With the emergence of the Deepseek R1 model, many of my previous assertions about large language models have been overturned. For instance, the development paradigm I used in "Alice Run," has already shifted.

Previously, when I presented a development request to a large language model, I had to trim my code, carefully extracting and laying out the core of the problem for the model to process. Now, I only need to paste several pages of code related to the business logic all at once, and the model can reference the relevant implementations on its own to complete the necessary development work. When using a relatively complex language like Rust, the model makes almost no mistakes. The few errors that do occur require only one or two simple corrections to produce highly usable results.

While marveling at the tremendous impact of the open source model on this generation of technology, as a writer in the field of education, I feel this is a good opportunity to discuss the influence of this technological leap on education and the changes we need to make.

Let's Start with Education

The Spiral Structure Hidden in Education

Before we begin this discussion, I want to start with a few questions: When we say we are learning, what are we learning? What have we learned? And what are exams truly testing?

The answer that might first come to mind is "learning knowledge," but I believe there is a higher dimension hidden above knowledge: the construction of cognitive abilities, which is the fundamental capacity to solve problems. Exams, then, are a means of selecting "individuals who possess the ability to solve complex problems."

From this perspective, education involves the enhancement of two intertwined dimensions: the accumulation of surface-level knowledge and the construction of deep-level cognitive abilities. These two dimensions do not exist in isolation but rather promote each other in a spiral structure. The accumulation of knowledge and skills is like pouring the foundation of a building, providing the necessary base for cognitive development.

To understand this process deeply, we must recognize that acquiring knowledge and skills is not simple information storage but a systematic process of capability construction. This process involves two key steps: first, internalizing knowledge into basic skills, and second, transforming these basic skills into tools for problem-solving.

In this transformation, repeated practice and application play a crucial role. Just as a skilled pianist needs countless hours of finger exercises, a brilliant physicist requires extensive problem-solving practice to build an intuitive understanding of physical laws. It is through continuous practice that we can convert surface-level knowledge into innate ability.

The improvement of ability is accompanied by a change in how we view problems. The process of learning is a series of problem-solving experiences, and every form of knowledge is necessarily nested within a specific problem framework. When we "master a piece of knowledge" through repeated practice, the corresponding cognitive process is this: you begin to accurately identify the specific problem contexts where this knowledge applies and understand its coordinates within the entire knowledge system. This is a process of first mastering basic skills, then breaking old cognitive patterns, and finally forming a deeper understanding.

This learning process is essentially about building cognitive surplus. When we internalize the multiplication table into an intuitive reflex or develop muscle memory in operating a microscope, our cognitive resources are freed from basic operations and can be directed toward higher-order thinking activities. These activities include identifying complex patterns, constructing deeper relationships between things, and exploring the unknown.

In the field of neuroscience, research also shows that through repeated practice, the corresponding neural connections in the brain become stronger, much like a path forming from walking the same route over and over. Repetitive practice creates specialized "expressways" in the brain. These pathways are located in the basal ganglia and help us turn learned skills into automated actions. It is like learning to drive: at first, you have to think, "press the clutch → shift gear → release the clutch," but with practice, it becomes second nature. This characteristic of the brain is known as neural plasticity.

This dynamic balance is reflected in education: the mastery of knowledge determines the lower limit of cognitive development, while the ability to explore the unknown determines its upper limit. In other words, a solid foundation of knowledge is the starting point for cognitive development, but the spirit of continuous exploration is the key to advancing it.

Conceptual Networks and Problem Spaces

Why is it that when you see an apple, you immediately think of the color red, a sweet taste, or even a certain phone brand? Why is it that when learning new information, connecting it to old knowledge makes it easier to remember? This is related to the way our brain stores knowledge: through a "conceptual network."

The organization of knowledge in the brain is much like a map. Each of your thoughts and pieces of knowledge is like a landmark in a city: some are tall towers (like "mathematics"), some are coffee shops (like "the aroma of coffee"), and others are bridges (like "friendship requires communication"). These landmarks are connected by countless roads. Some roads are wide and short (like "water" and "thirst"), while others are winding and long (like "quantum physics" and "cats").

This map, composed of "conceptual landmarks" and "relational roads," is the conceptual network.

When you see the word "dog," the "dog" concept node in your brain lights up as if a switch were flipped. At that moment, associated concepts like "pet," "loyalty," and "barking" also begin to glow faintly. This is activation and diffusion.

If the activation of "dog" is strong enough, related nodes will be triggered like a series of dominoes. For example, thinking of a dog might lead you to "walking the dog," then to "exercise," and then to "losing weight." This chain reaction is the root of your sudden flashes of insight when solving a difficult problem or the source of your creative ideas.

When you learn a new concept like "self-driving," your brain immediately connects it to existing nodes such as "cars," "artificial intelligence," and "traffic safety." The more knowledge you have, the more complex your map becomes, and the more routes you can take.

The brain periodically prunes unused connections, such as a forgotten phone number from a decade ago, while strengthening frequently used paths, like your daily phone unlock password. This decluttering makes thinking more efficient and prevents it from becoming a tangled mess.

All concepts exist to solve problems. Therefore, in a sense, a problem space can be seen as an abstraction of a conceptual space.

If the conceptual network is the "geographical map" of knowledge in your brain, then the problem space is the "maze navigation map" for solving problems. It is not content with static knowledge associations but dynamically constructs a battlefield to be explored: every fork in the road is a decision point, and every path leads to a different possibility.

Specifically, it is the mental structure formed when solving a particular problem. It contains the initial state of the problem (the starting point), the goal state (the destination), and all possible intermediate states and transitional steps. It is like solving a Rubik's Cube, where you must move from a scrambled state (initial) through a series of rotations (operations) to finally reach the solved state (goal).

Every step you take in the maze involves applying an operator to transform the current state into a new one. The whole process is like placing torches in a dark cave. Each torch can only illuminate a limited area, so we must keep trying different paths until we discover the one leading to the precious ore.

In master's and doctoral education, the construction of a problem space becomes a well-defined discipline. Researchers must read a vast amount of literature, trace the developmental trajectory of a problem in a certain field, identify the boundaries of the entire problem space, and then push those boundaries forward through their research, expanding human understanding of that field.

From this perspective, the conceptual space and the problem space share similar structures and properties.

The problem space relies on the conceptual space. When you face a new problem, your brain automatically activates relevant conceptual nodes. These existing knowledge structures help you understand the problem and plan a solution. For example, to solve a math problem, you need to draw upon relevant mathematical concepts and operational rules.

The problem-solving process, in turn, reshapes the conceptual network. Every time you successfully solve a problem, new connections are formed or existing ones are strengthened in your conceptual network. This explains why experienced experts can often find solutions to problems more quickly: their conceptual networks have already established more effective problem-solving paths.

These two spaces promote each other's evolution. When you encounter a bottleneck in problem-solving, you may need to learn new concepts to broaden your thinking. In turn, newly acquired concepts will help you discover new possibilities within the problem space. This virtuous cycle drives the enhancement of cognitive abilities.

The ability to explore a problem space is demonstrated by the adaptive adjustment of one's cognitive framework. When a doctor faces an atypical case, they need to utilize existing pathological knowledge while also constructing a new "symptom network." This is a classic example of skills and abilities advancing in an interactive spiral. This kind of "deliberately designed discussion of difficult cases" can disrupt existing cognitive frameworks, forcing the cognitive system to maintain a certain "elasticity" through reconstruction and pushing the boundaries of cognition.

The continuous expansion and reorganization of conceptual networks and problem spaces are accompanied by broader and more efficient problem-solving approaches. As the resolution of specific types of problems becomes "automated," a form of "encapsulation" begins to emerge. Just as a software engineer transforms complex underlying logic into directly callable functions, this kind of efficiency-driven encapsulation is constantly appearing.

The Historical Process of Cognitive Encapsulation

Since the Industrial Revolution, technological development has shown a significant characteristic of cognitive encapsulation. From the invention of writing to the advent of artificial intelligence, every technological leap has meant encapsulating the knowledge and skills of a specific domain into a callable tool. This encapsulation has dramatically increased overall societal efficiency but has also quietly altered human cognitive patterns: cognitive frameworks that once required long-term training to acquire are now compressed into "magic little boxes" that are ready to use out of the box.

The printing press encapsulated the dissemination of knowledge into a standardized template. The internal combustion engine encapsulated energy conversion into a mechanical device. Press a switch, and the television turns on. Place an order on Taobao, and the goods are delivered to your door. We no longer need to know the mechanisms behind this modern magic for it to work. Every major breakthrough in history has built new cognitive interfaces, transforming complex systems into operable black boxes.

This encapsulation mechanism has created unprecedented cognitive efficiency. When a user can drive a train without understanding the thermodynamic principles of a steam engine, or when a programmer can write code without delving into the physics of transistors, social productivity increases exponentially. The emergence of large language models has pushed this encapsulation to its extreme: they can encapsulate millennia of accumulated grammatical rules, literary expressions, and logical reasoning into a probability model and a natural language interface.

This directly shakes the foundations of traditional education. Writing skills that once took years of training are now reduced to a game of combining prompts. This encapsulation no longer remains at the tool level but directly intervenes in the core domain of cognitive construction. When algorithms can automatically generate frameworks for academic papers and write literature reviews, the foundational value of writing training begins to erode.

The crisis facing the education system is essentially a systemic risk brought about by the inability to sustain the spiral structure. Large language models are not only taking over the knowledge application stage but are also eroding the foundational stage of cognitive development. The encapsulation of programming skills deprives learners of the opportunity to train their algorithmic thinking. The popularization of intelligent writing tools weakens the cultivation of logical construction abilities. This trend could lead to cognitive stagnation: when knowledge acquisition becomes effortless, the motivation to explore problem spaces will diminish.

A more profound crisis lies in the interruption of intergenerational cognitive transfer. The traditional master-apprentice system emphasized the transmission of a "craft," which was not just about teaching operational skills but also about cultivating a problem-oriented consciousness. The intervention of large language models allows a new generation of learners to potentially skip the necessary cognitive climb and stand directly at the top of a technological black box.

I have friends who teach in middle and high schools, and they are already complaining about their students using large language models to do their homework. There are also many news reports about universities banning students from using large language model services to complete their assignments. Meanwhile, on many e-commerce platforms, services to clean up traces of AI generation have quietly appeared. It is clear that a conflict has emerged, and a game of wits has begun between students and educators.

When knowledge memorization and skill training can be accomplished by calling an API, how can education continue? Or rather, does education still have a reason to exist?

Four Types of Problems

To answer the question above, I would first like to distinguish with you how students perceive the "problems" they face, or more specifically, academic problems.

We can divide them into three different categories: Troubles, Questions, and Topics.

A Trouble is the most primitive way a student perceives a problem. At this level, a problem is seen as an obstacle that needs to be eliminated quickly, similar to a situation like failing an exam. At the cognitive processing level, these situations often activate the amygdala's stress response, triggering an emotion-driven coping mechanism. When facing a "Trouble," students tend to adopt avoidance and quick-fix strategies, such as copying homework, cramming at the last minute, or seeking shortcuts. The student is more focused on how to escape the current predicament as soon as possible, prioritizing the use of previously successful experiences, such as directly applying a formulaic template, rather than understanding the problem itself.

In this new era, this tendency may become even more pronounced. When students view their homework as a "Trouble," large language models are perverted into cheating tools. They might have the model write an entire essay or generate answers that they then slightly modify to evade detection.

This behavior not only undermines the fundamental purpose of education but also leads to a decline in cognitive ability, because in the process of avoidance, students lose the opportunity to build their knowledge systems through thinking and practice.

A Question represents a higher level of cognition. At this stage, students have realized that a problem is a puzzle to be solved, not merely an obstacle. They begin to focus on problem-solving methods and methodologies and try to establish connections between different pieces of knowledge. This type of problem corresponds to a standardized cognitive framework, and its solution path has been pre-encoded by the education system. The learner's core task is not to explore but to reproduce, as typified by textbook examples and standardized test questions. Here, the hippocampus and the prefrontal cortex work in coordination, primarily activating pattern recognition and procedural retrieval functions.

These students will think about things like, "What knowledge point is this question testing?", "What formulas are needed to solve it?", and "How are similar questions solved?"

When using large language models, these students tend to use them as auxiliary tools. They might ask the model to explain a difficult concept or to verify whether their own problem-solving approach is correct. This method of use preserves a space for independent thinking but remains confined within the framework of a predetermined answer.

The student's goal is to "get this question right," not to see the broader landscape of knowledge through the question.

A Topic is the highest level of problem perception. When students view a problem as a topic for research, they begin to focus on the principles behind the problem, its developmental context, and its deeper connections. These students will ask, "Why does this problem exist?", "How is this problem related to other fields?", and "Are there better solutions?" They are no longer satisfied with finding the standard answer but see every problem as a starting point for exploring knowledge.

For these students, large language model tools become powerful research assistants. They use these tools to gather information, organize ideas, and explore possibilities, but they always maintain a critical mindset. They understand the limitations of these services and know that technology is a tool to assist thinking, not a shortcut to replace it. In this process, they not only acquire knowledge but also cultivate the ability to conduct independent research.

These three cognitive approaches form a progressive chain. Moving from "Trouble" to "Question" and then to "Topic" reflects the deepening of a student's cognitive level and also reveals the true goal of education: to cultivate the ability for continuous exploration.

As we discussed earlier, the purpose of education is to cultivate the ability to explore boundaries. These boundaries can be the limits of the self or the limits of human understanding of the world.

This classification of problems is very important because the value of a technological tool depends on the user's cognitive level. When a student consistently perceives problems at the "Trouble" level, even the most advanced technology can only become a tool for opportunistic shortcuts.

More importantly, these three cognitive approaches determine a student's position in the dual spiral of knowledge accumulation and ability development. A student who sees problems as "Troubles" will struggle to form a systematic knowledge structure; their learning is often fragmented and short-term. In contrast, a student who reaches the "Topic" level can continuously refine their cognitive framework during their exploration, creating a virtuous cycle.

Finally, there is one more type of problem that is even more troublesome: viewing these three categories of "problems" as wrong.

When educators treat the three natural states of a student's engagement with problems ("Trouble," "Question," and "Topic") as Wrong, the education system irreversibly distorts its fundamental logic of existence.

When a student sees an exam as a "Trouble" to be avoided, teachers often resort to moral punishment instead of cognitive guidance. When a student fixates on finding the optimal solution at the "Question" level, the education system dismisses it as a lack of innovative spirit. When a student attempts to reconstruct a knowledge framework from a "Topic" perspective, they are questioned for deviating from the standardized curriculum.

We find that what such an educational system truly negates is not specific behaviors but the natural manifestations of the innate human drive to learn at different developmental stages. When an education system uses a single correct paradigm to constrain a student's way of thinking, it is essentially treating an organically growing living being with an "inorganic" mindset. After all, it is the large language models that are the "silicon-based lifeforms"!

It is this kind of arrogant certainty and the resulting prejudice and attacks that constantly stimulate the students' amygdala, turning the "love of learning" into something that only exists in fairy tales.

Affection

In my book, A Survival Guide for the Modern Student, I once cited the idea that "if a student enjoys learning, there's probably something a little off about them."

The logic hidden behind this is that the learning environment most students face is filled with suffering. To enhance their cognitive abilities, students must repeatedly push their cognitive systems to their limits, which is often accompanied by exhaustion and mental fatigue.

But something about this thinking seems off. The experience of suffering is also present in other activities, yet many people seem to enjoy exercise, feeling an unprecedented sense of exhilaration after pushing themselves to the point of exhaustion.

Why is this?

What is Affection?

To answer this question, let's start with the four neurotransmitters in the brain that govern "happiness": dopamine, serotonin, endorphins, and oxytocin.

Dopamine is the most familiar "reward chemical." It is released when we anticipate or actually receive a reward. When we complete a challenging task or achieve a goal, the brain secretes dopamine, making us feel a sense of accomplishment and pleasure.

Serotonin is the "mood stabilizer." It regulates our emotions, sleep, and appetite. Adequate levels of serotonin make us feel calm and content. Exercise, sun exposure, and good sleep all help to increase serotonin levels. When serotonin is too low, people are more prone to anxiety and depression.

Endorphins are the body's natural "painkillers." They are released in large quantities after strenuous exercise. This explains why we feel invigorated after a workout, as endorphins not only relieve pain but also produce a sense of euphoria.

Oxytocin is often called the "bonding hormone" or the "cuddle hormone." It plays an important role in close relationships and social interactions. When we receive recognition from others and feel a sense of belonging, our oxytocin levels rise.

When we say we "like to do something," what we actually mean is that we have found a "reliable way to promote the secretion of these neurotransmitters" through the things we enjoy doing.

Everyone's values are different, so the types and amounts of neurotransmitters secreted when completing different tasks also vary. This is the deeper reason why "different people have different tastes."

Why do most students struggle to experience this kind of pleasure in learning? The key is that the current education system has not established an effective "cognitive stress-reward" cycle. On the contrary, it creates a kind of "anti-human" experience.

When students are forced to deplete their dopamine reserves through mechanical repetition, suppress their serotonin secretion in a high-pressure competitive environment, and block the flow of oxytocin through isolated study, learning becomes a punishment system at a physiological level.

Take math practice as an example. In a traditional classroom, students often have to complete dozens of homogenous calculation problems. This process activates the automated processing mode of the basal ganglia, not the creative thinking of the prefrontal cortex. Dopamine is released only at the moment the homework is finished, creating a disconnected experience of "pain at the start, pleasure at the end."

The pleasure derived from exercise, however, follows a completely different neural mechanism. When a person engages in strength training, their muscle fibers sustain microscopic damage. This damage, along with the stimulation of the exercise itself, triggers a series of physiological responses.

These include the release of endorphins, especially during high-intensity exercise, which can help alleviate pain and discomfort. Completing training goals, feeling a sense of progress, and anticipating the results of the training can all stimulate dopamine release, bringing pleasure and a sense of accomplishment that reinforces the behavior. Exercise also increases serotonin levels, improving mood and reducing anxiety and depression.

This system, which is designed in accordance with neural principles and provides rewards for the same behavior across different time dimensions, makes exercise a sustainable and pleasurable experience.

In a traditional classroom, the cognitive stress students endure often comes from external evaluations (exam scores) rather than from an intrinsic desire to explore. It is like fumbling around in a dark cave; students neither know where their "cognitive load threshold" is nor do they receive immediate positive feedback.

Considering that "different people have different tastes," it is difficult for educators to design learning plans for every single student that can precisely promote the secretion of "happy neurotransmitters."

However, large language models, being relatively inexpensive and highly intelligent, offer a possible direction. There are different solutions for different educational scenarios. We cannot lay out a complete picture in a short article of a few thousand words, but I can give you an example that may offer some inspiration.

Constructing a Problem Space

A while ago, I was discussing the topic of problem spaces with a university professor. She mentioned that her child is at an age of explosive curiosity, asking all sorts of wonderful questions every day. I thought this was a great opportunity, as large language models could serve as a powerful educational aid. So, I suggested that she collect every question her child asks, find the answers online, and then use a model to transform them into children's stories, add pinyin, and print them out to read together with her child.

A very important point was to print them on loose-leaf paper, not on standard A4 sheets. Every so often, she could reread the new knowledge with her child, then take all the materials, spread them out, and try to organize them in different ways.

In the field of education, tools like Thinking Maps already exist to help children organize information more systematically. A parent can act as a guide, working with the child to construct their very first "problem space."

Thinking Maps categorize basic human thought patterns into eight types, each corresponding to a specific visual representation:

• A Circle Map, using concentric circles, helps learners define and understand the core meaning of a concept and its related associations.
• A Bubble Map, with a central radiating structure, guides learners to think about the specific characteristics and attributes of a subject.
• A Double Bubble Map, using two core bubble structures, compares the similarities and differences between two concepts.
• A Tree Map, with a tree-like structure, shows the hierarchical relationships between concepts, helping learners understand classification systems.
• A Flow Map, using a linear flowchart, displays the chronological sequence of events or steps in a process.
• A Multi-Flow Map, with a two-way flow diagram, helps analyze the causes and effects of an event.
• A Brace Map, using a bracket-like structure, shows the relationship between a whole and its parts.
• A Bridge Map, using a bridging structure, establishes analogical relationships between different concepts.

This method of making the "problem space" visible helps children naturally form a cognitive framework during the process of acquiring knowledge. When children ask questions, they not only get specific answers but also learn how to organize and connect these questions.

Let's illustrate this with a concrete example.

Suppose a child asks, "Why is the sky blue?" This question can trigger a series of explorations: starting from the refraction of light, moving to the composition of the atmosphere, and then extending to the unique characteristics of Earth's environment.

Using Thinking Maps, we can help the child connect these scattered pieces of knowledge into a preliminary cognitive network. The next time the child asks a question related to weather, optics, or the Earth's environment, they will subconsciously integrate the new knowledge into their existing framework.

In this process, the large language model plays a unique role. It acts like an experienced children's book author, translating complex scientific concepts into language that a child can easily understand. It can also write personalized learning materials based on the child's cognitive level and interests. This instant knowledge transformation and customization allow parents to better seize educational opportunities, turning a child's spontaneous curiosity into a systematic learning experience.

Innovation is an Illusion

When we talk about these "fashionable educational methods," many people might label them as "innovative." However, I do not subscribe to the concept of "innovation." I even believe that "innovation" is fundamentally a pseudo-concept.

Please do not misunderstand me. I am not denying the creation of new things. Instead, I am calling for a re-examination of our understanding of the nature of innovation. On the surface, technology is advancing at a rapid pace, with new products and methods emerging constantly. But if we delve into the level of problem spaces, we will find that these "innovations" are often just recombinations and optimizations of existing solutions within a relatively stable problem framework.

The myth of innovation crumbles when we misunderstand cognitive development.

From the horse-drawn carriage to the automobile, and from the automobile to the airplane, these appear to be revolutionary innovations. However, the core problem they solve has always been "how to achieve the efficient movement of people and goods." This fundamental problem space remains relatively stable, and technological evolution is simply the exploration of better solutions within this space. Even the latest self-driving technology is essentially still within this problem framework, seeking a better solution by incorporating new technological capabilities.

The shackles of a fixed cognitive framework often stem from a confusion between the "conceptual network" and the "problem space." Examining a constantly growing "problem space" through a fixed "methodological framework" is bound to be disorienting. The misapplication of innovation in education also stems from this spatial cognitive bias. Emphasizing surface-level methods like "divergent thinking" and "brainstorming" often overlooks a systematic understanding of the structure of the problem space.

True creative thinking is built on a deep understanding of spatial constraints, much like a Go master who creates new formations within the limits of the rules, rather than arbitrarily changing the rules of the game. The former is a breakthrough of one's own cognitive boundaries, while the latter is just a pointless self-indulgence.

From this perspective, "innovation" is not about conjuring up a fantasy world full of rainbow-colored bubbles out of thin air. Rather, it is:

• An accurate grasp of the problem space: True innovators are often able to accurately understand the essence and boundaries of a problem and see the evolutionary direction of the problem space.
• The reconstruction and optimization of solutions: Based on an understanding of the problem space, they recombine existing technologies and methods to find a better solution path.
• A keen perception of the space's boundaries: They are able to promptly sense new constraints and opportunities brought about by environmental changes and adjust their solutions accordingly.

I hope this understanding can help you break free from the mystification of innovation.

So-called "innovative geniuses" are often those who can keenly perceive the evolutionary patterns of a problem space and are adept at integrating existing resources. Their success is not an accidental flash of inspiration but is built on a deep understanding of the problem space.

In this sense, true educational innovation is not about shattering existing cognitive frameworks but about helping learners build more flexible cognitive structures. Such structures can continuously adjust as the problem space expands, maintaining the upward spiral of both knowledge accumulation and ability enhancement.

Oh, Philosophy! Oh, Philosophy!!

After deconstructing and re-articulating this entire problem space, we must confront the most fundamental question: As probability models continue to push the boundaries of cognition, where should the threshold of human wisdom be set? When large language models encapsulate knowledge into instant meals, and when algorithmic recommendation systems take over the selection of our cognitive paths, humanity faces an existential dilemma. The more conveniently we obtain answers, the harder it becomes to answer the fundamental question of "why we ask."

The answer to this question will determine whether education, this torch-passing ceremony that has lasted for millennia, will continue eternally or come to an abrupt end.

In this final section of the article, I would like to propose a more fundamental point, one that is widely lacking in modern education: philosophy.

Philosophy is the study of fundamental questions about existence, knowledge, truth, morality, beauty, and the mind. It not only focuses on "how to think" but also seeks to answer "why we think" and "what is the meaning of thinking." These reflections on the world around us are the necessary foundation for advancing our ability to "see" the essence of how things work (the problem space) and our own nature (as a thinker).

In an era where AI-generated answers are as natural as breathing, the absence of philosophical education is creating a silent cognitive crisis.

We are gradually being swallowed by a space "full of answers," immersed in nihilism. People are beginning to doubt their own value. People are starting to treat large language models as a religion, wildly swinging a great hammer and striking at everything that "looks like a nail."

When students are faced with a "perfect answer" generated by a large language model, the most dangerous cognitive trap is not a misunderstanding of the knowledge, but the illusion of understanding itself.

Philosophy can provoke thought, help us find a direction, and guide us to explore problem spaces and continue to break through boundaries. This training is not meant to deny the value of technological tools but to establish the subjectivity of the human thinker in human-computer collaboration.

As various large language models continue to break through technological ceilings, the most urgent task for human education is to cultivate the quality of "maintaining the courage to keep questioning in an ocean of answers."

It is this courage to explore the unknown and break boundaries, the joy that bursts forth in the brain at the moment a boundary is broken, and the strength to follow our hearts and keep exploring that allows us to stand here as human beings.

At the end of my book, A Survival Guide for the Contemporary Student, I left these words: "Do not stop thinking."

These four words, to this day, still shine with light.