Shigemi Hirata enrolled in university at eighty five. He graduated at ninety six, with a degree in ceramic art, and he is in the Guinness Book for it.
The number people quote is ninety six. The number that matters is eighty five, because that is the year he decided. Eleven years of turning up followed. He did not do it fast. He just did not stop.
That is the whole argument, and everything below is the practical half of it. The five from the video, then five more the video did not have room for, then the part that decides whether any of it survives the week: how to save them so they work in every new chat instead of only in the one where you first typed them.
Read this bit before the list, it takes twenty seconds. These are not buttons and they are not features. They are ordinary instructions that happen to be written with a slash in front of them, and they work because the AI reads the text and does what it says. That matters for one practical reason: an instruction only works as well as it is written, which is why the full version of each one is below, and why saving them properly is the difference between this working once and this working every time.
First, the honest bit
If you typed /quiz me and something reasonable happened, good. If you typed it and thought "that was just a normal answer," you were not doing it wrong. You had found the edge of the trick.
There is a real difference between two things that look identical on your screen:
- A real command. Built into the product. It appears in a menu when you type the slash, it does the same thing every time, and it is doing something the chat window cannot do on its own.
- An instruction wearing a command costume. Just text. Nothing pops up, nothing is registered anywhere, and it works purely because you asked clearly and the AI understood.
All ten below are the second kind. Which is not a problem, and I want to be clear about why: the second kind is most of what makes AI useful. Nobody built a "explain this to me like I am starting from nothing" button either, and that is one of the most valuable things you can type.
But it does have two consequences worth knowing:
- The wording carries the whole thing. A vague instruction gets a vague result. That is why each one below has a full version underneath it, and the full version is what you should actually save.
- It forgets. Type it in a new chat next week and you may get something different, because nothing has been stored. That is exactly what the saving section fixes.
For what it is worth, ChatGPT does have genuine slash commands, so the idea is not invented. Your own saved prompts appear in the slash list once you have saved them, which is the second saving route below, and some slashes open real built in tools. That is a much shorter list than the internet suggests, and none of the ten below are on it.
The five from the video, and what each one is actually for
Type the short version if you are in a hurry. Save the full version, because that is the one that behaves the same way every time.
1. /sticky-notes: get it down to something you can hold in your head
For the moment you have a chapter, an article or a training module in front of you and it is too big to think about. This breaks it into the smallest number of separate ideas it honestly has, and stops there.
The reason it works is not the notes. It is the last instruction in the full version, which forces it to tell you which two ideas everything else hangs off. That is the thing you were actually trying to find.
2. /question-predictor: find the part that matters most
Not everything in a syllabus carries the same weight, and the twenty percent that does tends to be the twenty percent that gets tested. This pulls that part out and tells you why it earned its place.
One honest limit, and it is in the wording of the prompt itself: nothing can predict a specific exam. What it can do is reason from how a subject is normally taught and assessed, which is genuinely useful and is not the same as fortune telling. If you are working towards a real certification, treat this as a prioritised reading order, not a leaked paper.
3. /study-guide: see it instead of reading it
The one people underuse, because they forget to ask for the picture. Ask for the visual explicitly and you get a diagram of how the parts connect, and then a walk through it one piece at a time.
This is the one to reach for when something has structure you cannot hold in a sentence. A process with stages. A system with parts that feed each other. Anything where "and then what happens" is the question you keep asking.
4. /quiz-me: the one that actually moves things into your memory
If you only use one of the five from the video, use this one. Reading something again feels like learning and mostly is not. Being asked and having to produce the answer is where it sticks.
The full version below carries the two instructions that make the difference: one question at a time, and the difficulty follows how you are doing. A list of twenty questions dumped on screen is a test. One question, then a response to your answer, then the next one, is a tutor.
5. /story-mode: the oldest memory trick there is
Turn the material into a story with people in it and a chain of cause and effect. People have been remembering things this way for as long as there have been things to remember, and it works particularly well on anything that is a sequence, or anything with a reason underneath it that a list flattens out.
The instruction ends by mapping each part of the story back to the concept it stands for, so you finish with the story and the actual material rather than just a nice story.
Five more that were not in the video
The five above are about getting material into a shape you can work with. These five are about something harder and more useful: finding out what you actually understand, and making it still be there next month.
If you only add one of these, add /teach-back. It is the most uncomfortable thing in this guide and it is the one that changes how fast you learn.
6. /ladder: find the exact rung where you get lost
You have probably tried asking for a simple explanation. The problem with a simple explanation is that it works. You read it, it makes sense, you feel like you have understood something, and you have not gained anything you can use.
The ladder fixes that by giving you the same thing at three heights: as you would tell a bright ten year old, as you would tell an intelligent adult who does not work in the field, and as one professional would say it to another. You read down until it stops making sense, and that rung is your gap, located precisely.
The instruction that does the real work is the last one. It asks which words appear at the top of the ladder that never appeared at the bottom, because those words are where the difficulty is hiding. Nine times out of ten, the thing you could not follow was one unexplained term doing all the load bearing.
7. /teach-back: the one that finds what you only think you know
This is the hardest one here and it is the one worth doing. You explain the thing, out loud or typed, in your own words, badly, without looking anything up. Then it tells you what you got wrong.
It is not the same as being quizzed. A question gives you a prompt and a shape to fill in. Explaining something from nothing gives you neither, and the gaps show up immediately. This is why people say they did not really understand something until they had to teach it, and it is named after the physicist Richard Feynman, who was famous for insisting that if you could not explain it plainly you did not have it yet.
The sharp edge is one specific instruction in the prompt: it has to tell you which parts you described using words you clearly cannot unpack. Repeating a phrase you have read without being able to explain it is the exact failure mode of reading something twice and feeling ready. You will not catch yourself doing it. It will.
8. /anchor: use the forty years you already have
Everything new gets explained to you as though you are starting from zero. You are not starting from zero. You have decades of pattern recognition from work, from running a household, from raising people, from money, from whatever you have actually done, and almost nobody teaching you anything is using it.
This one makes it ask what you already know first, and then explain the new thing entirely in terms of that. Not a generic analogy. Yours.
And it carries a safety instruction that most analogies do not: it has to tell you where the comparison stops being true. An analogy you trust past its breaking point is worse than no analogy, because you will confidently get the next thing wrong.
Worth saying plainly, since this whole guide sits under a video about being too old. This technique only works if you have a lot of life behind you to anchor to. Someone starting out at twenty two cannot use it. It is the one place where coming to this late is not the obstacle, it is the reason it works.
9. /space-it: the difference between a good session and actually knowing it
Here is the thing nobody tells you about a really good study session. It mostly wears off. Forgetting is not a personal failing and it is not an age thing, it is what memory does, and it does it on a fairly predictable curve.
The fix has been known for a very long time and it is boring: review the same material at widening gaps rather than all at once. Tomorrow, then a few days, then a week, then a fortnight, then a month. Each time you nearly forgot it and then dragged it back, it gets more permanent.
What makes this worth handing to AI is not the schedule, which you could write yourself. It is that the prompt makes it write the actual review questions for every future session now, weighted towards whatever you found hardest today. Future you does not have to prepare anything or remember what was difficult. Future you just answers eight questions and closes the laptop.
10. /traps: find out what everyone gets wrong, before you get it wrong
Every subject has three or four things that nearly everybody misunderstands on the way in, usually because the wrong version is more intuitive than the right one. They are rarely taught, because the person teaching you stopped making those mistakes so long ago that they have forgotten the mistakes were available.
This asks for them directly, and it asks for something more useful than a list of errors: why each wrong version feels right. Knowing that a misunderstanding is tempting, and knowing exactly what makes it tempting, is what lets you catch yourself in the act later.
Run this one early, before you have built anything on top of a wrong foundation. It is much cheaper than running it late.
Which one, when. Ten is a lot to hold, so: new material you cannot get through, /sticky-notes or /ladder. Too much material and not enough time, /question-predictor. It has parts that connect, /study-guide. You think you know it, /teach-back, then /quiz-me. It will not stick, /story-mode or /anchor. You want it still there next month, /space-it. Starting something new from scratch, /traps on day one.
How to save them, so this works in every new chat
Here is the part that decides whether you are still using these in a month. There are two routes and they do different jobs. Route one is the one I would set up today.
Route 1: Custom Instructions, once, and then they always work
Custom Instructions is a box that gets read at the start of every conversation. Define your ten shortcuts in there once, and from then on typing /quiz-me in any new chat does the full thing, because the definition is already in the room.
Go to Settings, then Personalization, then Custom Instructions, and find the box for how you would like it to respond. Paste this in underneath anything already there. It is long, and that is the point. You paste it once and never think about it again, and every one of the ten works from then on.
Paste into Custom Instructions: all ten, defined
I use a few shortcuts of my own. These are not features of yours, they are instructions I have defined. When I type one of them, follow its definition exactly. If I type one without naming a topic, ask me for the topic before you begin. /sticky-notes [topic]: Break the topic into the smallest number of genuinely separate ideas it has. Write each one as a single sticky note: a short heading, then no more than two sentences underneath in plain language. Define any technical term in the same note as the word. Number them. Do not pad the list to look thorough. At the end, tell me which two notes everything else depends on, and why. /question-predictor [topic]: Identify the material within this topic that matters most and is most likely to be tested or asked about. Order it most important first. For each item, give me one line on why it carries that weight, whether that is because everything else builds on it, because it is the part people most often get wrong, or because it is where the reasoning of the subject actually lives. Be explicit that you are inferring from how this subject is normally taught and assessed, and that you are not predicting any specific exam. Then tell me what you have deliberately left out, so I can decide whether I agree. /study-guide [topic]: Give me a visual first. Produce a diagram showing how the parts of this topic connect, using an image if you can generate one, and a clear text or ASCII diagram if you cannot. Then walk me through it one element at a time, in the order someone learning it should meet them, and say what each connection means. End with the single most common misunderstanding about how these parts fit together. /quiz-me [topic]: Ask me ONE question and then stop and wait for my answer. Do not give me a list. After each answer, tell me whether I was right, and if I was wrong, explain the actual reasoning rather than just correcting me. Start straightforward and get harder as I get them right. Get easier if I miss two in a row. Mix recall questions with questions that make me apply the idea to a situation. Keep going until I say stop, then summarise what I kept getting wrong and what I should look at again. /story-mode [topic]: Turn this material into one continuous story with characters and cause and effect, where the events happen in the order the concepts actually depend on each other. Keep it concrete and keep it moving. Do not use a metaphor that breaks down the moment it is examined. When the story ends, list which part of the story stands for which concept, so I have both. /ladder [topic]: Explain this to me three times, at three levels, one after the other. Level one, as you would explain it to a bright ten year old, with no jargon at all and nothing assumed. Level two, as you would explain it to an intelligent adult who does not work in this field. Level three, the way one person who works in this field would say it to another, using the real terminology. Then, and this is the part I actually want: list every term that appears in level three but never appears in level one, and explain each of those terms on its own. Those words are where the difficulty is hiding. /teach-back [topic]: I am going to explain this to you in my own words, from memory, without looking anything up. Do not help me, do not correct me, and do not add anything while I am doing it. Just wait until I say I am done. Then tell me, in this order: what I got right, what I got wrong, what I left out that actually matters, and which parts I described using words I clearly cannot unpack myself, where I repeated a phrase without demonstrating that I understand it. Be direct about that last one, it is the whole reason I am doing this. Finish by asking me two questions about the weakest part of my explanation. /anchor [topic]: Before you explain anything, ask me what I have spent my working life doing, and one or two other things I happen to know well, including things that are not professional. Wait for my answer. Then explain this topic using analogies drawn only from what I told you, not generic ones. For every analogy you use, tell me explicitly where it stops being true and what it would lead me to get wrong if I pushed it too far. I would rather have three analogies with their limits marked than ten that quietly break. /space-it [topic]: I studied this today. Ask me what I covered and which parts I found hardest, then wait. After I answer, build me a review schedule at widening intervals: tomorrow, then three days, then a week, then two weeks, then a month. Give me real dates counting from today so I can put them straight into a calendar. For every one of those sessions, write the actual review questions now, five to eight per session, no more than ten minutes of work each, weighted towards the parts I told you I found hardest. I do not want to have to prepare anything on the day. /traps [topic]: What do people most commonly get wrong about this while they are learning it? For each one give me four things: the mistaken version, why it is tempting and what makes it feel correct, what is actually true, and one specific question I can ask myself to catch myself making that mistake. Then ask me what background I am coming to this from, and tell me which of these traps I am personally most likely to fall into and why.
Two things to know about that box. It applies to new conversations, not the one you already have open, so start a fresh chat to test it. And it is shared across everything you do in there, so if you notice it being oddly studious about something unrelated, that is why. Everything above is written to sit quietly until you actually type one of the shortcuts.
Route 2: Save them as prompts, and they appear in the slash list
The second route is the one that turns these into real commands, in the only sense that counts. Saved prompts genuinely do appear when you type a slash, under your own name for them.
Save each of the ten definitions above as its own saved prompt, named the way you want to type it. From then on the slash menu is doing the remembering, and you pick from a list rather than trusting your memory of the wording. With ten of them, this route earns its keep.
Which route is better depends on you. Custom Instructions is set and forget, and it works even when you have forgotten the exact name. Saved prompts keep your settings clean and put the list in front of you. There is no reason you cannot do both, and if you are the kind of person who will forget these exist by Thursday, do both.
And one that is a real feature
Study Mode is built in, it is genuinely a feature rather than a prompt, and most people have never opened it. It is worth knowing about precisely because it does the thing the ten above are imitating: it walks you through a problem instead of handing you the answer, and it will not simply give you the solution when you ask for it.
The shortcuts and Study Mode are not in competition. Use Study Mode when you are working through something properly and want to be made to think. Use /quiz-me when you have twenty minutes and want to find out what has actually stuck.
Your first session, today
The gap between reading this and using it is where most people lose it, so make the first one small and make it now.
- Paste the Custom Instructions block in. Two minutes, and it is the only setup step in the whole guide.
- Open a new chat and pick something you already half know. Not the hardest thing you are avoiding. Something you have read once and could not explain to somebody else.
- Run
/sticky-noteson it. Five minutes, and now you know what the material actually is rather than how long it is. - Then
/teach-back, and do it properly. Explain it without looking. This will be uncomfortable and it is supposed to be. It is the fastest honest read you will get on what you actually know, and everything after it is aimed at what it finds. - Take the weakest part it identified and run
/anchoror/story-modeon that part only. Not the whole topic. The bit that fell over. - Finish with
/space-it. One minute, and it writes every review session between now and next month so you never have to plan one. This is the step that decides whether today mattered.
That is about half an hour and it is a complete loop: find out what it is, find out what you have, fix the gap, and make sure it is still there in four weeks. Everything else in this guide is a variation on that.
Then do it again tomorrow with the next thing. That is genuinely all this is.
Hirata took eleven years and the eleven years were not the impressive part. The decision was. You are not too old. You just have not decided yet, and those are two completely different problems.
Somewhere to point all this at. If you want something to study rather than a subject you already have, I have a separate guide to free courses from Harvard, Yale and MIT that you can start this week without applying to anything.
Sources & Notes
- All ten shortcuts are prompts, not product features. This is stated plainly in the guide rather than buried here, because it is the single most useful thing to understand about them. They work because the AI reads the text as an instruction. Nothing is registered, nothing is stored, and nothing appears in a menu until you save it yourself using route two.
- Shigemi Hirata. Guinness World Records lists him as the oldest university graduate, awarded a Bachelor of Arts on a Ceramic Arts course at Kyoto University of Art and Design, Japan, on 19 March 2016, at 96 years and 200 days. He enrolled at 85, so the degree took him eleven years. The record was set in 2016, and records get broken, so if you find a newer holder, the argument does not depend on which person it is.
- Nola Ochs, for readers in the US. She held the record before him, graduating from Fort Hays State University in Kansas in 2007 at the age of 95, and went on to complete a master's degree at 98.
- On "the twenty percent that matters". That figure is a rule of thumb, not a measured result, and there is no study behind it. It is used here the way it is normally used, as shorthand for the fact that material is not evenly weighted. The prompt itself is written to explain its reasoning so you can judge the ordering rather than take it on trust.
- Nothing can predict a specific exam. The question predictor infers from how a subject is normally taught and assessed. That is a reasonable inference and a useful prioritisation. It is not knowledge of your paper, and the prompt is deliberately written to say so out loud each time it runs.
- Story as a memory aid is ancient and uncontroversial, which is all that is claimed for it here. No specific research finding or percentage is being asserted.
- Explaining something to find out whether you know it is commonly called the Feynman technique, after the physicist Richard Feynman, who was well known for insisting that an explanation you cannot give plainly is one you have not understood. The named four step "technique" as it circulates online is a later popularisation rather than something he published as a method, so it is described here as the idea it is, not as a protocol he wrote down.
- Reviewing at widening intervals is one of the most consistently supported findings in how memory works, going back to Hermann Ebbinghaus in the 1880s and repeatedly confirmed since. The specific intervals suggested here (a day, three days, a week, a fortnight, a month) are a practical starting shape rather than an optimal schedule, because the ideal spacing depends on the material and on how long you need to keep it. No percentage improvement is claimed, because the honest answer varies enormously.
- Connecting new material to what you already know is well established as an aid to retention, which is the basis of the anchoring prompt. The claim made here is deliberately modest: analogies drawn from your own experience are easier to hold than generic ones. The prompt is written to mark where each analogy breaks down, because an analogy trusted past its limit reliably produces confident errors.
- Forgetting is not an age problem, and nothing in this guide suggests otherwise. Material fades without review at any age. That is why the spacing prompt exists and why it is framed as ordinary maintenance rather than compensation for anything.
- Menu paths were checked in August 2026 and they move. If Settings, Personalization, Custom Instructions is not where you find it, search the settings for "instructions". The route matters, the exact menu path does not.
- Where these came from. The idea for the first five came from another creator's video, which presented them as though they were built in features. The descriptions here, the full instruction wording, the saving routes and the correction about what they actually are were all written from scratch. The second five are not from that source at all. They are long established study techniques (levelled explanation, explaining it back unaided, anchoring to prior knowledge, spaced review, and surfacing common misconceptions) written here as prompts, with the reasoning for each one given so you can judge it rather than take it on trust.