Gaokao Is Over! McKinsey Warns 57% of Jobs Will Be Wiped Out by AI, and the More "Practical" Your Major, the Faster It Dies — Exposing the Study-Abroad Major Blacklist That Gets Rejected the Moment You Apply, the Hidden Landmine Universities Beyond the National Defense Seven, 5 Global Premium-Paying Occupations, and 4 Ultimate Meta-Skills That Ignore Your Diploma

This episode discusses the impact of the AI era on employment after the college entrance examination. A McKinsey report indicates that 57% of jobs will be automated by AI, with traditional practical majors and high-scoring conformists being the first to be affected. It also analyzes the hidden pitfalls of studying abroad (such as PP10043 visa restrictions, the "Seven Sons of National Defense" and entities on the export control list), the counter-intuitive phenomenon where the more practical the major, the faster it dies, and globally premium industries, while proposing four ultimate meta-skills for the AI era.

Video link:

Full text:

All things are as they are; reject fragmented cognition. The college entrance examination has ended. If your family has a candidate who performed well, well, congratulations. It's a good thing, right? But if you feel you didn't perform as well as you hoped, there's no need to feel any particular way. This thing's biggest impact on life is just an exam. Chinese teachers are particularly fond of putting pressure on students, saying this is a turning point in destiny. That's complete bullshit. Of course, this episode is not just for students after the college entrance exam; it's also helpful for those who are confused about choosing a career. A person's economic value is only related to their ability to meet market demand. Finding the entry point for value is the most important thing, and only then is it possible to enhance self-worth. The students cultivated by the CCP's current education system could still be worth something in the industrial era, working as programmers or in grassroots management. But the problem is, AI is here. McKinsey released an employment report in late last year, October 2025, which stated very plainly that with current AI technology, theoretically, nearly 57% of work time can be automated, and the hardest hit in this proportion are precisely those jobs with very clear rules and standardized processes. This part is precisely what those mass-produced by the college entrance examination assembly line are best at, such as accounting processes, data entry, technical coding, and formulaic writing. These four items are all among the top quarter with the highest automation exposure in the 6,800 skills scanned by McKinsey. To put it bluntly, AI will first consume those mass-produced by the college entrance examination assembly line. Are there any signs of this in the numbers? Indeed, there are. Look at this: the number of college entrance examination applicants reached a peak of 13.42 million in 2024, and then began to decline. In 2025, it was 13.35 million, and in 2026, it dropped to 12.90 million, a decrease of 450,000 from the previous year and a decrease of 520,000 from the peak.

Speaking of which, I have two heartfelt words to say. First, let's talk about those high-scoring students admitted to the 985 universities. This group is composed of the most excellent obedient individuals, but they are precisely the target that AI is most likely to replace. What is the essence of the CCP's education system? It is to turn people into retrieval carriers, then throw a question at them, and let the students spit out the answer. Whose job is this? Isn't this the job of large language models? And when it comes to doing this job, no one is better than these large language models. Therefore, this part is precisely not your moat, because the premise of this kind of education leads to a very serious problem: being too accustomed to the existence of correct answers. Let's not talk about anything else, let's just talk about the stock market. Where do Chinese retail investors waste the most time? It's in trying to find a universally applicable standard answer for a winning strategy, and then, this strategy can be applied everywhere. This is the lingering poison of the college entrance examination ideology in people's minds, philosophized into behavior. However, in real-world competition, most of the time there are no standard answers at all, only judgments that are more in line with the facts. So, people educated under this system have a characteristic: they can definitely win gold medals in games with answers. Now, you are suddenly asked to play a game without even a referee. The awkwardness and confusion in the face of the unknown can be clearly revealed. As for those who failed or had mediocre results, I won't say anything to comfort you. Your greatest asset is precisely that you have nothing to lose, nor have you been washed clean by the idea of standard answers. You are probably more accustomed to exploring and trying, and then finding a way after walking the path. This, precisely, is the rarest survival instinct in the AI era. If you look back at history, you will understand that those who have benefited the most from every technological revolution have never been the winners of the current rules, but the outsiders to the rules. Now, you are that outsider. Therefore, I say that AI is actually a true lifelong equalizer. It doesn't look at your academic qualifications; it only cares about whether you can ask good questions and whether you can judge whether what it spits out is reliable. On the matter of AI, the halo of the 985 universities cannot even scratch the surface of AI.

Talking only about China is not enough. Let's broaden our lens and look at the global job market. The World Economic Forum released a future of jobs report last year, interviewing over a thousand employers, covering a fairly comprehensive range, reaching fifty-five economies. By 2030, AI and this wave of technology will create about 170 million new jobs, while eliminating over 92 million jobs. The net increase is 78 million jobs. This net figure is positive, so it sounds less alarming, right? But you must not focus on this net figure; what's truly important is the angular momentum. Nearly 22% of existing jobs will be structurally redefined. In other words, about one in five people will have their work redefined. The report also contains a very important number: 63% of employers say that the skills gap is their biggest roadblock right now. So, you see, this is not an issue for any single country; the whole world is simultaneously changing tracks.

Okay, now suppose you are planning to go abroad or are a parent planning for your child to go abroad. There are a few pitfalls you must be aware of. You need to know now that this part is very, very realistic, and I already have friends who have fallen into this trap. It might not be comfortable to hear, but being uncomfortable is better than regretting it after falling in. The first pitfall is PP10043. Everyone remember this. This is an invisible wall. This thing was created by old Trump in May 2025 (Note: the original text refers to 2020), and this part specifically targets students who want to come to the United States to study. After old Trump signed this, the Biden administration kept it and did not move it. The current administration is still implementing it. What is it? It means that any Chinese national who has studied, worked, or received funding from a military-civil fusion institution will basically be scrutinized and likely rejected when applying for an F or J visa. Its actual scope of impact is much larger than you imagine. Let's look at this: Georgetown University estimates that 3,500 to 5,000 Chinese students are directly blocked by this hurdle each year, accounting for 17% to 27% of Chinese new graduate students in STEM fields in the United States. By May 2025, Rubio directly announced the complete revocation of visas for some Chinese students, and expanded the scope of scrutiny from military affiliation to any affiliation with the CCP, and then banned you from studying in various sensitive fields.

Let's talk about the landmine universities. Most people know that the Seven Sons of National Defense—like Beihang University, Beijing Institute of Technology, Harbin Institute of Technology, and Beijing University of Posts and Telecommunications—are all listed under Section 1286 of the US Department of Defense regulations. Graduating from these and then applying for STEM graduate programs in the US is basically stepping on a landmine voluntarily. But the truly deceitful part lies in the second tier, which many people are completely unaware of. For instance, Northwestern University in the US issued a warning starting last July, and the Department of Defense updated its entity list last June, adding a large batch of universities and institutions. This means that as long as you have done collaborative research at these institutions, even if it's just co-authoring a single paper, it poses an extremely high compliance risk. Another more covert issue is whether your undergraduate advisor's funding involves military assets of the Chinese Communist Party. If you participated in this project, or if your name even appeared in the acknowledgments of the paper, it is very likely to trigger scrutiny from the visa officer. As for majors, there is a category that gets put into administrative processing as soon as you apply. This blacklist is not specified in any official document; it has been reverse-engineered from a large number of visa rejection cases, and applying for these basically guarantees a visa rejection. Examples include microelectronics and semiconductors, quantum computing, quantum communications, as well as military-related and nuclear energy fields. I actually know people studying nuclear fields and from the Seven Sons of National Defense, and they are all trying to figure out how to avoid this detour. If you want to avoid this detour and are determined to go abroad, how on earth can you keep a low profile and survive? First, try to stick to foundational majors for your field of study, such as pure mathematics, theoretical physics, or statistics. These are not on any sensitive lists to begin with, and graduate schools in Europe and the US are particularly short of such people. You can get in first under the identity of a basic discipline, and once you land, you can decide your direction based on the environment. This is much smarter and has a much higher success rate than crashing into the wall right from the start with a quantum computing label. Moreover, these foundational disciplines are like universal wheels, you know? No matter where you want to pivot, you can pivot, and you can make it move, because no matter what you want to do in the future—whether it's basic research, finance, or even AI—the more fundamental something is, the less it can be replaced.

Next, regarding the universities themselves, Chinese universities as a whole are still producing people using the methods of the industrial era, churning out the kind of labor force demanded by the industrial era. This remark isn't me firing blindly at the map; it's a conclusion reached by Stanford economist Nicholas Bloom when evaluating China's education system. The original meaning was that Chinese students rely heavily on rote memorization just to prepare for the gaokao, resulting in a fundamental mismatch with the needs of technological innovation and critical thinking. At the same time, Scott also estimated that about 500 million people in China have severely deficient skills, meaning they cannot systematically participate in the high-skill and high-wage economy. This is a systematic educational falling-behind on the scale of 500 million people. Therefore, even if you get reassigned to a major you completely dislike, or if your school's textbooks are still stuck ten years in the past, do these things matter? As long as you don't accept it yourself, all of this fucking nonsense is just fleeting clouds.

Moreover, there is a very counterintuitive conclusion here: the more practical a major is, the faster it will die. This is also supported by data, still from that McKinsey report. Among the skill sets with the highest exposure to automation, the ones appearing most frequently are accounting, programming, data entry, and financial processing. You see that? These happen to be the core outputs of practical majors. Take code engineers first: the accuracy of AI in writing standard code already reached 70 percent by 2025, and now it can reach 85 percent. By the time this generation of students graduates, AI's code accuracy will effortlessly exceed 90 percent. It can be said that the value of junior programmers is depreciating at a speed visible to the naked eye. As for financial accounting and the like, that goes without saying—this is the most widespread arena for AI, because accounting rules are right there, and scanning directly according to the rules is all it takes. Conversely, mathematics, physics, and philosophy might seem useless, but they will definitely become more and more valuable because they train your underlying operating system, not some specific application. Mathematics trains your ability to recognize structures in a mess of garbled data, while physics enables you to build models with accurate predictions using very few assumptions. As for philosophy, it is even more important; it helps you spot loopholes in arguments so you don't fool yourself. These three things happen to be AI's most fundamental weaknesses right now. AI can execute logic, but it cannot critique its own logical premises. It can generate inferences, but it cannot judge the truth or falsehood of those inferences, nor whether those inferences actually hold under real-world physical constraints.

Please continue watching the World Economic Forum. Last year, it asked global employers a question: what are the core skills that employers value the most right now? Can you guess what ranked first? It is analytical thinking and critical thinking. Sixty-nine percent of employers consider it indispensable. Closely following that are resilience, flexibility, and adaptability, accounting for sixty-seven percent. Then come leadership and social influence, which account for sixty-one percent. Further down is creative thinking at fifty-seven percent. Self-awareness and motivation account for fifty-two percent. Have you noticed? The top-ranking skills happen to be the byproducts of mathematics, physics, and philosophy. Therefore, regardless of which university you attend or what your major is, there are three things you must do yourself during these four years of college, and no one can do them for you.

The first and most important thing is to perform surgery on your own brain, and burn down the default mode of eagerly finding the answers expected by the test creator with a fire in your heart. How specifically do you practice this? Whenever you accept a conclusion, whether taught by a professor from a textbook or spoken by themselves, force yourself to ask a few questions: What are the core assumptions it relies on? Is there any evidence contradicting it? And how was this evidence quietly disposed of? The most crucial step is to treat AI as your debating opponent rather than an answer machine. Have it argue a viewpoint, and then you specifically look for its loopholes. Ask it to provide evidence, and then you personally verify its sources. I dare say this exercise is much more useful than any classroom because it provides you with timely feedback. Once this set of thinking training is completed, even if the Chinese Communist Party (CCP) learns seventy-seven changes in the future, it will still be unable to fool you, or fooling you and your children will become a very difficult task.

The second thing is to understand the underlying logic of AI. Pay attention, this is not teaching you how to write prompts or anything like that; it is teaching you to see through what the essence of this machine's large model is. It is a probability prediction machine. When you give it a piece of input, what does it do? It predicts the next most likely word based on the statistical regularities in the training data. It is not understanding; it is simply making predictions based on statistics. At least at the current stage, large models are like this. This distinction is extremely important. It explains why large models can talk utter bullshit with a straight face. Sometimes, seriously, I used that interface before, and it was unbelievable—it literally talked absolute bullshit the moment it opened its mouth, because its optimization goal is to sound reasonable rather than to be actually correct. You don't need to understand the mathematical formulas inside, but you must build an intuition for yourself. Large models are compressed bodies of human language statistics, not digital mirrors of human wisdom. Understanding this allows you to know where you should trust it, where you should question it, and where you absolutely must not trust it. This kind of judgment is the difference in using AI. Everyone uses AI, but what does making better use of AI mean? It is not about how many prompt templates you have memorized, but about learning to have judgment.

The third point is hyper-specialization in a single point, which means pushing your advantages to the maximum. Oh, looking at this McKinsey report again, the numbers in it are very striking. Look at this: the demand for positions capable of skillfully using AI has surged nearly sevenfold within two years. In 2023, it only covered about one million workers; by 2025, it has covered nearly seven million, and now it has surpassed twelve million. What does this mean? It means understanding AI is rapidly changing from a scarce capability into a basic threshold. And once something shifts from scarce to ordinary, its premium vanishes. Therefore, what is truly valuable has never been the act of using AI itself, but the ability to use AI to achieve excellence in even a very narrow field, reaching an extreme that is difficult for others to replicate. This is the essence of hyper-specialization. It does not ask you to be a generalist, but to delve so deep into a specific niche that others can only look up in awe. What does it mean to reach the extreme? Let me give an example: this requirement doesn't mean how precise you make your quantification, that is not a standard of measurement; instead, what is it? It means you can specialize in high-frequency statistical arbitrage, figuring out the cross-arbitrage space formed between CME Treasury futures, cryptocurrency markets, and perpetual contracts due to the physical micro-delays of the two locations until you know the ins and outs like the back of your hand. People like this are so scarce that top-tier high-frequency quantitative funds are willing to pay four times the price to poach them. The narrower the niche, the wider your moat becomes.

Alright, let's tie everything we've discussed together. No matter what major you were transferred into, which city you go to, or how high or low your academic qualifications are, as long as you hold onto these points firmly, they will be the truly useful things that help you navigate the next twenty years. Let's look at this: the first thing is model decomposition. To put it plainly, it means you can see through any tool you are currently using, including AI. You need to see through its internal logic, its working mechanism, its structural assumptions, and its capability boundaries, rather than just knowing how to click around the interface or use some prompt engineering tricks. Why is this important? Because in every historical technological leap, the winners were not those who used the tool the earliest, but those who saw through the mechanism of the tool the earliest. For example, the winners of the internet in the 90s were not the first people to get online; they were the ones who realized early on that the internet would destroy information asymmetry. The same applies today: the winners are definitely not those who use AI the earliest, but those who understand early on that a large model is a probabilistic statistical machine. Only then will you know when the output looks completely correct but is actually very dangerous.

The second important thing, as McKinsey has made very clear, is that 57% of work time can theoretically be automated by AI. These tasks share a common trait: most of them can be scaled and replicated. So, what you need to do is find the points within the remaining less-than-half that haven't been fully attacked yet, and then use time and depth to carve out an irreplaceable niche for yourself. What is the third thing? Since 57% of work time will theoretically be replaced by AI automation, humans actually love doing two things: first, saving time better; second, wasting time better. This will create opportunities for many service industries, especially those involving emotional and communication services. Perhaps in the future, when you go for a manicure, you won't be looking for the technician with the best skills, but for the one who is the best at chatting and connecting. What kind of jobs can AI not eliminate? Those that require physical presence, hands-on work, and taking responsibility for the consequences afterward. McKinsey's human-centric professions are the category with the lowest automation risk. In the United States, the median annual salary for these roles is around $71,000, and they account for one-third of all jobs in the U.S.

I don't know about others, but I feel that especially when you use AI a lot, you get fucking annoyed by the words it spits out. Later, I realized why: you can't see any emotion in that language. I'd even prefer you to curse at me, but it's just fucking devoid of emotion. Now I mainly use this thing for calculations and data mining, which is fine. When I'm free, I'd rather shoot the shit with my fucking friends than look at this AI interface. The last ability, and also the most important one, is the ability to compile complex systems. The World Economic Forum lists analytical thinking and critical thinking as the top skills most valued by enterprises in 2025. The logic behind this is that in an accelerating world, the most valuable thing is not how much specific knowledge you possess, but how quickly you can apply new information to update your understanding of the world. AI can give you a pile of data, but it can't tell you what that data actually means within a complex causal chain. You have to build that model in your own brain. Then what can AI do? You can let it calculate and verify, and after verification, take it into the real world to confirm or falsify it. AI can't do this part, because it requires practice.

So what I want to say isn't motivational, nor is it some kind of toxic positivity. At the same time, it's not about trying to comfort friends who didn't do well on their exams; it's just the truth. The rules of the game in this world have never been what those Chinese schools told you. What do these garbage Chinese schools love to do? They love to economically bind exam performance to a person's life. Study hard, thousands of troops crossing a single-log bridge, get a stable job after graduation, and then you're safe. During the period of China's industrialization—ah, from the 1990s to the 2010s—this script was basically valid. But it hides a huge implicit premise: that the world must be stable, and the unchanging rules must be absolute. Whatever that diploma is worth today, it will be worth tomorrow. But by 2026, that premise has collapsed. The Chinese Communist Party is fucking slamming on the brakes. Everyone needs to realize that the valuation of a diploma is priced by the job market. Once the job market collapses, diplomas will naturally depreciate. But the genuine skills you learn—while lacking a standardized test—are the foundation for everyone to make a living. AI reshapes the underlying logic of an industry by an order of magnitude roughly every 18 months; you could say that everyone is currently on the same starting line.

At the same time, each and every one of us is among the first batch of people starting out under these new rules. This is bad, but it's also good. The bad part is that none of us has a reference script in our hands, and the experience of the older generation is basically useless because their experience will become obsolete at the speed of light in the face of the AI variable. The good part, however, is that this is a period of rule reset, and every reset period happens to be the fairest time in history. Why? Because the winners of the old rules are holding a bunch of chips that might be completely useless in the new game, while those who stumbled under the old rules carry no baggage from the old game. Therefore, what truly determines your destiny is not how many points you scored on this gaokao, but whether you can see the true underlying logic of this world in the next few years, and whether you have the guts to live according to that logic. To repeat: when all is said and done, the gaokao is just an exam. Whether you do well or poorly, your value in life does not depend on this exam. Also, what else? Whether you did well or not, take the CCP's history textbooks and those fucking political textbooks and burn them in a fire. Keeping that bullshit at home makes it too easy to get brainwashed and is too disgusting; it is simply the source of intellectual mutilation. Even though almost nobody believes that bullshit, burning it is equivalent to putting an exclamation mark on your journey.