Journal
   Likely offence: 9/10. It begins with cheating and therefore ensures that assessment design, disciplinary knowledge and institutional hypocrisy enter through the servants’ entrance, if at all. The headline alone could power three of your blog posts. (The Wall Street Journal)
3. “In the AI Era, How Do We Battle Cognitive Laziness in Students?” — Times Higher Education
   Likely annoyance: 8.5/10. “Cognitive laziness” turns a relation among student, task, assessment, institution and machine into a character defect located conveniently inside the student. Bloom’s taxonomy is also summoned, presumably because no educational emergency is complete without it. The article itself is more careful than its title. (Times Higher Education (THE))
4. “AI Won’t Replace Teachers—But Teachers Who Use AI Will Change Teaching” — Education Week
   Likely boredom: 8/10. The immortal “AI won’t replace X, but…” construction returns, having survived several thousand LinkedIn posts. It adds connection, care, wisdom, discernment and the calculator analogy: the full ceremonial fruit basket of moderate AI commentary. (Education Week)
5. “Will AI Replace Teachers? It Depends on How They Teach” — Forbes
   Likely offence: 8/10. The title quietly relocates responsibility for technological restructuring onto individual teachers: adapt correctly and perhaps your occupation will be spared. A fine example of structural change dressed as professional self-improvement. (Forbes)
6. “Surfing the Future: Why Education Needs to Embrace AI, Soft Skills and Self-Awareness” — World Economic Forum
   Likely annoyance: 9/10. It contains outdated systems, agile learners, future jobs, soft skills, an “AI tsunami” and instructions to surf it. It also invokes learning styles and frames teacher development as strategic investment. Almost unfairly well calibrated to annoy you. (World Economic Forum)
7. “Why AI Literacy Is Now a Core Competency in Education” — World Economic Forum
   Likely boredom: 8/10. Education becomes a delivery mechanism for skills that employers are projected to require, including prompt engineering and algorithmic thinking. “Core competency” is the phrase organisations use when they want an idea to become compulsory before anyone has worked out what it means. (World Economic Forum)
8. “AI Will Transform the Workplace. Will Education Keep Up?” — Forbes
   Likely annoyance: 8/10. The question assumes that education’s natural function is to chase the workplace while carrying its coat. Economic mobility supplies the moral justification; “dynamic educational systems” supplies the management-consulting atmosphere. (Forbes)
9. “AI-Driven Personalised Learning Will Become Mainstream in 2025” — Global Services in Education
   Likely offence: 9.5/10. This offers “true differentiation,” automated planning, instant feedback, improved outcomes and competitive advantage for international schools. The author’s organisation also develops and manages schools, making the article a useful example of a market describing the inevitability of its own products. (Global Services In Education)
10. “Making AI Work for Schools” — Brookings Institution
    Likely boredom: 7/10. AI integration becomes a playbook of transparency, accountability, human-in-the-loop design, equity and monitoring, supported by experimental learning-gain results. All defensible; all tidy. You may object that the school has been rendered legible enough to manage but not strange enough to understand. (Brookings)

My predicted winner: No. 6. It combines workforce instrumentalism, soft-skills soup, learning-style residue and a surfing metaphor. It is less an article than a bingo card that has achieved consciousness.

Not bad. Could be useful with better crafted prompts and as always, generates work for the LLM user.

Bibs & bobs #40

 Where to Look, and What Just Disappeared

Rory Sutherland made an interesting observation in a recent YouTube video. It was a point about leadership. He argued that he is less interested in telling people what to do than in telling people where to look. It is both annoyingly good and inviting given the common fare around leadership. It has the quality of something said casually over coffee while quietly reversing a forklift through three management theories and a strategic plan.

Most organisations are addicted to telling people what to do. Policies tell people what to do. Strategic plans tell people what to do, although usually in language that appears to have been assembled from wet cardboard, airport signage and the minutes of a committee that died in 2018 but has not yet been told.

Universities are especially gifted at this. They can produce a 42-page document whose main achievement is to make the reader feel both governed and abandoned. Sutherland’s line then becomes attractive.

Leadership is not just command. It is attention. The leader says: look here. This matters. That does not. This is evidence. That is noise. This is urgent. That can be safely ignored until it becomes a crisis with muffins. There is something right about this and also something slippery. Telling people where to look is not innocent. It can reveal what has been ignored. It can also redirect everyone away from the large smoking hole in the floor.

“Look here,” says the leader, pointing at innovation. Meanwhile workload is in the corner eating the furniture. “Look here,” says the consultant, pointing at efficiency. Meanwhile care has been converted into a spreadsheet and is now being asked to justify its colour scheme. “Look here,” says the AI strategy. Meanwhile judgment has slipped out the back door wearing a fake moustache and carrying a small suitcase labelled “professional discretion.” This is why the where to look line matters for GenAI.

A lot of the AI conversation still asks the dull question: What can the machine do? Can it write, can it summarise, can it draft a lesson plan, can it produce a grant outline, can it write a diplomatic email to someone whose main contribution to the project has been to repeatedly reopen settled matters like a cat proudly delivering dead mice to the ethics committee? The answer, of course is yes, it can do all those things, more or less. The better question is not what the machine can do. The better question is: where does it make us look?

GenAI pulls attention toward fluency, speed, coverage, neatness and plausible structure. These are not minor gifts. A blank page can behave like a domestic tyrant. A first draft can feel like a ladder with no final rung. A summary can stop a document setting like wet cement. A list of possibilities can loosen a thought that has jammed itself in the doorway. Used wisely, GenAI can help. But fluency is also its trick.

The sentence arrives before judgment has found its shoes. The answer appears before hesitation has had a chance to clear its throat. The category settles in before the case has even entered the room. Then comes the confident paragraph, wearing a tiny academic hat, nodding gravely, and hoping nobody notices it is only pretending to be knowledge. This is where academic work gets interesting.

A large part of academic craft is not content, it is attention. Knowing where to look. Knowing what to ignore. Knowing when a sentence is bluffing. Knowing when a concept has been dragged from another field and is now stands awkwardly beside the cheese table. Knowing which footnote is quietly holding up the argument. Knowing when “further research is needed” means “we have reached the end of our courage and would now like a coffee break.”

This is part of what might be called secret academic business [1]. We teach the visible rituals: literature reviews, referencing, methods, theoretical framing, argument structure. Fine. Necessary. OK. But the hidden craft is harder. It is knowing what smells wrong. It is knowing when a field has quietly agreed not to ask the awkward question. It is knowing when the student’s problem is not a student problem at all, but a task design problem wearing a lanyard. It is knowing when a beautiful paragraph has performed the intellectual equivalent of rearranging cushions while the house burns.

LLMs can mimic some of this. They can perform critique. They can say “problematise” with a straight face. They can generate a paragraph that looks as if it has read widely and slept badly. Mimicry does not support apprenticeship. The machine can point. It does not know why pointing there matters [2]. Or, more accurately, it does not care. And not caring is part of its charm. Also part of its menace. Like a vending machine that dispenses plausible interpretations and occasionally drops a can on your foot.

This is basically why middle managers are in such a strange position with GenAI. From above comes the great hymn of efficiency: streamline, automate, optimise, transform, scale. These words now roam freely through universities, grazing on sense and leaving small piles of key performance indicators behind them.

From below comes the less glamorous question: who does the work after the miracle? Who checks the output, fixes the errors, spots the subtle damage, and explains to the student, dean, committee, partner organisation or future ombudsman that the system was used “appropriately” in a setting where nobody had time to ask what appropriate meant? This is the part that often disappears. GenAI does not simply save labour. It redistributes labour.

The time saved by one person does not always disappear. It often reappears elsewhere as checking, cleaning, correcting, documenting, soothing, translating and apologising. Usually lower down. Usually invisibly. Usually on the desk of the person already holding three collapsing systems together with Outlook calendar invites, institutional memory and a small, renewable supply of spite.

The useful middle manager may not be the one with the grand AI strategy. It may be the one who keeps asking irritatingly practical questions. Not, “How do we use AI to go faster?” but, “What are we no longer noticing because we are going faster?” Not, “Was this written by AI?” but, “Can the person defend the choices?” Not, “How much time will this save?” but, “Whose time, and where does the unsaved time go to die?” That last question is unlikely to appear on a slide with a blue gradient, which is usually how you know it may be useful.

So, at this point, “where to look” is starting to admire itself in the reflective surface of its own cleverness. The thing is that sometimes people already know where to look. They can see the problem perfectly well. They know the workload is impossible. They know the policy is nonsense. They know the assessment invites AI mimicry. They know the committee has become a retirement village for unresolved decisions. They know the bold new initiative is last year’s bold new initiative wearing a different scarf.

The problem is not attention. The problem is permission, time, protection and authority. Telling people where to look is not the same as giving them the capacity to act on what they see. This is where leadership slogans begin to wobble. “Where to look” is useful, but it can become another elegant idea floating above the swamp in clean shoes. It makes leadership sound like a matter of insight, when often it is a matter of courage, resources and not punishing the person who says the boat is both sinking and described in the annual report as “aquatically agile.”

Maybe Sutherland’s line needs an extra clause. Leadership is not only telling people where to look. It is being responsible for what your pointing hides.

This applies to GenAI too. Reward speed and the work will get faster. Reward polish and it will become smoother. Reward compliance and the room will get quieter. Reward scale and people will start to shrink inside the machinery. Reward judgment and we might get something resembling education. No guarantees, obviously. This is still planet Earth. But judgment seems the better target.

That means asking students and staff to show the choices, checks, hesitations, refusals, revisions and reasons behind their work. Not merely “did AI write this?” but “what did you do with what it gave you?” Not “is this authentic?” but “can you account for it?” Not “where is the human text?” but “where is the human judgment?”

The bot can produce text. It cannot take responsibility for what the text does. That remains our problem. Lucky us.

So yes, Sutherland is right. Leadership is partly about telling people where to look. The more troublesome version is better:

Good leadership makes better things noticeable, gives people some capacity to act on what they notice, and remains accountable for what has been pushed out of view.

This is less elegant. It has too many clauses and will not fit neatly on a coffee mug, which is a serious disadvantage in a civilisation increasingly governed by mugs, slides and laminated wisdom. But it has the virtue of being less likely to become nonsense by lunchtime. It also gives us a useful rule for GenAI: do not ask only what the machine can do. Ask where it is making you look. Then ask what has disappeared. Then ask who has been left to clean it up.

                                                                                                    

Notes

[1] This blog post is a longer account of secret academic business and GenAI. 

[2] It is often easier to imagine a mind than explain the maths.

Bibs & bobs #39

 LLMs and secret academic business

There is an old problem in universities that can be given a useful name: secret academic business. The notion has been gestured to by a number of scholars. 

Barbara Kamler and Pat Thomson’s work [1] on abstract writing is useful here because it treats apparently small academic genres as learned, political and pedagogical practices rather than neutral technical form. They describe the strange business of academic abstract writing, one of those scholarly practices that everyone is somehow meant to know how to do, despite very little direct instruction. Their point was not simply that abstracts are hard. It was that academic work is full of small genres, tacit moves, institutional tricks and rhetorical conventions that are treated as if they are obvious, when they are nothing of the sort.

Marchant, Anastasi and Miller [2] used the term similarly in relation to doctoral students learning the hidden practices of publishing in journals. They argued that getting from thesis to article is complex, mysterious and often poorly taught at the micro level. Fredericks and colleagues [3] push the term harder, describing secret academic business as part of the unequally distributed rules of academic life: the things passed on to some and not others. This is the sharper meaning. Secret academic business is not just “tips and tricks.” It is the hidden curriculum of academic survival [4].

The hidden curriculum includes knowing how to read a call for papers, how to make a claim without sounding ridiculous, how to cite without looking like a tourist, how to write for reviewers, how to disagree politely enough to survive, how to signal membership of a field, and how to know when a paragraph is merely wearing an academic hat.

Some of this can be shared. We can teach writing. We can teach searching. We can teach source curation. We can teach the structure of an abstract, the anatomy of a journal article, the ritual phrases of peer review, and the deep sadness of “revise and resubmit.” But the shared version is usually an abstracted account. It is the recipe without the cook’s hands. It tells you the steps, but not the pressure, timing, smell, hesitation, and occasional panic that make the thing work. This is where LLMs become interesting.

LLMs appear to make some of secret academic business available. Ask one for an abstract and it will produce something abstract-shaped. Ask it for a literature review and it will produce a literature-review-like object. Ask it to make a paragraph more scholarly and it will put spectacles on the sentence, sit it near a window, and give it a faint air of methodological concern. This is not nothing.

For students, early career researchers, and people not already fluent in academic codes, LLMs can expose some of the choreography. They can show how a paragraph might move from context to problem to claim. They can generate examples of reviewer responses. They can help compare tones. They can explain what an abstract is doing. They can create a rehearsal space for academic moves that were once learned mainly through proximity, embarrassment, and a supervisor writing “unclear” in the margin like a tiny thunderclap.

LLMs do not reveal secret academic business so much as simulate its visible residues. They are good at the products of academic practice: the tone, the sequence, the shape, the gestures. But the more stubborn parts of academic work remain elsewhere. They remain in judgement.

Knowing that a source exists is not the same as knowing what it is doing in an argument. Knowing how to summarise a field is not the same as knowing where the bodies are buried. Knowing how to generate a plausible research question is not the same as knowing whether that question has legs, teeth, ethics clearance, and a chance of surviving contact with data. The machine can often produce the academic move. It cannot reliably know whether the move is warranted.

That’s important because secret academic business is not only about technique. It is also about taste, institutional memory, disciplinary feel, and embodied judgement. It is the almost physical sense that a sentence has become too smooth. That a citation is being used as wallpaper. That a concept has been asked to carry a piano up three flights of stairs. That the paper is pretending to have findings when what it really has is adjectives.

LLMs can mimic the forms of academic competence. Sometimes that mimicry is useful. Mimicry is how many people learn. We copy before we understand. We rehearse before we inhabit. We borrow the voice before finding out which parts of it make us stammer.

Mimicry becomes dangerous when it is mistaken for membership. The LLM can produce a passable academic surface without having undergone the formation that gives that surface responsibility. It can learn the handshake, the dress code, and several convincing noises of scholarship. It has not learned why everyone in the room is nervous.

There is a further complication. LLMs are now themselves becoming part of secret academic business. Many academics are using them. Some say so. Many do not. Or they disclose only in the safe, ornamental way: “used for editing,” “used for brainstorming,” “used to improve clarity.” These phrases may be true, but they are often as informative as saying that one used a word processor.

A student who uses a LLM may be asked to disclose, document, justify, and defend the use. An academic may use the same tool and call it workflow. The student “cheats”; the academic “iterates.” This distinction has all the moral grandeur of a parking permit. The asymmetry is worth naming. The machine has not ended secret academic business. It has been folded into it.

The new secret academic business is not only knowing how to write a journal article. It is knowing when to ask the machine, what to ask, what to ignore, what to rewrite, what not to admit, and how to make the final product look as if it arrived through respectable cognitive channels.

This is the bit that should make universities uncomfortable. Not because LLM use is inherently scandalous. It often is not. The scandal is the uneven honesty around it.

LLMs may also make some forms of secret academic business more teachable. Used well, they can turn hidden moves into objects of discussion. A supervisor can ask a student to compare three versions of an abstract and identify what each version is doing. A class can examine a LLM-generated literature review and ask where it becomes generic, where it overclaims, where it invents coherence, where it mistakes fluency for knowledge. A doctoral student can ask why one paragraph sounds like a grant application and another like a funeral notice.

In that sense, the LLM can become a “third thing” [5] in the room: not the student, not the supervisor, but a shared object around which judgement can be practised. This is the optimistic version. The LLM does not replace academic formation. It gives us something on which to practise formation.

That requires us to stop treating academic work as if its polished outputs are honest accounts of its making. They are not. Most academic writing is less like sculpture and more like plumbing done during a dinner party. Something has leaked, someone is pretending to remain calm, and by the end everyone agrees to call it structure.

The arrival of LLMs gives us a chance to talk more honestly about how academic work is made. Not just the official version: search, read, synthesise, draft, revise. But the actual version: drift, panic, overcollect, misread, imitate, cut, ask someone, move a paragraph, discover the argument halfway through, delete the clever bit, add the boring but necessary bit, and pretend this was the plan all along.

Secret academic business will not disappear. Some of it cannot be fully formalised because it lives in bodies, histories, institutions, disciplines, relationships, and scars. Some of it should be made explicit because secrecy protects advantage. Some of it will always remain craft. LLMs sit awkwardly across all three.

They expose some of the choreography. They cheapen some of the performance. They hide inside the workflows of those already fluent enough to use them well. They can provide rehearsal without apprenticeship.

So perhaps the question is not whether LLMs reveal secret academic business. They do, but only partly. Nor is the question whether they destroy it. They do not. The better question is: what kind of academic business becomes more secret when everyone has access to the performance of competence? My guess, for now, is judgement.

The future secret academic business is not “how to write an abstract.” The machine can already produce one, sometimes annoyingly well. The future secret academic business is knowing when the abstract is empty, when it is overpromising, when it has mistaken fog for complexity, and when it has made the research sound far more certain than the researcher has any right to be. That is not a technical skill. It is academic conscience with a red pen. And no, I don’t think we have been teaching that well either.

Notes

[1] Kamler, B., & Thomson, P. (2002). Abstract art or the politics and pedagogies of getting read. Australian Association for Research in Education Annual Conference, Brisbane, Australia.

[2] Marchant, T., Anastasi, N., & Miller, P. (2011). Reflections on academic writing and publication for doctoral students and supervisors: Reconciling authorial voice and performativity. International Journal of Organisational Behaviour, 16(1), 13–29. https://researchportal.scu.edu.au/esploro/outputs/journalArticle/Reflections-on-academic-writing-and-publication/991012820658002368

[3] Fredericks, B., White, N., Phillips, S., Bunda, T., Longbottom, M., & Bargallie, D. (2019). Being ourselves, naming ourselves, writing ourselves: Indigenous Australian women disrupting what it is to be academic within the academy. In L. M. Thomas & A. B. Reinertsen (Eds.), Academic writing and identity constructions: Performativity, space and territory in academic workplaces (pp. 75–96). Palgrave Macmillan. https://doi.org/10.1007/978-3-030-01674-6_5 

[4] I clumsily attempted to address this issue in a postgraduate course a long time ago.

[5] Rancière calls it the thing in common in Rancière, J. (1991). The Ignorant Schoolmaster: Five Lessons in Intellectual Emancipation. Stanford University Press.  

Bibs & bobs #38

 Give the bot a job

I bodysurf very occasionally now. More often than not the images I use for backgrounds have a breaking wave, an old bloke’s memory of times long gone. The images of the waves I use are not just any breaking wave, but one that arises when ocean swell meets an offshore breeze. That is what gives the wave its clean look. In Victoria, at 13^th beach that meant a northerly. 

The trick with bodysurfing is not to defeat the wave, or command it, or express your human agency over it. The trick is much less than that. You have to get your body to something like the speed of the wave. If the face is steep enough, you can swim down the front of it. For a few seconds, maybe longer, if you get it right, the wave picks you up and takes you to the shore.

You are not simply using the wave. You have entered a temporary association with it. The wave gives you speed. You give it angle, stiffness and direction of body and a contribution to its breaking. You become, briefly and wonderfully, part of the event. Then it might dump you or peter out.

This may or may not be a useful introduction to my ongoing puzzling, maybe obsession, about delegating work to nonhumans. When we act, we never act alone. We act with chairs, keyboards, calendars, forms, software, waves, doors, roads, recipes, school timetables, and badly designed online claim systems that appear to have been assembled by a committee of teaspoons.

The usual mistake is to ask whether these things “have agency”. That question sounds profound, but often produces epistemic fog. A better question is: what changes when this thing is brought into the arrangement?

A wave has no plan for me. The wave is not trying to transport an aged and creaky bodysurfer toward shore. It is not being helpful. It is not my aquatic intern. But in the right association, for a few moments, it does something with me that I cannot do alone, something almost magical.

This isn’t just true of water. It’s also what happens when I write with a language model. As I am doing now.

Large language models are nonhumans that arrive already over-imagined. They are miracle, fraud, parrot, alien, intern, oracle, plagiarist and that’s just before breakfast. This makes them difficult to make familiar, because they are strange from the start and seem good at remaining so.

Venkatash Rao’s line from a 2025 post is useful here: much of the public argument about AI gets trapped between hype and dismissal, whereas the more interesting stance begins with curiosity about what the technology actually does in practice. Rao’s post on mediocrity puts it nicely: current AI is neither a magical god-being nor a scam, but a useful technology that is likely to stay around and therefore needs practical attention.

That is where questions about giving the bot a job begins. We don’t ask is it intelligent, or is it conscious or will it replace us? The better question is what job have I given it, and what job is it actually doing?

If you do not give the bot a specific job, it will still do one. Usually the wrong one. It will become a fluent filler of space. It will fill in the social situation it thinks it is in [1]. It will smooth over uncertainty. It will produce the beige, plausible, laminated version of whatever you nearly asked for. It will be the person in the meeting who says, “That’s a really interesting question.”

It is not sycophancy so much as momentum. Fluency is the bot’s wave. Swim lazily and it may still pick you up, but not for long. That is why job specifications matter.

A language model can be given the job of summariser, critic, translator, adversary, explainer, pattern-finder, metaphor generator, copy editor, question asker, or first-draft donkey. Those are not the same job. They require different instructions, different checks, and different forms of human judgement.

When I ask it to draft, I am not asking it to think for me. I am asking it to produce material I can push against, rework, reorganise and be curious about why the model is taking a particular path rather than another.

When I ask it to summarise, I am not asking it to decide what matters. I am asking it to give me a crude map of the territory, preferably with the crocodile-infested swamp clearly labelled.

When I ask it to critique, I am not asking it to be right. I am asking it to be useful enough to annoy me, make me think harder about the basis of the LLM generated criticism.

This is why the current obsession with prompt tricks often misses the point. A prompt is not an incantation. It is more like body position on a wave. Too flat and nothing happens. Too late and you get smashed. Too theatrical and you are merely performing for the seagulls.

The better question is not “what magic words should I use?” but “what is the relationship I am setting up?”

There is a connection here to mediocrity. Rao’s argument [2], at least as I read it, is not that mediocrity means uselessness. It means that much of AI’s power sits in the ordinary, good-enough middle: the banal but powerful uses that become normal rather than mystical. The examples in the transcript of the interview are deliberately unglamorous: identifying electronic components, suggesting toy circuits, acting as a formulary for skincare experiments. These are not thunderbolts from Olympus. They are specific jobs.

The bot is better understood as abundant, cheap, fluent mediocrity. That sounds like an insult. It’s not meant to be. A great deal of work runs on mediocre competence: finding examples, making lists, checking consistency, generating alternatives, rewriting at a different level, producing something provisional enough to be improved. The danger is not mediocrity. The danger is mistaking mediocrity for judgement.

This is where taste enters the mix of ideas. Stephany Tyler’s post on taste is useful. She argues that in an age where AI can generate almost anything, the question shifts from “can it be made?” to “is it worth making?” It frames taste as discernment: the capacity to choose what matters when abundance becomes overwhelming. It’s not a matter of did the bot write this but of whether or not the text is any good.

That matters because language models collapse an old scarcity. Producing words is no longer difficult. Producing more words is trivially easy. Producing words with some shape, some pulse, some accuracy, some purpose, and some restraint remains most difficult. The scarce thing is not text. The scarce thing is judgement, or taste, if you prefer the less pompous word.

Taste is knowing when the draft sounds like a committee has swallowed a thesaurus. Taste is knowing when the bot has made your argument smoother but stupider. Taste is knowing when a sentence has been optimised into a beige corridor with handrails.Taste is knowing what to ignore. 

The “Taste Is the New Intelligence” piece [3] puts this as curation: when anyone can make anything, the live question becomes what to ignore, what to trust, and what to allow into your mental environment.

That is also true of working with LLMs. The machine can produce possibilities. It cannot care which possibilities belong in your work. It can imitate tone. It cannot know what you are prepared to stand behind. It can generate a paragraph. It cannot be embarrassed on your behalf, which is unfortunate, because embarrassment remains one of the great engines of improving prose.

So let’s give the bot a job. Give it a small, clear job that leaves you responsible.

For example, don’t prompt: “Write me something good about AI and education.” Instead, try prompts like these: “Give me three ways this paragraph is lazy” or “Find the hidden assumption” or “Rewrite this without the inflated claims” or “Generate five examples, but make two of them bad so I can see the difference” or “Act as a sceptical reader who thinks I am over-romanticising the wave metaphor.” That last one does hit home, the sceptical reader would likely be correct.

The bodysurfing metaphor only works if it keeps the wave dangerous. Bodysurfing has a built-in correction: the wave can smash you into the sand if you make a bad decision. Language models are much more polite. They often fail in ways that look and feel helpful. This is the problem. A wave that dumps you is honest. A bot that flatters you into publishing sludge is a menace in a beige cardigan.

All of this means is that the association has to be managed. The bot gets a job. The human keeps the judgement. The text becomes the site where both actors have left traces. A place where the human has some sense of capacities exchanged. 

I have made an argument about this elsewhere: when we delegate work to machines, we do not simply offload a task; we redistribute capacities, obligations, risks and forms of judgement between human and nonhuman actors.

That is the part I find interesting. Not whether the machine has agency “by itself”. Nothing has agency by itself. Not the wave. Not the swimmer. Not the keyboard. Not the school timetable. Not the LLM. Agency is not a private possession. It is an effect of association, of an association of many actors.

The question is what kind of association we are willing to enter. With a wave, the job is simple: catch it, hold the line, enjoy the brief borrowed speed, avoid the sand. With a language model, the job is harder: borrow some of the fluency and resist all of the sludge.

Do not worship the bot. Do not turn it into civilisation’s dark squid. Give it a bounded job. Then watch what it does.

                                                                                                    

Notes

[1] It is difficult to write about LLMs without anthropomorphising them. It’s also a dangerous crutch to rely on. 

[2] Sheffield, M., & Rao, V. (2025, July 22). Why mediocrity seems to be the key to innovation in evolution and technology. Flux. https://plus.flux.community/p/why-mediocrity-seems-to-be-the-key.

[3] Tyler, S. (2025, April 23). Taste is the new intelligence. WILD BARE THOUGHTS. https://wildbarethoughts.com/p/taste-is-the-new-intelligence.