I'm sure you have, at some point, seen those little ads for "essay writing services" in the bathroom stalls of a university administration building. You and I should have a consensus on this business: having someone else complete your thesis is unethical. In an era where the AI storm is sweeping the world, the one doing the "ghostwriting" no longer needs to be a real person. A quick search for online large language model services will reveal a plethora of flashy websites.

Initially, schools maintained a high level of vigilance against this trend, banning students from using such services for homework and papers, and adding an "AI writing detection" step to the graduation thesis review process. However, as prompt engineering becomes more refined, and various new models and fine-tuned versions appear one after another, the early detection services designed for a single model like GPT will likely become unable to handle these complex situations. Furthermore, as the conversational style of large language models continues to permeate the internet, the writing style of readers will also begin to converge with that of generated text. All these changes point to a clear outcome: it is becoming increasingly difficult to determine whether the author of a piece of writing is a real person.

Oh, holy cow, it looks like the end of the world is coming! We can no longer prevent students from "slacking off and cheating on their homework"! The future of our nation has been defiled! The education system is going to collapse! Our future is doomed to be bleak! Humanity will soon be replaced by AI!

Wait a minute, wait a minute. Something's not right here.

If I understand correctly, the purpose of a tool is to solve a problem. But according to this line of reasoning, it seems like the tool is "solving" the person. What problem were we originally trying to solve when we developed AI systems in the first place?

This article is a prelude to "When Will the Homunculus in the Flask Awaken?," "The Next Step in Education," and "AI Will Not Eat You." I encourage you to read the first three articles before continuing with this one.

Solving Problems

Let's start with some simple psychological concepts. The reason we want to "do something" is because of the "motivation" behind it. What triggers motivation is a "need," and this "need" is a "problem" that we pose.

When you think of "motivation," an image of a wicked archvillain might pop into your head. In a castle glowing with purple light, he chuckles, rubbing his hands together with a sinister grin: "Aha! What is my motivation!" But in reality, motivation itself does not carry such a heavy moral weight.

If we ascend Maslow's hierarchy of needs, we find that "needs" simply reflect our everyday lives.

For example, it's late at night and you're hungry. Your need is "hunger," so you pose a question: "How can I fill my stomach?" You start to solve this problem. You open the fridge and find nothing appealing, so you grab a packet of instant noodles from the cupboard. You tear open the packaging, put the noodles, egg, and seasoning into a bowl, add some water, and microwave it for a few minutes. A fragrant bowl of noodles is ready. You enjoy your meal, then lie down comfortably in bed, ready for a good night's sleep.

As you toss and turn in bed, a question suddenly pops into your head: "Did I lock the door after taking out the trash this evening?" Now you have a need for "security." To answer this question, you get out of bed and check the door lock. "It really wasn't locked! Good thing I checked!" Your heart, which was in your throat, settles down, and you can finally sleep soundly.

In a state of half-sleep, a happy memory of eating cake with your friends comes to mind. You really want to hang out with your friends again. This is a need for "love and belonging." To answer this "problem," you think to yourself that you'll message your friends in the group chat tomorrow to arrange a meal and have that delicious cake again.

Beyond these, you also have needs to "relax," to "be seen, recognized, and feel a sense of accomplishment," to "explore yourself," and to "write poetry in the vegetable garden." Every action we take corresponds to a need. These needs are translated by the brain into motivations and are thus perceived by us in the form of "problems." Once the brain poses a problem, we must begin to "write the answer." If you have read the previous articles in this AI series, you will surely know the rest of the story, so we won't repeat it here.

So, for a student, what kind of need does "writing a paper or an essay" correspond to? If we think about this question from a "great, bright, and upright" perspective, the answer would surely be "to convey one's thoughts" and "to explore unknown knowledge." But the reality is often bleak. For many students, the problem they are trying to solve might be the fear of "failing a course" or the anxiety of disapproval from low grades. At this point, the problem becomes a trouble, and this kind of "misalignment of expectations" is a major challenge facing the current education system.

The book A Survival Guide for the Modern Student has systematically explained the reasons behind this phenomenon, but for the sake of a sound argument, we will briefly mention them here. The teaching scenarios in East Asian societies are obsessed with efficiency and scores, and they advocate educational methods centered on behaviorism, with "punishment" as the core of promoting learning behavior. Whether it is being forced to copy answers for getting a question wrong, running laps on the playground for a bad quiz score, or the head teacher peeking into the classroom through the back door window and publicly announcing students' scores to humiliate those with poor grades, all these methods, without exception, create an "unsafe" learning environment.

Why do educators do this? If you ask them directly, you might get answers like "I'm doing this for your own good," or "How can I manage so many students in the class?" But if we dig deeper, in the eyes of these educators, the real problem is no longer "how to improve students' abilities" or "how to help students discover what they love." The students themselves have become the problem.

When the problem we are trying to solve becomes the "person," things get complicated. This kind of educational space, with "naive behaviorism" at its core, attempts to satisfy higher-level motivations like "acquiring knowledge" and "exploring the self" by undermining lower-level motivations. Perhaps we should all sit down, calm down, and consider whether it is truly worth destroying these precious things for the sake of a few points on a test.

A Story

If your motivation for writing is to "get this homework over with quickly," you can turn left at the door and head to some big shot's AI course. A "prompt library" or a "universal prompt formula" will be more helpful to you in solving your immediate problem than this article. Those "prompts" can help a large language model figure out what a "person who can't ask a clear question" really wants, instead of getting frustrated with a "difficult" customer.

Please don't misunderstand; I don't mean to belittle any motivation here. Even I, while writing this article, have fallen into this "troublesome" trap to some extent. Hmm... that sounds a bit abstract. Before we continue, why don't I tell you a story about a "big trouble" I encountered?

I haven't been doing well lately. I've had sleep problems for several months now. Every evening, my smartwatch reminds me, "You've been under a lot of stress today, you need to rest early." But whether I go to bed early or late, I wake up the next morning feeling like death warmed over.

"Sleep quality: failing. Thanks!" After a half-hour workout in the morning, another day of "prefrontal cortex offline" begins!

I completed my article "The Next Step in Education" in this mental state. For those few days, I was outlining the narrative structure of the article while doing aerobic boxing. When I felt I had a good enough idea, I sat down at my computer to start writing. But my head was really fuzzy, and I didn't have the energy to write word by word. So, I wrote a very detailed outline, found reference articles, and fed them all to R1. It generated forty to fifty articles in one go. I picked out the usable text fragments and stitched them together with the outline to create a draft.

Next, for the parts that R1 couldn't explain clearly, I used other models (I mainly use Claude and Gemini¹) to write on a given topic, such as explaining the concept of a conceptual network more vividly or providing more examples for the "four happy chemicals."

It was a piece filled with R1's flavor, sparkling with the light of "ancient Greece" and "quantum mechanics." (If you've ever used DeepSeek R1 to write, you'll know that it brings ancient Greece and quantum mechanics into everything. Its entire family is ancient Greek quantum mechanics, and its quantum mechanics is absurd.)

Next came the revision process. I read through the draft from top to bottom, and wherever the logical connection was off, I started to patch it up myself. For some parts that were "troublesome" to fix, I handed them over to Claude. The specific method was to cut the logically disconnected part in half, give it to Claude to reconnect, and after generating three or four versions, I would pick the usable parts and reassemble them myself.

I used Bing Search and Perplexity to check the key neuroscience and psychology knowledge, and wherever I found an inaccurate statement, I made adjustments. After two rounds of this cycle, I did the final manual fine-tuning. I started writing around one or two in the afternoon and finished at midnight, which was a new record for me.

Finally, I read through the entire article and deleted everything that "I absolutely couldn't have written without an LLM" to uphold my value of "not pretending to be someone I'm not." In the end, the article's viewpoint was mine, the development of the viewpoint was controlled by me, and the core concepts of psychology, neuroscience, and education were all woven in by me. Therefore, I could be responsible for the article's viewpoint and accuracy.

I have to say, R1's writing is really strong. Thanks to its divine power, the "literal writing" of the entire piece was much better than my other articles. Also, because I allocated most of my attention to the overall structure of the content rather than the specific execution details, the logical flow of the argument was much smoother than in my other articles.

But when I shared the first draft with my friends, I received the feedback that it "lacked the Rori flavor." Specifically, it lacked the taste of that "typo-ridden smartass." After sharing the article with a wider audience, the feedback was that the article was obscure and not as easy to read as my other pieces. However, readers with a professional background in psychology said that the article was very well-written and a pleasure to read.

What exactly happened?

The Author's Responsibility

My answer is: I did not take "responsibility" for this article very well.

We act to satisfy a certain need. The action may be directed at our own internal state or at the people around us. The purpose of the action is always to "create an impact" (which is the sense of control mentioned multiple times in the previous articles). Once an impact is triggered, the accompanying responsibility arises.

When we write an article, what is our motivation? Barring any "misalignment of motivation," I believe it should be described as:

The author conveys a viewpoint to the reader.

Each syntactic component of this sentence corresponds to a responsibility of the author. The author needs to be responsible for the object, "a viewpoint." They must ensure that their viewpoint is clear and accurate, the arguments are smooth, and the deduction process is logically rigorous.

The author needs to be responsible for the adverbial target, the "reader." If the author's target reader is only themselves (like in a diary), then they only need to make sure they'll understand what they were on about years later. If the intended audience is professionals, you can significantly shorten the logical argument by using abstract professional terminology. But if the audience is the general reader, then more effort must be put into the writing: showing empathy, placing oneself back in the darkness of the unknown, and thinking about what the process of exploration should really be like.

Including myself, who penned "The Next Step in Education," many popular science writers can hardly be considered passing in this regard. And there are many reasons for this problem.

First and foremost is the "misalignment of motivation." If the author's primary goal in writing is not to convey a viewpoint to enlighten the reader, but to show off their knowledge, consolidate their expert status, chase clicks, earn a fee, or meet a quota, then the quality of the resulting text is naturally hard to praise. The focus of the writing often shifts from "what the reader needs" to "what I want." This deviation in internal motivation will inevitably be reflected in the final written expression, making it difficult for the article to truly reach and influence the target audience.

I have seen some written works where the author never even considered the reader from the very beginning. Such authors often overestimate the clarity of their expression and underestimate the difficulty of the reader's comprehension. We think we have made ourselves clear, but in reality, we are just talking to ourselves within our own mental framework, writing words that miss the mark.

There are also works that fail to form a clear picture of the reader from the outset. The writing style and narrative structure for a blog post, a journal article, a piece for middle and high school students, and an article for digital enthusiasts are all different. The background information the author needs to provide also differs.

This "inattention" to the reader, though it may seem "arrogant," is sometimes² not rooted in arrogance but in a kind of cognitive inertia. Our brains default to a "minimum energy consumption" mode of operation, and "empathy" happens to be a "high-energy" activity that requires active participation from the prefrontal cortex. From this perspective, people naturally tend to start from their own point of view, subconsciously using their own thought processes and language habits as a template for understanding others.

As the author of a popular science article, my failure in "The Next Step in Education" was that I wrote a draft with the mentality of "quickly jotting down my thoughts," but this draft was not responsible to the audience of a "popular science article." Specifically, the biggest failure in the article's argumentation was that a large number of examples were interspersed within the logical reasoning, breaking the coherence of the reading experience. The reason why people in the same profession could "read it with great pleasure" was that they could naturally skip the arguments while reading and did not need to expend cognitive resources to understand the obscure professional terms. The most brilliant part was that because the article was very long and R1's writing style gave the whole piece a grand "aura," those who finished it felt an exaggerated sense of accomplishment, which to some extent masked the major flaws in the argumentation.

I am very ashamed of the way I used examples in this article. Their existence is like a university professor who doesn't understand the subject but is forced to give a lecture. This poor soul can only read a passage from the textbook, give an irrelevant example, read another passage, and then give another example that is completely unhelpful for your understanding. What's even more frustrating is that I do understand this knowledge, but I forgot to use my best argumentative structure to deconstruct it: introducing a problem through an example, and then explaining the viewpoint by solving that problem.

But why did this happen? I think a very important point is that the author needs to be responsible for the subject of the writing act: the "author." An author really shouldn't force themselves to produce content when their prefrontal cortex isn't working well. This might sound like a platitude, similar to "to stay healthy, please get enough sleep." But when we introduce AI into the writing process, this sentence hides a deeper trap. In a state of a "disconnected prefrontal cortex," I completely lost my nuanced grasp of "reader empathy" and my control over my unique style—that personal imprint my readers call "Rori flavor."

Subjectivity

The other day, a reader raised a very interesting question: Is it you writing the article, or is the AI writing you? Good question. That's a really good question.

A terrifying image of living people being "possessed" by AI "spirits" came to my mind. As of early April 2025, we can still say with relative confidence that "AI has no personality." But in the face of increasingly powerful models, if the user fails to fully grasp their own subjectivity, they are very likely to become a puppet in the hands of these models.

It's not just artificial intelligence; many convenient modern magics have this quality, like short-form video platforms and social media. However, because technology is inherently neutral, we cannot say that these "contemporary magics" are "evil." But the more powerful the tool, the more thought we need to put into it, and the greater the responsibility we must bear.

For my own use of artificial intelligence, I have a few requirements for myself:

• Do not use them to solve problems beyond my own capabilities, because facing a mess I can't handle could easily lead to things spiraling out of control.
• Ensure that the questions I ask the AI are clear. If they are not clear, I will spend time clarifying them first. Because if I don't even know what I'm doing, how can I expect my audience to know what I'm trying to do?
• Sufficiently process and handle the final product, turning "what belongs to the machine" into "what truly belongs to me." If it's a writing task, I need to do as much fine-tuning and editing as possible. If it's an image generation task, I need to at least drag it into an image editing software and clean up the obvious flaws.

The core of these requirements is to maintain "my" subjectivity in the creation.

The exercise of subjectivity reflects our ability to have autonomous control over our own actions, thoughts, and creations. In the creative process, the expression of subjectivity means the author's control over the work—from conception to expression, from motivation to final presentation. The author is the initiator of the idea, the decision-maker, and the one who bears responsibility. Subjectivity gives a work its unique character, making "you" into "you," and making the work carry the author's way of thinking, perspective, emotions, and intentions, rather than just being a pile of words.

This is also the core of the "large model creation" ethics in my eyes:

"Your" creative act should be inspired by "the motivation for creation." Your purpose should be to convey thoughts, explore knowledge, and communicate with the reader.

"You" should be responsible for the viewpoint (clear, accurate, and logically rigorous), for the reader (considering the reader's understanding and having empathy), and for yourself (ensuring you have the ability to master the creative process).

"You" should lead the core of the creation. You should clearly know what problem you are using the large language model to solve, not just generate content aimlessly.

"You" should sufficiently process, modify, and verify this content to truly integrate it into your own thinking and style, and you should be able to take responsibility for the content's accuracy and viewpoint.

Although various models can quickly generate text and provide inspiration, they are ultimately not "you." I want to see "you," and many readers also want to see "you." No one wants to see a "mean model of a large sample." Perhaps R1 is very eloquent, the images drawn by gpt-4o-all are beautiful, and the speech and music generated by models sound good at first listen. But a large amount of content with the same flavor, "vomited" into every corner of the internet without thought and organization, will naturally make people feel sick.

As models become more and more powerful, people subconsciously let go of "me" and throw the entire creative process to the model. The tragic result is that the novelty of a new model, that "emotional dividend," has gone from lasting a few months at the beginning to a few weeks, a few days, or even a few hours.

Instant noodles are pretty good, but if you eat them every day, you're going to get sick of them. Not to mention that you're eating this "instant noodle" more than three times a day, more than six times, maybe even more than ten times. You open a video site, click on a random work, and the first four words are "Imagine if..." You immediately close the video because that's a unique turn of phrase of generated text. And an author who hasn't properly handled these "flavor compounds" is unlikely to have added much of their own story to the content.

Let's answer that question one more time: Why do you hate that AI-generated content?

Because I can't see "you." I want to see the questions "you" ask. I want to see "your" answers. Not seeing these things disappoints me. It makes me angry. It makes me feel disrespected.

How strange, how strange. Where are "you"? And where will "you" go?