A good AI teacher is not in a hurry to give answers.

This article was originally published in “Vista Newsletter (opens in a new tab)” No.160
Hello, hello!
This article is written on a rare Thursday.
There is nothing on the calendar today, but the classes for the next few days are very full.
So these days, I’m doing briefings almost all the time. In one day on Wednesday, the three briefing projects combined pushed out more than a dozen updates. When I reviewed it in the evening, I wrote a sentence: Finish the teaching briefing before class, and you can concentrate on teaching on the weekend.
In between working on these newsletters, I read a piece of news that I thought was worth sharing with you. It just made it very clear something that I have been talking about repeatedly in class for a while.
An AI that deliberately does not give answers
Last week, the 2026 Information Month and Taiwan Education Technology Exhibition (opens in a new tab) debuted at Taipei World Trade Center Hall 1.
Before the launch, Central News Agency reported on the AI learning partner “edegree (opens in a new tab)” that the Ministry of Education will display at the exhibition venue. This is a tool developed by Yincai.com, and four new functions have been released this time. There was a passage in the report that I read twice:
AI tools on the market often give answers directly, and the outside world is worried that students will become dependent on them. The approach of e-degree is that when students ask questions, it does not directly give answers. Instead, it uses examples, guides concepts, and combines explicit, implicit and detailed steps to build a learning scaffolding for students step by step.
Another new feature is to analyze each student’s learning stuck points based on their answering status, learning speed and common mistakes, and help students find where they are stuck.
I’m not going to comment here on whether this tool is useful or not, because I haven’t actually used it yet. But I very much agree with the design idea behind it: perhaps the most important ability of a good AI teacher is to know when not to rush to give answers.
This sentence sounds simple, but it is difficult to do. Because what AI is best at is giving answers quickly and beautifully.
The most common reminder for graduate students
This Tuesday, I posted a new article on the researcher base, titled “Use AI to find literature, don’t call it “give me ten articles” (opens in a new tab)>.
I wrote this article because I have seen similar scenes from my classmates too many times in class: open the AI tool and enter “Please give me ten documents on a certain topic” in the dialog box.
The AI is also very cooperative. The ten bibliographies are neat and tidy, and the author, year, journal, and volume are all available.
The problem is, you can’t find several of them in any academic database.
This is not me trying to scare you. There is a study published in Scientific Reports in 2023. GPT-3.5 and GPT-4 were asked to write a short literature review each on 42 topics, and then check 636 references one by one. As a result, 55% of the documents listed in GPT-3.5 were fabricated; GPT-4 also had 18%.
That was a model two or three years ago, and today’s tools have advanced a lot. But using AI as a database still has its own risks.
The AI is not broken, we misplaced it.
Think of searching for documents as drawing a star map
I used a metaphor in that article: literature is a star map.
Each paper is a star, and the citations between papers are connections. A group of papers that cite each other gathered together form a constellation, which is also a research field. What you have to do is draw this picture yourself. The most suitable role for AI is to be a guide who accompanies you to look at the map.
Specifically, I broke it down into five steps:
- Look at the panorama first. When you are in an unfamiliar field and you can’t even tell what to look up, first use a tool like Stanford University’s STORM and let it compile a long article with citations. This long article is only used to read the outline, trace the source, and circle unfamiliar terms, without direct quotation.
- Break down the keywords. This step is where general AI such as ChatGPT, Gemini, and Claude really come in handy. Ask it to help you break the question into several groups of English keywords, and then form a Bollinger search query.
- Search somewhere else. Check the search terms in academic databases such as ERIC, Taiwan Doctoral and Master’s Thesis Knowledge Add-On System, and Huayi.
- Follow the reference line. Pick a key paper as a seed and use Connected Papers to find its related research.
- Incorporate into the literature library. Put the found documents into Zotero (opens in a new tab), and put the core PDF into Gemini Notebook (opens in a new tab) to follow up, and click back to the original text for each answer.
I gave an example in that article: the title is “Will parents swiping their mobile phones affect the parent-child interaction of young children?”
Many people will directly use “parents”, “mobile phones” and “parent-child interaction” to translate them into English, only to find a bunch of irrelevant things. But if you ask AI to help break down the keywords, it will tell you that there is a special word in the academic world called technoference, which refers to the phenomenon of technological products interfering with human interaction.
Just knowing the word makes a completely different search result.
You see, it is also AI. One way to use it is to let it directly give you ten documents. What you get is a list that looks complete but cannot be trusted. Another way to use it is to let it tell you how to find it. What you get is a method that you can use yourself.
This is the same as e-degree deliberately not giving an answer.
stuck points, usually not where you think they are
Another new function of e-degree is to diagnose learning stuck points. After reading this, I thought about my own experience of taking classes in the past few years.
Where students get stuck is often not where they think they are.
Graduate students think they are stuck because they cannot find the literature, but in fact they are stuck because they don’t know what this phenomenon is called in the academic world. Office workers think they don’t know how to use AI, but in fact they don’t know what they want. The teacher thought he was stuck because he couldn’t write programs, but actually he was stuck because he was not sure which thing was worth entrusting to AI.
This is why, when I prepare lessons now, I spend a lot of effort thinking about: Where are these people most likely to get stuck?
This weekend through next Tuesday, I’ll be dealing with four very different audiences.
On Saturday at a certain university, I met a group of doctoral students from the Department of Education. The topic I want to talk about is AI-assisted literature exploration. Most of the trouble for doctoral students is not that they don’t know how to use tools, but that after using them, they don’t know how much to trust them. So that whole day, I will spend a lot of time checking and checking.
On Sunday at a nursing school, the audience was teachers and students with nursing background. The situational error tolerance of medical treatment is very small. Any suggestion given by AI must clearly understand its origin and limitations.
Next Monday and Tuesday, there will be a two-day teacher workshop, one day dedicated to AI academic research and writing, and the other day dedicated to Vibe Coding. I prepared a briefing for each of the two days. Teachers’ stuck point is usually time: they have to prepare lessons, correct homework, and deal with students’ affairs every day. There is very little time that can be used to learn new tools. Therefore, every exercise in these two briefings must be able to be used in your own classroom immediately after class.
In addition, I also made a rich presentation for a career lecture at a university of science and technology, titled “Stop Busting Around: Ten Habits Highly Effective Workers Practice”. The audience of this session are college students who are about to enter the workplace. Their problem is often that they have many things to do but cannot distinguish which ones are the most important.
Four types of listeners may encounter four different bottlenecks. The tools in the brief may overlap, but the approach is completely different.
I think this is the most difficult part of the teacher’s job to be replaced by AI. AI can organize the content very well, but to judge where the group of people in front of them are stuck and where to start talking, someone still needs to be watching their expressions on the spot.
Before proofing, the last chance to touch up
There is another very important thing this week: my new book “AI Unmanned Company” has entered the last revision before proofing.
The editor of the publishing house reminded me that after proofing, I could only correct typos but not polish the sentences. In other words, this was my last chance to tweak the book.
I must say that after the fourth or fifth proofreading, one gets tired. After reading the same paragraph many times, my eyes will automatically skip over familiar sentences, and my mind will keep thinking: This should be enough.
At this time, I will deliberately ask AI for help, but not ask it to help me revise the draft.
What I asked it to do was to find questions: Are there any contradictions between the statements in the previous and following chapters? The same noun is written differently at the beginning and at the end? Which numbers and dates need to be reconfirmed? For every item listed in it, I will go back to the original manuscript and read it before deciding whether to change it.
It’s up to me to decide whether to change it or not. What AI is responsible for is helping me see places that I can no longer see.
This is actually the same thing as the e-degree and literature star chart mentioned earlier. The most valuable time for AI is often not when it completes the work for you, but when it helps you see where you are stuck.
Tomorrow, share a book on the radio
Tomorrow morning, I will go to the radio station for an interview and talk to the host about a good book I wrote about last month: “Live a Year Without Relying on Humans”.
Author Joanna. Stern is a technology reporter for The Wall Street Journal. She spent a whole year handing over health, transportation, housework, work, and even psychological counseling to AI, tested more than 100 products, and interviewed nearly 200 people.
I wrote in the book review that the most worthwhile thing to take away from this book is not which tool is easy to use, but the three principles she set for herself: always test, set an evaluation baseline, and record the cost.
The third one is the one I quote most often in class. There must be a price for handing something over to AI. Sometimes the price is money, sometimes it’s time, and sometimes it’s your own ability that slowly deteriorates.
e-degree does not directly give students answers. To a certain extent, it records the cost for students: if they get the answer directly to each question, the price students pay is that they lose the process of coming up with the answers themselves.
If you haven’t read that book review, you can read it here:
Reflections after reading “Live a Year Without Relying on Humans”
Three things you can take away from this article
Finally, let’s sort out the key points of this article:
- Use AI to find literature. Don’t ask it to give you ten papers directly. Ask it to help you break down the keywords and search terms, and then check it yourself in the academic database.
- When you get stuck while learning something new, first ask yourself: Am I really stuck where I think I am? Many times, the real sticking point is a term you don’t know yet, or a need you haven’t made clear yet.
- When you get tired of revising, ask AI to help you find problems instead of asking it to help you revise. After you find the problem, it’s up to you whether you want to change it or not.
Recent events
Five steps to use AI to find literature, the full version is here
The literature star map mentioned in this article, the original text contains the current status of the tools for each step, prompt words that can be copied directly, and a two-and-a-half-minute teaching animation. If you have classmates who are writing papers, please feel free to forward them to them.
A complete set of four graduate student handbooks, available for purchase until December 31
“Gemini Notebook Research Workflow Manual”, “AI Academic Research Prompt Manual”, “Graduate Academic Briefing Package” and “Thesis Oral Examination Preparation Package” are purchased together for NT$2,399. From reading literature, writing and reviewing, to giving presentations and degree oral examinations, everything is ready in one go. The content of the four books is exactly the same as the single book, with free updates within one year, and the combined purchase price is only available until December 31.
December 20, Academic Research and Writing Workshop, early bird until November 29
If you are a graduate student or university teacher, on the morning of December 20 (Sunday), I will have a three-hour practical class in Taipei to teach you to use AI Agent as a research co-pilot to read faster, make better corrections, and at the same time maintain academic ethics. The verification and verification discussed in this article will be fully practiced in the workshop. Early bird pricing available until November 29th.
See the complete introduction and registration of the workshop
Seven-day challenge camp, registration deadline is November 28
The seven-day challenge camp starts on November 30th and registration closes on November 28th. One task letter a day, and in seven days you can make a small tool that you can use every day. You can put it on the Internet and get the URL without having to write a program.
See the introduction to the seven-day challenge camp (opens in a new tab)
New book “AI Unmanned Company” will be released in December
There is a five-question quiz on the new book page, and you can also sign up for release notifications, which will notify you as soon as the book comes out.
Finally, I want to hear your story
The last time you got stuck using AI, where was it? Looking back later, was it really the place you thought it was at the time?
Feel free to leave a message and tell me. See you next time!
Vista
Next step
Bring this topic to your team
Vista Cheng speaks and runs workshops on AI workflows, writing and content systems for companies, schools and public agencies.