Attention Is All You Need

The email arrived at 4.47 on a Wednesday afternoon, which is roughly when all institutional crises announce themselves in the Territory — late enough that nothing can be done about them today, early enough that they will gnaw at you all night. The subject line was cheerful. “Quick favour!” it said, with the punctuational confidence of someone who has never had to deliver one.

“Quick Favour!”

I read it three times. I made tea. I read it again. Then I did what any professional would do when faced with a request that violated several laws of physics: I stared at the wall for forty minutes and considered a career in horticulture.

The exclamation marks were doing extraordinary work. They were the sole cargo of optimism in an otherwise structurally pessimistic communication. They were carrying the entire emotional payload of a request that, translated into its actual meaning, read: “We have decided that the most consequential technological transformation of our lifetimes can be conveyed in approximately the same amount of time as a long lunch, and we’d like you to make it sparkle.”

I closed my laptop. I opened it again. I considered, briefly, whether the sender had read any of my actual workshop materials, or whether they had simply absorbed the existence of the workshop osmotically — the way one absorbs the weather forecast — and concluded that surely the highlights would do.

This is not the first such email. It will not be the last. This is, I have come to understand, the fundamental shape of my professional life: an unending procession of cheerful requests that I deliver, in compressed form, content that loses approximately ninety percent of its meaning the moment you compress it.

The all-day workshop is loved. Everyone says so. People walk out at 4 p.m. with a kind of dazed gratitude — the look of a person who has spent eight hours in an industrial cold-storage room and is just remembering what their feet are for. They have experienced something. They have, in some cases, been moved. They have, in nearly all cases, formed at least one opinion they did not previously hold.

A one-hour version of that experience is not a workshop. It is a microwave dinner where the original recipe was eight hours of slow braising. You can call it the same dish if you like. It will not nourish anybody.

Did the Manhattan Project Have a Taster Session?

I want to ask a question, and I want everyone in the back to hear it.

Did the Manhattan Project have a taster session?

I have checked. There was no sixty-minute pre-conference webinar called So You’re Curious About Fission. There was no TLDR titled The Implications of Splitting the Atom: A Five-Minute Read for Busy Executives. Oppenheimer did not, to my knowledge, offer a “lite” version of the implications, with cocktails and a Q&A, sponsored by Lockheed.

This was, I would gently submit, correct.

I am not, to be clear, comparing AI to a weapon. I am comparing the magnitude of the technological displacement — the speed of it, the breadth of it, the way it reorganises every cognitive task that has previously been priced into white-collar wages — and observing that historically, when technologies of this scale arrive, we do not normally insist that the people most responsible for understanding them be brought up to speed during a single recurring catered lunch hour. We send them away. We make them read things. We assume that comprehension takes longer than digestion.

But it is 2026, and we have somehow decided that artificial intelligence is best understood at the same operational tempo as a gym induction.

A Brief Taxonomy of What Actually Happens in the One-Hour Webinar

I have delivered enough one-hour AI webinars to construct a small but reliable taxonomy of what happens inside them. I offer it here as a public service.

In the first ten minutes, the audience expects me to solve the academic integrity crisis. They believe — and I want to be tender about this, because they believe it sincerely — that I have, somewhere in my pedagogical satchel, a method for definitively detecting AI-generated student submissions, and that I am, for reasons of professional discretion, withholding it. I am not. There is no such method. The research has been clear for some time that humans are unreliable at detecting AI writing, and what people are actually noticing is the subset of students who are bad at using AI — a point made, with admirable patience, in a recent Inside Higher Ed piece by Justus and Janos. This is selection bias dressed up as expertise. I will say this aloud. The audience will look at me as though I have just confiscated their last seatbelt.

In the next twenty minutes, having absorbed that there is no shortcut to detection, the audience pivots to expecting me to teach them how to use the tools. This sounds reasonable. It is not, as it turns out, the question they are actually asking. The question they are actually asking is: “Can you give me the prompt that produces the right answer, the first time, every time, without my needing to think about it.” The answer to this question is no. The answer has always been no. The answer was no in 1995 when people asked the same question about Boolean search operators, and it is no now, and it will be no in 2030 when whatever we are calling these things is integrated into the kettle.

In the final twenty minutes, having absorbed that I will not be giving them the magic prompt, the audience reaches what I have come to think of as the despair phase. This is the part where someone — and there is always one — asks whether AI is going to take their job, and someone else — there is also always one — asks whether AI is sentient, and a third person asks whether they should be using ChatGPT or Copilot or Gemini, as though this is the kind of question that can be answered without three follow-up questions about what they are actually trying to do. I respond as best I can. I am, by this point, sweating slightly. The Q&A runs over. Nobody has been taught anything. Everybody is afraid.

This is not a successful intervention. This is a fire drill conducted in real fire.

On the Question “How Do I Use It?” Asked Sincerely, By Adults

There is a question I am asked, in some form, in roughly every professional conversation I have, and I want to address it here, in writing, so that I can stop addressing it in person.

“How do I use it?”

You ask it something.

That is the answer. That is the whole answer. There is not a more sophisticated answer hiding behind that one. There is no secret antechamber. You go to the chat box. You type a question. You read the response. If the response is wrong, you tell it the response is wrong. If the response is right but boring, you ask for it again, differently. If the response is full of em-dashes and you have decided that em-dashes are now incriminating evidence of AI involvement — congratulations, you have correctly identified an em-dash — ask it to write again without them.

If you cannot think of a question, ask it to help you think of one. Ask it why you cannot think of one. Ask it to write a 500-word personal reflection on why, in 2026, a fully grown adult with a postgraduate degree is sitting in front of a chat interface that responds to plain English and asking somebody else how to use it.

Then read the response. Then push back on the response. Then, when you have done that for approximately three hours over the course of a fortnight, you will have learned more about how to use it than any one-hour webinar can possibly convey, because you will have done the only thing that ever produces fluency, which is use the thing.

I am sorry to be the one telling you this. I genuinely am. There are days when I wish I could hand out the secret prompt. But the secret prompt is “ask it something, and then ask it something else.”

The Justus and Janos piece I mentioned earlier makes a related point that bears repeating: many of the AI-classroom tactics being earnestly recommended on teaching-and-learning websites — embedding hidden white-text instructions inside essay prompts, asking about events post-knowledge-cutoff, requiring “personal reflection” as though that defeats prompting — have not worked for more than a year. The advice persists because the people giving it have not actually used the tools. They learned a trick from a TikTok video in 2024 and they are still teaching it, with the wounded confidence of a person who knows, knows, that this one weird hack is going to save them. It is not. It was already failing when they first heard about it.

But you cannot learn this in an hour. You can learn it in approximately two weekends of curiosity, of which the institutional norm permits zero.

The Grown-Up Students

There is a category of person at every educational institution I have worked at, and I have a private name for them which I will now make public: the grown-up students. These are my colleagues. They are lecturers, mostly, and a sprinkling of administrators. They are intelligent, accomplished, often very good at their jobs.

They are also, almost without exception, treating the existence of generative AI the way one might treat a small fire in a neighbour’s house: with polite concern, a vague sense that someone ought to do something, and an absolute refusal to acknowledge that the wind is blowing in their direction.

These are the people who will not attend the all-day workshop. Who will not attend the one-hour taster. Who will not, when offered a personalised one-on-one session for free, find it in their calendars. They have decided that they will simply wait this out — the way previous generations waited out the internet, the smartphone, and Microsoft Teams — until eventually the institution does something so clumsy that engagement is forced upon them at a point well past the moment when graceful adaptation was possible.

These are also the people whose students are turning in lavishly ChatGPT-assisted assignments, complete with the conversational artefacts left over from the original prompt — the “Certainly! Here is your essay on...” preamble that nobody has bothered to delete — and these are the people who are then asking me, with the wounded sincerity of the genuinely surprised, what they are supposed to do about it. The answer, I am afraid, involves a workshop. A long one. The kind they have been declining for two years.

I cannot say any of this in a one-hour taster session. There is no time. There is barely time to introduce myself.

The Part Where I Stop Being Gentle (Briefly)

I want to be honest for a paragraph.

I am scared.

Not for myself. Not for the institution. I am scared for the people who are about to vote, and parent, and make medical decisions, and run small businesses, in an information environment where the average citizen’s primary news source is a stranger doing a fifteen-second monologue on TikTok over an autoplaying clip of Subway Surfers. I am scared because I can see, with the weary clarity of someone who teaches this for a living, that the gap between people who can pressure-test an AI output and people who cannot is going to become one of the great civic literacies of our century — and we are training one of those groups and not the other.

I am scared because the people most likely to fall for AI slop are precisely the same people who are confidently telling me they don’t need a workshop, they just need the highlights.

The one-hour taster is not just inadequate. It is actively dangerous — because it produces the worst possible outcome, which is the confidence of competence without the substance of it. A person who has spent zero hours on AI knows they don’t know. A person who has spent eight hours knows quite a bit. A person who has spent one hour believes they have understood the field, and believes it loudly, in meetings.

I would prefer the zero-hours person. They are at least teachable.

A Modest Proposal: The Drop-In Therapy Support Group

In the spirit of innovation — which is, I am told, one of our institutional values — I have a counter-proposal.

I do not want to deliver another one-hour AI webinar. I want to host a drop-in therapy support group. Tuesdays, 4 p.m., bring your own grievance. Open to: educators who are tired of students using AI badly, educators who are tired of colleagues pretending AI doesn’t exist, and any grown-up student who has reached the point in their week where they need to cry about it in a professional setting.

There would be a circle of chairs. There would be tissues. There would be a whiteboard with the words I CANNOT FIX THIS FOR YOU written across the top in permanent marker.

We would not solve anything. We would not generate any deliverables. We would not cover the highlights of anything. We would simply sit, briefly, in the shared knowledge that we are all professionals being asked to perform the impossible at an institutional pace that has not yet acknowledged the existence of the impossible. I would offer biscuits. The biscuits would be the deliverable. People would feel seen. KPIs would not be met. Everybody would leave slightly less alone.

This is, I want to stress, a serious proposal. I will be submitting it through the appropriate channels, which is to say I will be ignoring all the appropriate channels and doing it anyway, the way most useful work in regional Australian education has historically been done.

Attention Is All You Need

In 2017, a paper was published — Vaswani et al., now famously titled Attention Is All You Need — that introduced the transformer architecture. It is, more or less, the foundational paper of the entire modern AI era. Every chatbot you have ever spoken to, every image generator you have idly amused yourself with, every helpful and unhelpful and accidentally racist Microsoft product you have stared at — all of it descends, architecturally, from that paper. The title was a technical observation about a particular kind of neural network mechanism. It has, over time, taken on a second meaning.

Attention is all you need.

Pay attention. To this. To the thing happening in front of you, which is the most consequential technological transformation in working memory, and which is being treated, at most institutions, with roughly the same energy as an upcoming software update.

I do not have a one-hour version. I do not want to make one. I want to sit in a room with people for a whole day and let them feel the shape of what is changing, and then I want them to go home and use the tools, and then I want them to come back next month, and we will sit together again, and we will do the slow, unglamorous, long work of becoming literate in a new medium — the way every generation has done with every new medium since the printing press. There is no taster for this. There is no highlight reel. There is no I’ll just need the cheat sheet.

Attention is all you need. It is also, I’m sorry to say, the only thing you can give me. And it has to be more than an hour.

The email is still in my inbox. I have not yet replied.

I’m going to make more tea.