There is something strange about investing. If an item in a supermarket drops from 100 yuan to 70, most people see it as cheaper and become more willing to buy, as long as the product itself has not changed. Financial markets often work in exactly the opposite way. An asset rises from 100 to 150, and people begin to think it is worth buying. At 200, more people start talking about it. At 300, even those who were never interested begin doing research. Only when everyone seems to be making money do many finally convince themselves: “This thing really is good.” The reverse is also true. They like it at 100, begin to doubt it at 90, revisit all the risks at 70, suddenly find problems everywhere at 50, and finally sell at the moment of greatest pessimism.

If the asset itself has not changed, this behavior is genuinely strange. Once the price rises, the asset is clearly more expensive, yet people like it more and more. Once the price falls, it is clearly cheaper, yet people increasingly dislike it. Investing may be one of the few markets where a higher price makes consumers want to buy more. That is why I have come to think that the hardest part of investing may never have been analyzing the asset. It is analyzing yourself.

Before they begin investing, many people believe they are highly rational. They study financial statements, industries, and the macroeconomy. They calculate valuations, set an entry price, and even tell themselves in all seriousness, “If the market falls, I definitely will not panic.” All of that is easy to say while the market is calm. It feels very different when the fall actually comes. When the 100,000 yuan in an account becomes 80,000, something in the brain begins to shift. What once looked like “short-term volatility” suddenly becomes “risk.” Negative news that used to be ignored becomes impossible to miss. Problems that once seemed trivial suddenly look severe. The investor starts searching constantly for bearish arguments and, one day, finally gives in and sells.

After selling, people often feel immediate relief. That relief matters because it suggests that many sell decisions are not really solving investment risk. They are solving psychological pain. Human beings are acutely sensitive to losses. The happiness of gaining 10,000 yuan is usually nowhere near enough to offset the pain of losing 10,000. If an investment rises from 100,000 to 120,000 and then falls back to 100,000, the investor has not actually lost any money. Psychologically, though, it rarely feels as if nothing happened. It feels more like a loss of 20,000, because at some point the 120,000 was quietly treated as money that already belonged to them.

This creates one of the most common illusions in investing: an unrealized gain easily becomes private property in the mind before it has ever been realized. When the price falls from 120,000 to 110,000, the brain does not say, “My investment is still up 10,000.” It says, “I just lost 10,000.” If the price keeps falling, the pain keeps accumulating. Eventually, decisions stop being organized around future returns and start being organized around a different question: “What can I do to feel better?”

The same thing happens on the way up. People are often cautious when an asset first begins to rise, but the more it gains, the less cautious they become. Someone who once thought a 10 percent annual return was excellent starts to see 20 percent as normal after a run of gains. Someone who planned to sell after making 30 percent reaches that target and begins thinking about 50 percent. At 50 percent, they start asking why it cannot double. The target keeps moving further away. That is the most troublesome thing about greed.

Greed does not suddenly arrive and honestly announce, “I am becoming greedy now.” It usually disguises itself as a perfectly reasonable explanation: “The fundamentals have improved.” “The addressable market is larger than I thought.” “This time is different.” “The trend is so strong that I may never get back in if I sell.” Any one of those statements may be reasonable on its own. The real danger is that it is hard to tell whether you are reassessing the facts or simply becoming more optimistic because the price has risen.

Market prices have an extraordinary ability not only to change your wealth, but also to change the way you interpret reality. When an asset is rising, good news feels especially convincing. A new product can be read as the beginning of enormous growth. A new industry policy can be read as confirmation of a long-term trend. Institutional buying can be read as approval from smart money. A competitor leaving the market can be read as an improvement in the competitive landscape. Once the same asset begins to fall, every one of those events can be reinterpreted. The product may fail. The policy may be uncertain. The institutions may only be trading in the short term. Competition may be getting worse.

Often, the facts have not changed at all. Only the price has changed. The price changes the mood, the mood filters the information, and the selected information then proves that the mood was right. It forms a loop that is difficult to notice. People usually believe they see evidence first and form an opinion afterward. In reality, the order is often reversed: the opinion comes first, then the search for evidence. In a rising market, a holder actively looks for every reason the rise should continue. In a falling market, the same person begins looking for every reason to sell. The internet makes the problem worse, because almost anything you want to prove can be supported by more than enough material.

If you think the market will rise, you can find dozens of bullish reports. If you think it is about to collapse, you can find dozens of carefully reasoned bearish analyses. There is now too much information for information alone to become an answer. What people see in the market is often not reality itself, but only the part of reality they are willing to see.

Cost basis is another fascinating thing. People become almost emotionally attached to their purchase price. Buy an asset at 100 and watch it fall to 80, and one of the most common thoughts is, “I will sell once it gets back to 100.” That sounds perfectly natural, but on closer examination it has very little logic behind it. The market does not know you bought at 100. The company’s management does not know. Other traders do not know. Future cash flows will not change because of your cost basis.

If the asset is now worth only 50, a rise from 80 back to 100 does not become more reasonable just because you once paid 100 for it. If it is actually worth 200, selling at 100 merely because you have “broken even” makes just as little sense. A cost basis is only a historical record, but in the mind it quickly becomes a psychological anchor. Profit and loss are redefined around that anchor. There is nothing inherently red or green about an asset priced at 80, but the moment a trading app places “-20%” beside it, the experience changes completely.

Much of the time, we are not judging what an asset is worth. We are judging how far it is from “my price.”

There is another powerful force in a bull market: watching other people make money. Losing money yourself is painful, of course, but sometimes it is also painful to watch others profit while you do not. You may have had no intention of investing in something. Then one friend makes 30 percent and you think nothing of it. Another makes 50 percent. Social media starts discussing it, the news begins covering it, and more and more people post screenshots of their returns. One day you start wondering whether you have missed something. The higher the price climbs, the stronger that feeling usually becomes. By the end, the real reason for buying is no longer expected return. It is the fear of being left behind for good.

The memory-chip bull market in the first half of this year gave me a strong sense of FOMO. Whether I opened Douyin or Twitter, it only took a few posts or videos to find people showing off their gains. Shares of companies such as SanDisk, SK hynix, and Micron had risen severalfold. I felt that I did not understand the sector, though, so I did not buy.

This is also why many people show no interest in an asset while nobody wants it, then suddenly develop an intense desire to research it after it has risen severalfold. A rising price creates credibility of its own. Human beings rely heavily on group judgment. In many parts of life, that mechanism works well. A long line outside a restaurant usually means the food is not too bad. A product with 100,000 positive reviews is usually more reliable than one with no reviews at all. Financial markets have a complication: when everyone buys because “other people are buying,” their purchases push the price higher. The rise then validates the rise, more people join in, and eventually the price itself becomes the strongest argument.

The same mechanism works in a decline. Everyone is selling, so the price falls. The falling price then appears to prove that the risk is high, so more people sell. A crowd can create both mania and panic. People standing inside that crowd rarely feel that they are following it. They feel that they have reached an independent conclusion.

Another widespread tendency is to assume that whatever happened recently will keep happening. After a market has risen for several years, people begin to treat rising prices as the normal condition. After it has fallen for a year, they begin to feel that the bad days will never end. History has repeated this pattern countless times. When prices are at their best, the news usually looks its best as well: corporate profits are growing, the economy is thriving, investor confidence is strong, and everyone can find a reason to stay optimistic. When prices are at their worst, conditions are often genuinely terrible: the economy is in recession, companies are laying people off, bad news keeps arriving, and everyone can find just as many reasons to remain pessimistic.

This makes investing deeply counterintuitive. A truly cheap asset rarely feels “safe.” If every problem had been solved and everyone felt confident, the price would probably no longer be cheap. The most expensive moments, by contrast, often feel very comfortable. Your account rises every day, the news is good, the people around you are making money, and the future looks bright. Risk rarely appears wearing a shirt with the word “RISK” printed across it. More often, it looks like certainty.

Investors have an even more troublesome weakness: after making money, they tend to overestimate themselves. If someone enters the market for the first time and makes three good purchases in a row, it is difficult for them to conclude, “Maybe I was just lucky.” A much more common thought is, “I may actually be good at this.” So the original 10,000 yuan becomes 50,000. Relatively simple assets give way to more complicated ones. Someone who once avoided leverage starts to think that using a little probably cannot hurt.

Profit brings more than wealth. It also raises confidence, and confidence encourages greater risk-taking. That is why one of the most dangerous beginnings in investing is sometimes not losing money, but making a great deal of it. A loss at least forces someone to question their method. A string of gains can quickly persuade a person who has never lived through a full market cycle that they have discovered a reliable pattern. By the time serious volatility arrives, the risk they are carrying may be far greater than it was at the start.

This is why talking only about “mental toughness” is not enough. An investor’s psychology is largely shaped by position size. A 30 percent fall in a 10,000-yuan investment and a 30 percent fall in someone’s entire life savings are completely different psychological experiments. The first may make the asset look cheaper. The second may leave the person refreshing prices at three in the morning.

In theory, the same person should reach the same conclusion about the same asset. In practice, a change in position size can turn them into a different person. Every small decline in the market may push a real-life goal a little further away: the money for a home, a child’s education, years of savings, or retirement. Once those goals become tied to the number on a screen, even a calm person will struggle to treat the problem as pure mathematics.

Many so-called problems of investment psychology are therefore problems of risk management at heart. If an ordinary market fluctuation is enough to make someone abandon their original investment logic, one possibility is not that their willpower is weak. Their position may simply have been too large from the beginning. The idea is easy to understand. A person can balance comfortably on a board one centimeter above the ground. Put the same board between two buildings one hundred meters in the air and their body will immediately stiffen. Their ability to balance has not suddenly disappeared. What changed was the cost of making a mistake.

Investing is the same. Many people are exceptionally rational in a paper-trading exercise and become someone else as soon as real money is involved. There is nothing strange about that. Real money creates real emotion.

For the same reason, I have always thought that “stay rational” is almost useless as investment advice. Nobody becomes irrational because they did not know they were supposed to be rational. Fear during a crash needs no teacher. Greed during a long rise requires no training. These responses existed long before modern financial markets. Fear helped our ancestors avoid danger. Following the group improved the odds of survival. Loss aversion helped people protect resources they already had. In the wild, adjusting quickly to recent information may have worked better than slowly building a probability model.

The problem is that we bring a brain designed for survival into a financial market built from numbers, probabilities, and long-term compounding. Many instincts that once helped human beings stay alive can become weaknesses here.

The most effective approach to investing may not be training yourself to become a person without emotions, because such people barely exist. A more realistic approach is to accept that emotions will come, then design rules in advance so those emotions have less power. Before buying, you should answer a few questions. Why am I buying? What exactly do I see in this asset? What would prove my original judgment wrong? What is the largest loss I am willing to accept? If this position loses half its value, will it affect my normal life? If the price rises sharply, has the underlying value really changed, or has only the price changed? If the price falls sharply while the original logic remains intact, what exactly should I do?

These questions are usually easy to answer before buying, when money has not yet begun to influence emotion. Answering them after you hold the asset is much harder. People begin revising their answers, and they usually refuse to admit that they are doing it.

This may be the most important purpose of an investment plan. It is not a prediction of the future, because nobody knows the future. It is more like a contract written by your calm past self for your emotional future self. After the market rises, that future self may believe every asset can keep going up. After the market falls, the same future self may believe the world is about to end. If the emotion of the moment is allowed to rewrite the rules every time, then no investment strategy really exists. There are only immediate reactions.

Of course, this does not mean someone should cling forever to an original view. Refusing to admit a mistake is not discipline. If new facts prove that the original judgment was wrong, changing your mind is the right thing to do. The difficult part is separating two things: did the facts change, or did only the price change? Did the investment thesis fail, or did losing money simply become painful? Is the asset genuinely becoming more valuable, or do you want to believe it is because you have made money?

This may be one of the hardest questions in investing, because we are both judge and defendant. We have to decide whether our judgment has been distorted by emotion, and the decision is still being made by the same brain.

That is why I do not believe that learning a few terms such as “loss aversion,” “confirmation bias,” and “anchoring effect” is enough to overcome them. Knowing that a bias exists does not make it disappear. Many people know that staying up late is bad and still do it. Knowing that a high-sugar diet is unhealthy does not stop someone from wanting dessert when it appears. Human beings naturally seek benefit and avoid harm, and we easily become absorbed in whatever makes us feel comfortable. Investing is no different.

Theory is easiest to understand after the market has closed. The real exam begins when the account is shrinking every day, or when everyone around you is saying, “This time really is different.”

In the end, investing may not be a contest between a person and the market. It is more like a contest between a person and themselves. The market merely keeps producing prices. Fear, greed, regret, hope, envy, and confidence are produced by the person. There is a version of you who becomes more greedy as prices rise, another who becomes more afraid as they fall, and still another who begins to believe they are smarter than everyone else after a run of gains. All of them live in the same body, yet each follows a completely different investment logic.

The person trading against you is never just someone else in the market. It is also your future self.

What a good investment system may really need to do is not predict the next rise or fall with perfect accuracy, but limit the power of that completely different self before they appear.

Because the market’s greatest talent has never been making people lose money directly. It changes their emotions first, then lets them make the losing decision with their own hands.

投资有一个很奇怪的地方：如果一家超市里的商品从一百块降到七十块，大部分人会觉得它变便宜了，只要商品本身没有发生变化，人们通常会更愿意购买；但金融市场经常完全相反，一项资产从一百块涨到一百五十块，人们开始觉得它值得买了，涨到两百块，讨论它的人越来越多，涨到三百块，原本不感兴趣的人也开始研究，等到所有人都在赚钱的时候，很多人才终于说服自己：“这个东西确实不错。”反过来也一样，一百块的时候觉得它很好，跌到九十块开始有一点怀疑，跌到七十块开始重新研究风险，跌到五十块，突然发现这个东西浑身都是问题，最后在最悲观的时候卖掉。

如果资产本身没有发生变化，这种行为其实非常奇怪，因为价格上涨以后，东西明明变贵了，人反而越来越喜欢；价格下跌以后，东西明明变便宜了，人却越来越讨厌。投资可能是少数几个会让消费者因为涨价而增加购买欲望的市场，而这也是为什么我越来越觉得，投资真正困难的地方可能从来不是分析资产，而是分析自己。

很多人在开始投资以前，都觉得自己很理性，他们会研究财报，研究行业，研究宏观经济，计算估值，制定买入价格，甚至认真告诉自己：“如果市场下跌，我绝对不会恐慌。”这些话在市场平静的时候都很容易说，真正跌下来以后却是另外一回事，因为当账户里的十万元变成八万元时，人脑中的很多东西都会开始变化，原本被认为只是“短期波动”的东西，突然开始变成“风险”，以前忽略的负面新闻变得异常刺眼，之前认为无关紧要的问题也会一下子显得非常严重，于是人开始不停搜索悲观观点，最后在某一天终于忍不住卖掉。

卖掉以后，人通常会立刻松一口气，而这种松一口气的感觉其实非常重要，因为它说明很多卖出行为真正解决的问题可能并不是投资风险，而是心理痛苦。人对亏损极其敏感，赚一万元带来的快乐，通常很难抵消亏一万元带来的痛苦；一项投资从十万元涨到十二万元，再跌回十万元，最终资产并没有减少，但人的心理体验往往并不是“什么也没发生”，他更容易觉得自己亏了两万元，因为十二万元已经在某一个瞬间被他默认为属于自己的钱。

这也是投资里一个非常常见的错觉：没有卖出的浮盈，在心理上很容易提前变成私人财产，于是价格从十二万跌到十一万的时候，大脑并不会理解成“我的投资还赚了一万元”，而是理解成“我刚刚损失了一万元”，如果价格继续下跌，这种痛苦就会不断累积，最后很多投资行为也就不再围绕未来收益展开，而是开始围绕“怎样才能让自己舒服一点”展开。

上涨的时候同样如此。一项资产刚刚开始上涨时，人往往还很谨慎，可是涨得越多，谨慎反而越少，原本觉得年化百分之十已经很好的人，在连续上涨以后会开始觉得百分之二十也很正常，原本计划赚百分之三十就卖的人，真正赚到百分之三十以后，却会开始想百分之五十，到了百分之五十，又会觉得为什么不能翻倍，目标不断向后移动，这就是贪婪最麻烦的地方。

贪婪并不会突然出现，然后非常诚实地告诉你：“我现在开始贪婪了。”它通常会伪装成一套非常合理的解释，比如“基本面变好了”“市场空间比以前想象得更大”“这次不一样”“趋势这么强，卖了以后可能再也买不回来”，这些话单独拿出来都可能有道理，真正危险的地方在于，一个人很难判断自己究竟是在重新评估事实，还是仅仅因为价格上涨以后变得更加乐观。

市场价格有一种非常强的能力，它不仅会改变你的财富，还会改变你对现实的解释。资产上涨的时候，利好消息会显得特别可信，公司发布一个新产品，可以被解释成巨大的增长空间，行业出现一个新政策，可以被解释成长期趋势确认，机构开始买入，可以被解释成聪明资金认可，竞争对手退出，则可以被解释成市场格局改善；但同一项资产一旦开始下跌，这些事情又会被重新解释，新产品可能失败，政策存在不确定性，机构可能只是短期交易，竞争可能越来越激烈。

很多时候，事实根本没有发生变化，变化的只是价格，而价格改变情绪，情绪重新筛选信息，最后这些被筛选出来的信息又反过来证明自己的情绪是正确的，于是形成一个很难察觉的循环。人通常以为自己是先看见证据，然后得出观点，现实中却经常刚好相反：先有观点，再去寻找证据。上涨的时候，一个持有者会主动寻找所有支持继续上涨的信息；下跌的时候，同一个人又会开始寻找所有证明自己应该卖出的理由，而互联网让这个问题变得更加严重，因为无论你想证明什么，几乎都能找到足够多的材料。

你觉得市场要涨，可以找到几十篇看多报告；你觉得市场要崩，也能找到几十个逻辑严密的悲观分析。信息已经多到无法自动成为答案，一个人在市场里真正看到的，经常不是现实本身，而只是自己愿意看到的那一部分现实。

还有一个非常有意思的东西是成本价。人对自己的买入价格有一种近乎执着的感情，一百块买入一项资产，跌到八十块以后，最常出现的想法之一就是：“等它涨回一百，我就卖。”这句话听起来非常自然，但仔细想一下，其实并没有太强的逻辑，因为市场不知道你一百块买的，公司的管理层不知道，其他交易者不知道，未来的现金流也不会因为你的成本价发生任何变化。

如果这项资产现在只值五十块，那么它从八十块涨回一百块，并不会因为你曾经在一百块买入就变得更加合理；反过来，如果它实际上值两百块，那么因为刚刚“回本”就在一百块卖掉，同样没有什么道理。成本价本来只是一个历史记录，但在人脑里，它会迅速变成一个心理锚点，盈亏也会围绕这个锚点被重新定义，于是八十块的资产本身并没有红色或者绿色，交易软件只要给它加上一个“-20%”，人的感受马上就完全不同了。

很多时候，我们不是在判断资产值多少钱，而是在判断它距离“我的价格”还有多远。

牛市里还有另外一个非常强大的力量，那就是别人赚钱。自己亏钱当然痛苦，但别人赚钱而自己没有赚到，有时候也很痛苦。你原本根本不打算投资某个东西，后来一个朋友赚了百分之三十，你觉得没什么；又一个朋友赚了百分之五十，社交媒体开始讨论，新闻开始报道，越来越多的人晒收益，某一天你就会开始怀疑自己是不是错过了什么，而价格涨得越高，这种感觉通常越强，最后真正推动买入的已经不再是预期回报，而是害怕自己被永远留在车外。

今年上半年的存储牛市就让我觉得非常FOMO。那时候抖音上也好推特上也好，随便刷几个帖子和视频都能看到一堆人晒单。像闪迪、海力士、美光那些股票有人甚至都翻了好几倍了。但我那时候觉得自己并不了解这个东西，所以就没有买。

这也是为什么很多人在一个资产无人问津的时候没有兴趣，等它涨了几倍以后才突然产生巨大的研究热情，因为上涨本身会制造可信度。人类是一种非常依赖群体判断的动物，在很多生活场景里，这种机制其实非常有效，比如一家餐厅门口排着长队，通常说明它不会太差，一个商品有十万条好评，通常也比一个没有评价的商品更可靠；但金融市场有一个麻烦，当所有人因为“别人也在买”而买入时，他们自己的买入又会继续推高价格，于是上涨证明上涨，越来越多人加入，最后价格本身成了最大的理由。

这种机制在下跌的时候同样成立，所有人都在卖，所以价格下跌；价格下跌又证明风险很大，于是更多人卖。人群既可以制造狂热，也可以制造恐慌，而站在人群里面的人通常不会觉得自己正在从众，他会觉得自己只是做出了独立判断。

还有一个非常普遍的问题是，人特别喜欢相信最近发生的事情会继续发生。一个市场连续涨了几年以后，人们很容易觉得上涨是正常状态；连续跌了一年以后，又会觉得坏日子永远不会结束。这种事情在历史上重复过无数次，因为价格最好的时候，新闻往往也最好看，企业利润增长，经济繁荣，投资者信心充足，每个人都能找到继续乐观的理由；价格最差的时候，情况往往真的很糟，经济衰退，公司裁员，坏消息不断，所有人同样都能找到继续悲观的理由。

这使投资变得非常反直觉。真正便宜的资产通常不会给人一种“很安全”的感觉，因为如果所有问题都已经解决，所有人都充满信心，价格通常也不会继续便宜；反过来，真正昂贵的时候往往非常舒服，账户每天上涨，新闻都是好消息，身边的人都在赚钱，未来看起来一片光明，而风险很少会穿着写着“风险”两个字的衣服出现，很多时候，它看起来更像确定性。

投资者还有一个更麻烦的弱点：赚钱以后，很容易高估自己。如果一个人第一次进入市场就连续买对三次，他很难得出“我可能只是运气不错”这个结论，更常见的想法是“我好像真的很擅长这个”，于是原来投入一万元，后来投入五万元；原来只买相对简单的资产，后来开始研究更复杂的东西；原来不用杠杆，后来觉得适当使用一点似乎也没关系。

赚钱带来的并不只有财富，它还会增加自信，而自信又会进一步增加风险承担，所以投资里有时候最危险的事情不是一开始亏钱，反而是一开始赚了很多钱。亏损至少会逼一个人怀疑自己的方法，连续盈利则可能让一个根本没有经历完整市场周期的人迅速相信自己掌握了某种规律，等真正的大幅波动出现时，他承担的风险可能已经远远大于最开始的时候。

这也是为什么只讨论“心理素质”其实不够，因为一个人的投资心理很大程度上是由仓位决定的。一万元的资产跌百分之三十，和一个人全部积蓄跌百分之三十，是完全不同的心理实验，前者可能让人觉得价格更便宜了，后者却可能让人凌晨三点还在刷新行情。

理论上，同一个人应该对同一个资产有相同的判断，但实际上，当仓位发生变化以后，他甚至会变成另外一个人，因为市场每跌一点，生活中的某个目标可能就离自己更远一点，买房的钱、孩子的教育费用、几年的积蓄、养老资金，一旦这些现实目标和账户里的数字连接起来，再冷静的人也很难完全把它当成一道数学题。

所以很多所谓的投资心理问题，本质上其实是风险管理问题。如果一个正常的市场波动就足以让一个人彻底改变原来的投资逻辑，那么有一种可能并不是他意志力太差，只是仓位从一开始就太大了。这件事情其实很好理解，如果一个人站在一厘米高的木板上，他可以非常轻松地保持平衡，把同一块木板放到一百米高的两栋楼之间，身体会马上僵硬，他的平衡能力并没有突然消失，变化的是犯错的代价。

投资也是一样。很多人在纸面模拟的时候异常理性，真钱进去以后马上变成另外一个人，这并不奇怪，因为真正的钱会产生真正的情绪。

也正因为如此，我一直觉得“保持理性”是一条几乎没有意义的投资建议，没有人是因为不知道应该理性才变得不理性。一个人在暴跌的时候恐惧，不需要任何人教；在连续上涨的时候贪婪，也不需要学习。这些反应在人类形成现代金融市场以前就已经存在，恐惧帮助祖先躲避危险，跟随群体能够提高生存概率，损失厌恶能够让人珍惜已经拥有的资源，快速根据最近的信息调整判断，在野外可能比慢慢建立概率模型更加有效。

问题只是，我们拿着一套为生存设计的大脑，走进了一个由数字、概率和长期复利组成的金融市场，很多曾经帮助人类活下来的本能，在这里恰好可能变成弱点。

所以投资里真正有效的方法，可能并不是训练自己成为一个没有情绪的人，因为这种人基本不存在；更现实的方法，是承认自己一定会有情绪，然后提前设计规则，让情绪没有那么大的权力。在买入之前，就应该回答一些问题：为什么买，到底看中了什么，什么情况出现以后原来的判断会被证明是错的，最多愿意承担多大的损失，这个仓位如果跌掉一半会不会影响正常生活，如果价格上涨很多，是因为价值真的发生了变化，还是单纯价格上涨，如果价格下跌很多，而最初的逻辑没有变化，自己究竟应该做什么。

这些问题在买入以前通常非常容易回答，因为那时钱还没有真正影响情绪；真正持有以后再回答，就困难得多，因为人会开始修改答案，而且通常不会承认自己正在修改。

投资计划最重要的作用可能就在这里。它并不是预测未来，因为没人知道未来，它更像是一份过去那个冷静的自己，写给未来那个情绪化的自己的合同。市场上涨以后，未来的自己可能会觉得所有资产都还能继续涨；市场下跌以后，未来的自己又会觉得世界马上要结束；如果每一次都让当时的情绪重新制定规则，那么所谓投资策略实际上根本不存在，只剩下即时反应。

当然，这不意味着一个人应该永远坚持最初的观点，死不认错不是纪律，如果新的事实证明原来的判断错误，改变观点当然是正确的。真正困难的是分清两件事情：到底是事实变了，还是价格变了；到底是投资逻辑失效了，还是自己因为亏钱开始难受了；到底是资产真的越来越有价值，还是因为自己赚钱以后想相信它越来越有价值。

这可能是投资里最难回答的问题之一，因为人既是法官，也是被告，我们需要判断自己的判断是不是受到情绪影响，而做出这个判断的依然是同一个大脑。

所以我不太相信一个人仅仅学会几个“损失厌恶”“确认偏误”“锚定效应”这样的名词，就真的能够克服它们。知道自己存在偏见，不等于偏见就会消失，很多人都知道熬夜不好，还是会熬夜，知道高糖饮食不好，也不代表看到甜食的时候就不会想吃，因为人天然都是趋利避害的，容易沉迷在让自己舒服的事情中，投资同样如此。

理论往往是在市场关闭以后最容易理解的东西，真正的考试发生在账户每天缩水的时候，或者发生在所有人都告诉你“这次真的不一样”的时候。

投资最终可能不是一个人与市场之间的比赛，它更像是一个人与自己的比赛。市场只是不断给出价格，真正产生恐惧、贪婪、后悔、希望、嫉妒和自信的是人自己；上涨的时候，有一个越来越贪婪的自己，下跌的时候，有一个越来越恐惧的自己，连续赚钱以后，还有一个开始觉得自己比别人聪明的自己，这些人住在同一个身体里面，却拥有完全不同的投资逻辑。

真正和你交易的，从来不只是市场上的其他人，还有未来的你自己。

一个优秀的投资体系真正要做的，也许不是准确预测下一次上涨和下跌，而是在那个完全不同的自己出现以前，提前限制他的权力。

因为市场最擅长的一件事情，从来不是让人亏钱，而是先改变人的情绪，再让人亲手做出那个让自己亏钱的决定。

While looking for work recently, one fact has become harder and harder to ignore: people who know how to do useful work are no longer scarce. Open a job app and pick any half-decent position, and hundreds of applications are no longer unusual. A new graduate can use Office, edit videos, make PowerPoint slides, write a little Python and Java, pass China’s CET-4 or CET-6 English exam, and use ChatGPT to help write code. Ten or fifteen years ago, that person might have counted as a capable university graduate. Today, they are simply an entirely ordinary one.

People have not become less capable. Quite the opposite: everyone can do too many of the same things.

That sounds counterintuitive. Schools teach us that a person should keep learning skills and that the more skills they master, the more competitive they become. The market clearly does not work that way. It does not pay a higher price because something was difficult to learn, or because you stayed up for many nights learning it. What the market really wants to know is very simple: why does it have to choose you?

Suppose a job receives 1,000 applications and 700 applicants know Java. “Knowing Java” is certainly a skill, but it is no longer much of a competitive advantage. If another 300 know both Java and Python, Python will not magically make them valuable either. Their real competitors are the other 299 people who look much the same. When people cannot find a job, their first response is often to keep stacking more skills onto themselves. If Java is not enough, they learn Python. If Python is not enough, they learn Go, then Docker, Kubernetes, Redis, and microservices. Their resumes keep getting longer, but the market does not pay noticeably more for them.

This is a lot like investing. The price of something is never determined by how much effort went into producing it. It depends on how many people want it and how many people can supply it. Diamonds are not expensive simply because miners work hard to extract them. A roadside stone does not attract buyers just because it is heavy to carry. Skills are goods in essentially the same sense. Once you understand that, many seemingly strange outcomes stop looking so strange.

Photography used to be a professional skill. A photographer had to understand exposure, shutter speed, ISO, lenses, and composition, then buy a pile of expensive equipment. Today, someone with no knowledge of photography can take a phone photo that an ordinary enthusiast twenty years ago would have struggled to produce. Video editing has followed the same path. Making a decent video once meant learning Premiere or Final Cut and studying timelines, transitions, and color grading. Then Jianying, known internationally as CapCut, arrived. Now AI can generate captions automatically, remove pauses, match music, and even create a video directly from text.

Programming is going through the same thing.

I studied computer science as an undergraduate, so the change feels especially obvious to me. A few years ago, being able to build a complete website, from the database and backend to the frontend, still counted as a clear technical ability. Today, having AI build a website is hardly something worth showing off. You do not even need to understand React or Vue properly. If you have a rough idea of what you want and spend a few hours talking it through with AI, something that runs will probably emerge.

That does not mean programmers will disappear, nor does it mean programming has no value. What is happening is that the words “can write code” are themselves losing value, and that is a very different claim. The Industrial Revolution did not eliminate manufacturing; it reduced the premium on a great deal of purely manual labor. Excel did not eliminate finance, but it stripped many forms of manual calculation of the premium they once commanded. AI will not eliminate every kind of knowledge work either. The first thing it is likely to compress is the premium on mental work that has already been highly standardized and can be clearly described and repeatedly executed.

Once a skill can be clearly described, systematically taught, and assessed with measurable criteria, it becomes easy to produce at scale. Universities happen to be especially good at this: write a textbook, define a syllabus, design an exam, then produce hundreds of thousands of people with similar knowledge every year. When I now see headlines such as “This major is extremely popular” or “This skill will face a shortage of millions of workers,” my first reaction is no longer to rush out and learn it. I ask a different question: if everyone knows there is a shortage, will there still be one five years from now?

A publicly known opportunity rarely lasts very long. When internet companies paid high salaries, huge numbers of people began studying computer science. When independent content creation became profitable, huge numbers began making short videos. Now that artificial intelligence is booming, almost everyone wants to learn AI. This is completely normal. High prices attract supply. The more profitable something becomes, the more people enter the field, and the more quickly its original scarcity can disappear.

That is why I think the things truly worth learning are not “difficult skills,” but “scarce skills.” The two ideas are often confused. Advanced mathematics is difficult, but mastering it does not guarantee that someone will pay you a high salary. There is also a great deal of obscure, highly complex knowledge that only a few hundred people may understand. If nobody needs it, it still has little economic value.

Scarcity has to exist alongside demand.

If ten thousand people urgently need a problem solved and only one hundred know how to solve it, that is valuable scarcity. When I judge whether a skill is worth learning, I now prefer to look at three things: how many people need it, how many people can provide it, and how large an outcome it can affect once mastered. Following that logic, the scarcest things in the future may not be any particular software package or programming language at all.

The first thing becoming scarcer is the ability to find the problem.

AI is very good at answering questions, provided you know what to ask. Imagine a company announcing, “We need to use AI to cut costs and improve efficiency.” It sounds correct, but in practice it says almost nothing. An ordinary person may immediately begin researching large language models, agents, and RAG, then produce a fifty-page deck explaining how powerful artificial intelligence is. A genuinely valuable person may first examine the company’s workflow. How much does customer service actually cost? Which step wastes the most labor? Where is the error rate highest? Does the problem need AI, or could a fifty-line script solve it? Is it even possible that efficiency is not the real problem and that nobody wants the product in the first place?

There are many people who can answer questions and far fewer who can discover the real question. As AI makes “answers” cheaper, “questions” naturally become more expensive.

The second thing is judgment.

The internet has no shortage of information. If you are considering a stock, AI can generate a report thousands of words long in a minute. If you are looking for work, it can analyze ten companies. If you want to start a business, it can list one hundred business models at once. Then what? Is the asset actually worth buying? Should you take that job? Will anyone really pay for that business? AI can give you twenty reasons in favor and immediately produce twenty reasons against, but someone still has to press the final button.

The cheaper information becomes, the more important judgment becomes. Information itself used to be a barrier, and some people made money simply because they knew something others did not. That advantage is disappearing quickly. What separates people in the future may no longer be whether they can find information, but whether they can reach a better conclusion when everyone has roughly the same information.

The irritating thing about judgment is that there is no “Judgment 101” course you can finish and receive a certificate for. It comes mostly from years of experience, failure, and feedback, and from whether someone is willing to keep admitting that an earlier judgment was wrong. Many abilities improve quickly through repetition. Judgment often develops only after the real world has thrown your mistakes back in your face again and again.

The third thing barely counts as a skill in the traditional sense: taking responsibility.

I used to find it strange that society paid some managers so much. Certain managers appeared unable to do anything concrete. They did not write the code, make the slides, or compile the data, so why did they earn more than the people below them? Over time, I began to see that the valuable part is often not “doing,” but “deciding.”

Programmers can propose three technical approaches, but the CTO has to choose one. An analyst can write dozens of pages, but the fund manager has to decide whether to buy. Ten people can sit in a meeting room discussing a project, but eventually one person has to stand up and say, “This is what we are doing. If it goes wrong, I am responsible.” Execution, analysis, judgment, decision, ownership of the result: the further you move along that chain, the fewer people you usually find. Most people like authority, but very few genuinely like responsibility.

When a decision works, everyone is happy. When it fails, someone may be criticized, fired, or lose money. In serious cases, they may even face legal liability. Responsibility is therefore a scarce resource in its own right. AI can generate plans for you and list dozens of possibilities, but it will not go to prison in your place or absorb a ten-million-yuan loss for the company. In the end, a real person still has to put their name on the decision.

Another form of scarcity that I think is often underestimated lies in combinations of skills.

Each skill may be common on its own. Many people speak English. Many can write code. Many understand sales, and people who know something about new-energy industries are not rare either. But someone who understands software, can communicate with overseas clients in English, knows something about new-energy equipment, and can solve the customer’s problem on site is much harder to find. The advantage created by that combination may matter far more than pushing Java from “proficient” to “expert.”

Many people compete with others along a single axis. Programmers compete over who has solved more algorithm problems. Designers compete over who has mastered their software more thoroughly. Video editors compete over who can create more complicated effects. As long as everyone competes on the same axis, the contest will eventually turn into a price war. The current IT job market makes that obvious. Unless a student comes from a particularly competitive university or has an exceptional background, even finding a decent internship can be difficult. Large numbers of applicants present highly repetitive projects and submit nearly identical resumes. Human resources staff may already be numb from seeing them every day.

Sometimes the simplest answer is not to keep fighting harder on the same axis, but to choose another one. An ordinary programmer with sales ability might become an excellent SaaS salesperson or solutions consultant. An ordinary engineer with English and overseas experience might move into international technical support. Someone who understands AI and also truly understands a specific process in healthcare, law, logistics, or manufacturing may be worth far more than another person who only studies prompts.

No single ability has to be the best in the world. If several abilities complement one another, the combination may place you in a much smaller group. Suppose 50 percent of people can program and 40 percent communicate well. If the two abilities are roughly independent, only about 20 percent may possess both. Add English, industry knowledge, and project experience, and the number who satisfy every condition shrinks further. Reality is not as perfectly independent as a mathematical model, of course, but the way of thinking matters. You do not have to defeat everyone on one dimension. You can add new dimensions until fewer and fewer people are genuinely competing with you.

There is also the ability to deal with people. It can sound vague, almost unlike a technical skill, but I increasingly think it is one of the abilities least likely to lose value in the age of AI. It includes negotiation, sales, building trust, aligning interests, persuading people, and managing conflict. At their core, these activities are not simple exchanges of information. If exchanging information were enough, an email would do the job.

Why would a client award you a one-million-yuan contract instead of choosing another company with similar technical specifications? Why would an exceptional person join your team? When two companies have conflicting interests, why would each agree to compromise?

Even an ordinary job interview is an exercise in dealing with people. Someone may be technically strong but unable to explain an idea clearly, perhaps even stumbling over every sentence. If they cannot make the interviewer believe they can cooperate, communicate, and solve problems, much of their technical ability may never become employability. Conversely, someone who is not the strongest technically but can express themselves clearly, respond quickly, understand a need, and establish trust will often find it easier to persuade others to give them a chance.

Underneath all of this are trust and strategic interaction.

AI can teach you what to say and even generate a flawless negotiation script, but the other person ultimately has to trust a human being, not a piece of generated text.

The final ability I consider genuinely scarce sounds almost laughably simple: actually finishing things.

There are too many smart people online. Many discuss macroeconomics; few make money consistently over the long term. Many discuss startups; few actually launch a product. Many study AI; few build something people want to keep using. Almost everyone has saved dozens of “must-read tutorials,” “deep dives,” and “industry trends,” only to end up with a bookmarks folder richer than their actual body of work. I fall into this trap often myself.

Research creates a comforting illusion that you are already taking action, when often you are only consuming information. The genuinely difficult work is usually boring: build the first version, send it to people, get criticized, fix the problems, and send it again; submit dozens of job applications, get rejected, and revise them; after studying English, actually open your mouth and speak to someone instead of saving another speaking tutorial.

The world has never lacked people who know what should be done. It lacks people who can move something forward piece by piece without a teacher watching, without a standard answer, and without even knowing whether there will be a reward at the end. I do not know the most precise Chinese name for this ability. There is an English word I like: agency. It can mean initiative, or a person’s ability to act on the world and change reality.

I think this may be one of the hardest things for AI to replace. AI can lower the threshold for completing a task, but a lower threshold does not mean everyone will actually do it. Fitness advice has been free on the internet for years, yet most people still do not have their ideal physique. English-learning materials are immeasurably richer than they were decades ago, yet only a minority can use a second language fluently. Startup advice is available in almost unlimited quantities, yet most people will never sell their first product.

Information is not the biggest bottleneck.

Much of the time, we are.

If I were to rethink what it means to “learn a skill” today, I would no longer ask only which skill is hottest. Popularity itself may be a warning sign. Once something becomes popular enough that every short-video creator is teaching it, every training company is selling a course on it, and every university is preparing a major around it, the new supply is usually already on its way.

The questions worth watching are different. Which problems will become more serious over the next few years? What are many people willing to pay to solve? Why are so few people solving it now? What exactly is the barrier: knowledge, experience, capital, a license, language, geography, willingness to bear risk, or simply the fact that most people find the work troublesome, unglamorous, or beneath them?

I especially like the last possibility.

Human beings have a remarkably stable tendency to chase opportunities that look impressive. Everyone wants to be an investor, product manager, AI engineer, or founder. Nobody wants to handle customer complaints. Nobody wants to repair machinery in the middle of the night. Nobody wants to travel to a factory in an unfamiliar country to fix an equipment failure. Nobody wants to deal with a pile of dirty, trivial problems that offer nothing worth showing off on social media.

The market does not care whether a job sounds cool.

If other people refuse to do it and customers genuinely need it, that can be the most direct form of scarcity there is.

Scarcity does not last forever, of course, and that is precisely the difficult part. Once a skill becomes profitable enough, smart people enter the field. Training companies appear, tutorials appear, software appears, and eventually AI may arrive as well. Yesterday’s scarcity can easily become tomorrow’s basic requirement.

Trying to find one skill and live comfortably from it for forty years feels increasingly unrealistic to me. The thing that works over the long term may not be mastering one skill that remains scarce forever. It may be developing an ability to keep discovering scarcity: seeing a growing need before other people notice it, working in the field for several years before the crowd arrives, building experience, clients, capital, relationships, and a reputation, then looking for the next curve once the field becomes crowded.

In a sense, this is not very different from arbitrage. In financial markets, once everyone discovers an arbitrage opportunity, the available spread quickly disappears. The labor market works the same way. If everyone knows a skill pays well, sooner or later it will no longer be so scarce.

That is why I increasingly dislike asking other people, “What should I learn for the future?”

The question itself may be wrong.

Nobody can give you a list that says Python is worth 10 points, English 8, and AI 9, then promise that your life will be secure once you finish learning them. The better question is this: over the next five years, which problems will become increasingly important? Why are they difficult to solve? Who is solving them now, and why are those people so expensive? Most importantly, why has everyone else chosen not to do it?

Find those problems, then become one of the few people who can solve them.

That, roughly, is what a scarce skill means.

最近找工作的时候，我越来越强烈地意识到一个问题：这个社会上会做事情的人其实已经太多了。打开招聘软件，随便找一个稍微正常一点的岗位，几百个人投简历已经不是什么稀奇的事情，一个刚毕业的大学生会用Office，会剪视频，会做PPT，会一点Python、Java，英语过了四六级，还能用ChatGPT帮自己写代码，把时间往前推十几年，这种人可能算是一个能力不错的大学生，现在却只是一个极其普通的毕业生。

不是大家变差了，恰恰相反，是大家都会的东西太多了。

这听起来有点反常识。按照学校教给我们的逻辑，一个人应该不断学习技能，掌握的技能越多，他的竞争力就越强，但市场显然不是这么运行的。市场并不会因为一个东西很难学，或者你为了学习它熬了多少个晚上，就给你更高的价格，它真正关心的东西其实非常简单：我为什么非得找你？

假设一个岗位有1000个人投简历，其中700个人都会Java，那么“会Java”当然是一项技能，但它已经很难成为你的竞争优势；如果另外300个人不仅会Java，还会Python，那么Python也不会神奇地让他们变得值钱，因为他们真正面对的竞争对手，是另外299个和自己差不多的人。很多人找不到工作之后的第一反应还是继续往自己身上叠技能，Java不够就学Python，Python不够就学Go，然后继续学Docker、Kubernetes、Redis、微服务，最后简历越来越长，人却没有明显变贵。

这个问题和投资其实很像。一个东西值多少钱，从来不取决于生产它花了多少力气，而取决于市场上有多少人想要，以及有多少人能提供；钻石并不会单纯因为挖它的人很辛苦才贵，路边的石头也不会因为搬起来很重就有人花钱买，技能本质上也是一种商品。一旦把这个东西想明白，很多看起来奇怪的现象就没那么奇怪了。

摄影曾经是一项专业技能，一个人要知道曝光、快门、ISO、镜头、构图，还要买一堆昂贵的设备，现在一个完全没有摄影知识的人拿着手机，也可以拍出二十年前普通摄影爱好者很难拍出来的照片；剪辑也是一样，以前想剪一个像样的视频，你得学Premiere、Final Cut，研究时间轴、转场、调色，后来有了剪映，再后来AI可以自动识别字幕、删除停顿、匹配音乐，甚至直接根据文字生成视频。

编程正在经历同样的事情。

我本科就是计算机专业，所以对这件事的感觉尤其明显。几年前一个人会写一个完整的网站，从数据库到后端再到前端，至少还能算一个比较明确的技术能力；现在让AI写一个网站已经不是什么值得炫耀的事情了，你甚至不需要真正理解React或者Vue，只要大概知道自己想做什么，再和AI聊几个小时，一个能运行的东西大概率就出来了。

这当然不代表程序员要消失，也不代表编程没有价值，真正发生的事情是，“会写代码”这几个字本身正在贬值，而这两件事差别非常大。工业革命没有让制造业消失，只是让大量纯手工劳动的价值下降了；Excel没有消灭财务，但是让许多手工计算失去了过去的溢价；AI同样不会把所有脑力劳动全部消灭，它首先压缩的，大概率是那些已经被高度标准化、能够被清楚描述和重复执行的脑力劳动的溢价。

一个技能只要能够被清楚地描述、被系统地教学、被量化地验收，它最终就很容易被批量生产，而大学最擅长干的恰好就是这件事情：先编一本教材，规定教学大纲，再设计考试，最后一年生产几十万名掌握相似知识的人。所以我现在看到“某某专业非常热门”“某某技能未来缺口几百万人”这种新闻时，第一反应反而不是赶紧去学，而是想另外一个问题：如果所有人都知道它缺人，五年后它还缺人吗？

一个公开的红利通常活不了太久。互联网行业工资高的时候，大量的人开始学计算机；自媒体赚钱的时候，大量的人开始拍短视频；人工智能火起来以后，现在几乎每个人都想学AI，这其实非常正常，因为价格本身就会吸引供给，一个东西越赚钱，越多人涌进去，原本的稀缺性也就越容易被消灭。

所以我认为真正值得学习的并不是“难的技能”，而是“稀缺的技能”，而这两个概念经常被混在一起。高等数学很难，不代表学好高等数学就一定有人给你高工资；世界上还有大量非常冷门、非常复杂的知识，可能只有几百个人知道，但如果没有人需要它，它照样没有多少经济价值。

稀缺必须和需求同时存在。

有一万人急着解决一个问题，只有一百个人会解决，这才是真正有价值的稀缺。所以判断一项技能值不值得学，我现在更愿意看三个东西：有多少人需要它，有多少人能够提供它，以及掌握它之后到底能够影响多大的结果。沿着这个逻辑继续往下想，我反而觉得未来最稀缺的东西，可能根本不是某一种软件，也不是某一门编程语言。

第一个越来越稀缺的东西，是发现问题的能力。

AI非常擅长回答问题，但前提是你知道应该问什么。比如一家公司说：“我们要用AI降本增效。”这句话听起来很正确，实际上基本等于什么都没说，一个普通人可能马上开始研究大模型、Agent、RAG，然后做一份几十页的PPT告诉老板人工智能有多么厉害；真正有价值的人却可能先跑去看公司的业务流程，客服到底花了多少钱，哪个环节最浪费人，错误率最高的地方在哪里，这个问题到底需要AI，还是写一个五十行的脚本就能解决，甚至有没有可能公司真正的问题根本不是效率，而是产品压根没人买。

能够回答问题的人很多，能够发现真正的问题的人少得多，而当AI让“答案”越来越便宜以后，“问题”自然就会变贵。

第二个东西是判断力。

现在互联网最不缺的就是信息。你想买一只股票，可以让AI在一分钟里面给你生成一份几千字的分析报告；你想找工作，可以让它分析十家公司；你想创业，它甚至可以一次给你列出一百个商业模式。问题是然后呢？这项资产到底值不值得买，这个工作到底该不该去，这个生意到底有没有人付钱，AI可以给你二十个理由支持，也可以马上再给你二十个理由反对，但最后那个按钮还是要有人按。

信息越便宜，判断反而越重要。过去信息本身就是门槛，有些人赚钱只是因为他知道别人不知道的事情，现在这种优势正在快速消失，未来真正拉开差距的可能不再是你能不能找到信息，而是在所有人拿着差不多的信息时，你能不能得出一个比别人更好的结论。

而判断力这个东西很讨厌，因为它没有一门“判断力101”课程可以让你学完拿证，它更多来自长期的经验、失败、反馈，以及一个人是否愿意不断承认自己以前的判断是错的。很多能力可以通过重复练习迅速进步，判断力却往往要靠真实世界一次又一次把错误拍到你脸上，才能慢慢长出来。

第三个东西甚至都不能算传统意义上的技能，那就是承担责任。

我以前一直觉得社会给一些管理者那么高的工资有点奇怪，有些经理看起来什么具体的事情都不会干，代码不是他写，PPT不是他做，数据也不是他统计，凭什么赚得比下面的人多？后来我慢慢发现，真正值钱的部分往往不是“做”，而是“决定”。

程序员可以提出三个技术方案，CTO必须选一个；分析师可以写几十页报告，基金经理必须决定买还是不买；十个人可以坐在会议室里面讨论一个项目，最后必须有一个人站出来说：就这么干，出了问题我负责。执行、分析、判断、决策、承担结果，这条链越往后，人通常就越少，因为大多数人都喜欢权力，但是很少有人真正喜欢责任。

做对了大家都开心，做错了可能被骂、被开除、亏钱，严重的时候甚至要承担法律责任，所以责任本身其实就是一种稀缺资源。AI可以替你生成方案，可以帮你列出几十种可能性，但是AI不会真的替你坐牢，也不会替公司承担一千万的损失，最后总得有一个真实的人把自己的名字签上去。

还有一种我觉得经常被低估的稀缺，是技能之间的组合。

单独看每一样东西可能都不稀缺。会英语的人很多，会写代码的人很多，懂销售的人也很多，懂新能源的人也不是什么珍稀动物，但如果一个人同时懂软件、能够用英语跟海外客户沟通、了解一些新能源设备，还能在现场把客户的问题真正解决掉，那么符合条件的人一下就少了很多，而这种组合所产生的竞争优势，可能比单纯把Java从“熟练”刷到“精通”更有意义。

很多人一直在同一条轴上和别人竞争，程序员拼谁刷的算法题更多，设计师拼谁的软件用得更熟，剪辑师拼谁的特效做得更复杂，问题是只要所有人都在同一条轴上竞争，最后迟早会卷成价格战。看看现在的IT就业市场就能看出来，只要不是特别有竞争力的院校和履历，很多学生连找到一份像样的实习都不容易，大量人拿着高度重复的项目，投递高度相似的简历，也许HR每天看到这些简历都已经有点麻木了。

有时候最简单的办法反而不是继续卷下去，而是换一条轴。一个普通程序员加上销售能力，可能变成一个很强的SaaS销售或者售前人员；一个普通工程师加上英语和海外经验，可能变成海外技术支持；一个懂AI的人，如果再真正理解医疗、法律、物流或者制造业中的某一个具体流程，他的价值可能远高于另外一个只会研究Prompt的人。

单项能力根本不需要做到世界第一，只要几个能力恰好能够互补，组合之后就可能进入一个人数明显更少的区域。比如假设50%的人会编程，40%的人擅长沟通，如果这两项能力近似独立，那么同时具备两种能力的人可能只剩20%左右；继续叠加英语、行业知识、项目经验之后，真正同时满足所有条件的人会进一步减少。现实当然不会像数学模型一样完全独立，但这种思路本身很重要：你不一定非要在一个维度上击败所有人，也可以通过增加新的维度，让真正和你竞争的人越来越少。

还有人与人打交道的能力。这个东西看起来很“虚”，甚至不像技术，但我越来越觉得它可能是AI时代很难贬值的一类能力，包括谈判、销售、建立信任、协调利益、说服别人、管理冲突，这些事情的核心并不是简单的信息交换，如果只是交换信息，发一封邮件就够了。

一个客户为什么愿意把一百万的合同交给你，而不是另外一个技术参数差不多的公司？一个优秀的人为什么愿意加入你的团队？两家公司利益冲突的时候为什么愿意各退一步？

甚至最普通的一场求职面试，本质上也是人与人打交道能力的一种体现。一个人技术能力很强，但完全不能把自己的想法讲清楚、甚至说话都结巴的人，也无法让面试官相信他能够合作、沟通和解决问题，那么他的技术能力很可能无法完整地转化成就业竞争力；反过来，一个技术未必最强，但能够清楚表达、临场反应快、理解需求、建立信任的人，往往更容易让别人愿意给他一次机会。

背后都是信任和博弈。

AI当然可以教你怎么说话，甚至可以替你生成一份完美的谈判话术，但对方最后信任的是一个人，不是一段生成出来的文字。

最后一个我认为非常稀缺的能力，说起来甚至有点可笑，那就是把事情真的做完。

网上有太多聪明人了，谈宏观经济的人很多，真正长期赚到钱的人很少；讨论创业的人很多，真正把产品发布出去的人很少；研究AI的人很多，真正做出一个有人愿意持续使用的产品的人也很少。每天收藏几十篇“必看教程”“深度好文”“行业趋势”，最后电脑里的收藏夹比自己的实际成果丰富得多，这种事情几乎每个人都干过，我自己也经常掉进这种陷阱。

研究一个东西会产生一种很舒服的错觉，好像自己已经在行动了，其实很多时候只是在消费信息。真正困难的事情往往非常无聊：把第一版做出来，发给别人，被骂，修掉问题，再发一次；投几十份简历，被拒绝，再去修改；学英语以后真的张嘴和别人说，而不是继续收藏口语教程。

这个世界上从来不缺知道应该怎么做的人，缺的是在没有老师监督、没有标准答案、甚至不知道最后有没有回报的情况下，还能自己把事情一点一点往前推的人。我不知道该给这种能力起一个什么最准确的中文名字，英文里有一个词我很喜欢，叫Agency，你可以把它理解成主动性，也可以理解成一个人主动改变现实的能力。

我反而觉得这是未来最不容易被AI替代的东西之一，因为AI可以把完成一件事情的门槛降低，但门槛降低并不代表所有人都会真的去做。健身的方法在互联网上早就免费了，大部分人照样没有理想的身材；英语学习资料比几十年前丰富不知道多少倍，真正能够流畅使用第二语言的人仍然只是少数；创业教程几乎无限供应，大部分人一辈子也不会真正卖出自己的第一个产品。

信息不是最大的瓶颈。

很多时候，人自己才是。

所以如果现在让我重新理解“学习技能”这件事情，我不会再单纯问什么技能最热门，因为热门本身甚至可能是一个危险信号。一个东西一旦热门到所有短视频博主都在教，所有培训机构都开始卖课，所有大学都准备开专业，它的供给通常已经在路上了。

真正值得观察的问题应该是另外几个：未来几年什么问题会越来越严重，什么东西很多人愿意付钱解决，为什么现在解决这个问题的人还很少，这个门槛到底是什么？是知识，是经验，是资本，是牌照，是语言，是地理位置，是承担风险的意愿，还是大家单纯嫌它麻烦、不体面、不愿意干？

我尤其喜欢最后一种。

人类有一个很稳定的特点，就是喜欢追逐那些看起来光鲜的机会。大家都想做投资人、产品经理、AI工程师、创业者，没人喜欢处理客户投诉，没人喜欢凌晨修机器，没人喜欢跑到陌生国家的工厂里面解决设备故障，也没人喜欢处理一堆肮脏、琐碎、没有任何社交媒体炫耀价值的问题。

但市场并不在乎一件事情听起来酷不酷。

如果别人不愿意干，而客户又真的需要，有时候这反而就是最直接的稀缺。

当然，稀缺本身也不会永远存在，这恰恰是最麻烦的一点。一个技能一旦足够赚钱，聪明人就会涌进来，培训机构会出现，教程会出现，软件会出现，最后AI也可能进来，昨天的稀缺很可能就是明天的标配。

所以试图找到一种技能，然后靠它舒舒服服吃四十年，我觉得越来越不现实。真正长期有效的东西，可能不是掌握某一个永远稀缺的技能，而是拥有一种不断发现稀缺的能力：在其他人还没注意的时候，看见一个需求正在增长，在人群大规模涌进来以前先做几年，积累经验、客户、资本、人脉和信誉，等这个领域逐渐拥挤之后，再去寻找下一条曲线。

某种意义上，这和套利没有太大区别。金融市场里，当所有人都发现一个套利机会之后，套利空间就会迅速缩小；劳动力市场也是一样，如果所有人都知道某个技能工资高，它迟早就不会那么稀缺。

所以我越来越不喜欢问别人：“未来我应该学什么？”

这个问题本身可能就问错了。

没有人能够列一张清单，然后告诉你Python值10分、英语值8分、AI值9分，照着学完人生就安全了。真正应该问的是：未来五年，会出现哪些越来越重要的问题，这些问题为什么难解决，现在谁在解决，他们为什么这么贵，最重要的是，为什么其他人没有去做？

找到这些问题，然后让自己变成少数能够解决它们的人。

这大概才是所谓的稀缺技能。

Bitcoin can seem mysterious. One bitcoin is already worth tens of thousands of dollars, and it is familiar to many institutions and investors, yet very few people understand how it actually works. Most people are still puzzled by why it costs so much. Some even think it was invented as a scam or a Ponzi scheme. I see it as a utopian company created by a vast number of equal "employees." It offers a one-of-a-kind currency and an entirely new way to save. I believe that even a complete beginner can understand what Bitcoin really is after reading this article.

Put simply, Bitcoin is a distributed ledger. Traditional bookkeeping is usually handled by one person or a small number of institutions, separate entities such as banks and securities firms. That creates serious trust problems: companies can falsify their accounts, and banks can refuse to honor people's deposits. A few years ago, a township bank in Henan refused to recognize its customers' deposits. Residents who went to demand their money back were even beaten. Bitcoin does not have this problem. Every miner on the network helps maintain the ledger. If one miner's copy differs from everyone else's, the network identifies it as a malicious node and rejects it. This is how the network protects its credibility.

The Bitcoin network consists of blocks linked together in a chain, which is the blockchain most people have heard of. Each block records transaction data. If A sends B one BTC, for example, that transaction is written into a block. A block can hold roughly 4,000 transactions. As of August 1, 2026, the number of blocks is approaching one million. "Mining" is tied directly to these blocks. Bitcoin miners use their computers to work out the hash, a kind of character string, for the next block. The first miner to get it right earns the right to record that block. The electricity used in mining scales with the number of miners. Tens of thousands of miners now operate around the world, making an attack extremely expensive and helping keep Bitcoin secure. Mining difficulty also adjusts as the number of miners changes, keeping the average time between blocks at about ten minutes. A transfer usually takes around 20 to 30 minutes because most institutions and applications consider a transaction safe only after at least three confirmations. Once a miner includes a transaction in a block, each new block counts as one confirmation. Users choose their own transaction fees, and miners give priority to transactions that pay more.

Besides miners, the Bitcoin network has two other roles: developers and users. Developers are the core group responsible for maintaining the security of the network. They put forward ideas in the community, and an idea is added to the project only after community members agree on it. Most ideas never make it that far. Bitcoin is a system with a huge number of users. Constant updates would make the network less stable, and both developers and users are bound to disagree. The wishes of one group cannot simply override everyone else's.

In fact, you can count the major updates in Bitcoin's history on one hand: Segregated Witness around 2015, the Lightning Network in 2016, and the Bitcoin Cash fork, known as BCH, in 2017. Each had a practical purpose. Segregated Witness was designed to save space in each block. A block used to be limited to 1 MB, and the update increased its effective capacity. The Lightning Network makes small payments faster and allows them to settle almost instantly. BCH came out of a disagreement within the community. Some people thought blocks were too small and transactions too slow, so they wanted to fork a new coin with 4 MB blocks. Others believed that larger blocks would reduce decentralization.

Users are the final participants in the Bitcoin ecosystem. They decide whether to buy and use bitcoin. If the number of miners falls and the network becomes less secure, users can refuse to use it, while developers can look for ways to improve the network. If developers were captured by big capital and changed the rules to favor the powerful, miners could refuse to mine and users could stop buying, causing the price to fall. A falling price would also make hackers less interested in attacking the network. Few people would take on that much work for so little reward. Miners, developers, and users therefore form a subtle separation of powers. Bitcoin will not disappear simply because one of these groups collapses.

More importantly, Bitcoin's total supply is capped at 21 million by its code. No person or institution has the authority to issue more. That makes it an almost perfect currency. Anyone familiar with the monetary history of different countries can appreciate how precious that is: the Gold Yuan notes issued by the Republic of China government a century ago, the modern Turkish lira, and the Iranian rial, whose exchange rate recently collapsed because of the war between the United States and Iran. I could spend an entire day listing currencies that have suffered similar collapses throughout history.

As long as you keep your bitcoin secure, no third party or institution can freeze it. This stands in sharp contrast to the heavy-handed behavior of some banks. If a modern bank suspects that an account contains questionable transactions, freezing it is easy. That can protect users, but it can also cause a great deal of inconvenience.

One last thing is Bitcoin's price volatility. Its trading is not subject to any controls, so the price can move violently. On some days, it swings by more than 10 percent. Anyone who wants to invest in bitcoin needs to control their position size, or it may be bad for the heart. I do not think the volatility matters very much. Stretch out the price chart of almost any asset and you will find that none has escaped major swings. Both the Nasdaq and the S&P 500 have suffered drawdowns of more than 70 percent. Many investors call Bitcoin risky while also betting that supposedly stable assets will never run into trouble. Bitcoin's high volatility is a good thing because it creates a naturally antifragile structure: repeated small losses help prevent a much larger loss in the future. For more on antifragility, I recommend Nassim Nicholas Taleb's Antifragile. It is an important book for any investor.

比特币这个东西似乎很神秘。虽然价格已经达到几万美元一颗了，并且很多机构和投资者都知道它，但是了解它原理的人却很少。大部分人对它为什么这么贵还是一头雾水，甚至觉得它被发明出来就是为了诈骗和庞氏骗局。我个人认为它是由众多平等的员工创造出的一个乌托邦公司，提供了一种独一无二的货币，并且是全新的储蓄方式，通过我写的这篇文章，我相信即使是一个新手也能理解比特币到底是个什么东西。

简单来说比特币是一个分布式账本。传统的记账通常是一个人或者一些机构，这些一般都是独立的个体，比如银行、证券公司等。这就导致了很多信任问题：公司财务造假，银行不兑现居民存款；前几年河南某乡镇银行就出现了不承认用户存款的问题，居民去讨债甚至还遭到殴打了。而比特币就不会有这个问题，它是由网络上所有矿工共同记账的，只要其中一个矿工的账本和其他人不一样那就会被识别成恶意节点而被踢出，这样就能保证整个网络的信用。

比特币的网络是由区块组成的，由链串在一起，这样就构成了大家熟知的区块链。每个区块都用来记载交易数据，比如A转给B 1个BTC，这个交易数据就会写在区块上，一个区块大概能存4000条交易。截至目前（2026年8月1日）区块数已经快达到100万个了。大家知道的“挖矿”就和区块有关，比特币的所有矿工都在用自己的计算机算出下一个区块的哈希值（一种字符串）是什么，第一个算对的矿工就有了记账权。“挖矿”期间消耗的电力和矿工数量成正比，现在全球有数万个矿工同时挖矿，这就使得黑客攻击的成本极高，保证了比特币的安全。挖矿难度也会随着矿工人数动态调整，使得平均出块时间控制在10分钟左右。而每笔转账的时间大概在20-30分钟，因为大部分机构和应用认为一笔交易至少要经过三次确认才是安全的（矿工打包交易后，每出一个区块视为一次交易确认）。交易手续费由用户自己设定，矿工会优先打包手续费高的交易。

比特币网络中除了矿工之外，还有另外两个角色：开发者和用户。先说程序员，这个群体是维护整个网络安全的核心角色，他们负责在社区里提出idea，在社区成员一致同意后更新到项目中。当然大部分idea都没有实现，因为比特币是一个用户量很庞大的系统，如果动不动就提交一个更新那网络的稳定性就不能保证了，而且开发者也好用户也好，很容易产生分歧，不能为了一部分用户的想法忽略另一部分。

其实历史上比较大的更新一只手都数的过来，2015年左右的隔离见证、2016的闪电网络、还有2017的比特现金（Bitcoin Cash，简称BCH）分叉。而且这几个更新都是具有实用性的，隔离见证是为了节省每个区块（Block）的空间，过去一个区块只有1MB，这个更新之后提高了有效容量；闪电网络的用处是提高小额支付的速度，实现即时结算的效果；而BCH是因为社区内意见不合，当时有些人认为区块空间不够，导致交易速度慢，所以想分叉出一个新的币（BCH每个区块空间为4MB），但是另外有些人又觉得这降低了去中心化的程度。

用户则是比特币生态的终端角色，他们可以选择是否购买和使用比特币。如果因为某些情况使矿工数量减少，导致了比特币网络的安全性减弱，用户有权利拒绝使用它，开发者也可以想办法优化网络；如果开发者被大资本控制，为了利益更新了有利于权贵的一些规则，那么矿工可以拒绝挖矿，用户也可以拒绝购买从而导致比价下跌。并且在这期间，价格下跌了也会打消黑客攻击网络的想法，毕竟没人会做这种吃力不讨好的事情。从这些现象我们可以看到，矿工、开发者、用户这三者构成了一种三权分立的微妙关系，比特币不会因为其中一个群体的崩溃而消亡。

更关键的是，比特币的总量被限制在了2100万枚，这是由代码规定的，任何人、任何机构都没有权力增发。这也使它成为了一个近乎完美的货币。如果你了解世界上一些国家的货币史，你就会体会到这是何等珍贵的一种货币，远到一百年前中华民国政府发布的金圆券、现代的土耳其里拉、还有近期因为美伊战争汇率崩盘的伊朗里亚尔。如果要我一一列举历史上诸如此类汇率暴跌的货币恐怕一整天也写不完。

还有，只要你个人保管得当，没有任何第三方人和机构能冻结你的比特币。这就和一些银行的流氓行为形成鲜明对比，现代银行如果怀疑你的账户上有可疑交易，想冻结那是轻而易举的事情，虽然这对用户的安全有好处，但有时候还是会造成很多不便。

最后，我想提一嘴比特币的价格波动。因为它的交易没有任何管制，这就导致了它的价格起伏非常剧烈，有时候甚至日内波动能达到10%以上，所以想投资比特币的人一定要控制好自己的仓位，否则对自己的心脏不太好。我个人认为它的价格波动不是一件重要的事，你可以试着把所有资产的价格曲线拉长，没有一个是没经历过大波动的。纳指也好，标普500也好，最多也跌出70%以上的跌幅。很多投资者认为比特币风险高，但他们同时也在赌那些稳定性资产不会出问题。比特币的高波动性是一件好事，因为它形成了一个天然的反脆弱结构，通过连续的小损失避免了未来的大损失。关于“反脆弱”这个概念我推荐看纳西姆塔勒布的《反脆弱》，这对投资者来说绝对是一本很重要的书。

Abstract: I recorded seven IELTS listening and reading sessions. I took a combination of L-theanine and caffeine for four of them and took neither for the other three. The final entry came from the real exam. Average listening and reading scores in the combination group were 0.54 and 0.75 bands higher, respectively, while the average across both sections was 0.65 bands higher. The combination was used on the days when I scored better, but this does not prove that it directly improved my IELTS score. The experiment involved only one person and had no randomization, blinding, or placebo control.

1. The Short Version

L-theanine and caffeine may improve alertness and attentional control for a short period, helping me stay focused through sustained mental work such as IELTS listening and reading. Randomized controlled studies have also found small improvements in response speed or accuracy on some attention tasks. The effect varies with the task and the time of testing, however, and does not appear consistently across every measure.

A more accurate conclusion is this: the combination may reduce fatigue and fluctuations in attention, indirectly helping someone make better use of the ability they already have. It is not a score-boosting drug, and it cannot replace language skills, sleep, or practice.

2. The Combination and My Method

2.1 What I Took

• L-theanine: about 200 mg each time, from a common, reputable brand.
• Caffeine: a target dose of about 100 mg, giving an approximate 2:1 ratio of L-theanine to caffeine.
• Caffeine source: Swanson guarana capsules, a plant extract that naturally contains caffeine.
• Test tasks: IELTS listening and reading mock tests.

The caffeine content of guarana can vary between batches, so the number of capsules cannot be converted directly into an exact caffeine dose. A more reproducible experiment should use the milligrams of caffeine stated on the product label, rather than recording only the number of capsules.

2.2 How I Recorded the Results

The experiment covered routine practice and preparation for the real exam. I used the combination in four of the seven sessions and took neither substance in the other three, which served as the control. Each time, I completed the same types of IELTS listening and reading tasks and recorded my scores.

3. Results

Across the seven records, the combination group had a higher average in both sections. The difference was more noticeable in reading, while listening scores varied less. The chart keeps the score from every session visible so that the averages do not hide the differences between individual tests.

Several other explanations remain possible. The test sets may not have been equally difficult. Practice alone may have led to improvement. Sleep, mood, and testing time were not controlled. I knew whether I had taken the combination, which may have created an expectation effect. With such a small sample, one high score can also shift the mean substantially. The chart therefore shows a correlation and a personal experience, not a causal finding.

4. Why It Might Work

Caffeine raises alertness by blocking adenosine receptors. It often reduces drowsiness, but it can also cause nervousness, a faster heart rate, and excessively narrow focus. L-theanine is an amino acid found in tea. Some research suggests that it may soften the subjective tension caused by caffeine and improve certain measures of attentional control. The aim of taking them together is not to feel more stimulated, but to make wakefulness feel steadier.

The available evidence broadly fits that explanation:

These studies measured cognitive performance in laboratory tasks, not IELTS scores. They show only that the combination may help with attention; they do not support a fixed increase in test scores.

5. Safety Limits

1. Do not try the combination for the first time on the day of the real exam. Test your physical response first with a mock exam at the same time of day and a similar level of intensity.
2. Caffeine can cause insomnia, anxiety, heart palpitations, stomach discomfort, and trembling. Anyone who is sensitive to caffeine should use a lower dose or avoid it.
3. The U.S. FDA notes that, for most healthy adults, 400 mg of caffeine per day is not generally associated with negative effects. Individual responses vary widely. This is a reference limit for safety, not a recommended dose.
4. Avoid pure caffeine powder and highly concentrated caffeine liquids. They are difficult to measure accurately, and an overdose can cause serious harm.
5. People who are pregnant, have cardiovascular disease, anxiety, or a sleep disorder, or take medication should consult a doctor first. Supplements should not be treated as entirely risk-free.

6. How I Could Test It More Reliably

For another round of testing, I would try to keep the source of the questions, testing time, sleep duration, and diet consistent. I would randomly assign combination days and control days in advance, then record the actual dose in milligrams, perceived focus, adverse effects, and scores over many more sessions. If possible, identical-looking placebos and third-party coding could reduce expectation effects.

For me, these seven records justify cautious further observation, but they are nowhere near enough to claim that the combination definitely raises scores. Reliable gains still come from language ability, timed practice, and sleep. At most, L-theanine and caffeine may help performance stay closer to its normal upper limit.

References

1. Haskell CF, et al. The effects of L-theanine, caffeine and their combination on cognition and mood. Biological Psychology. 2008.
2. Giesbrecht T, et al. The combination of L-theanine and caffeine improves cognitive performance and increases subjective alertness. Nutritional Neuroscience. 2010.
3. Einöther SJL, et al. Effects of L-theanine and caffeine on attention and task switching. Appetite. 2010.
4. Williams JL, et al. The Cognitive Effects of L-Theanine and Caffeine: A Systematic Review and Meta-Analysis. 2025.
5. U.S. Food and Drug Administration. Spilling the Beans: How Much Caffeine is Too Much?

摘要：我记录了 7 次雅思听力与阅读训练，其中 4 次服用茶氨酸与咖啡因组合，3 次未服用。最后一次是真实考试的场景。服用组的听力、阅读均分分别比未服用组高 0.54 和 0.75 分，两科合并均分高 0.65 分。这个结果说明组合的使用与更高成绩同时出现，但由于样本只有 1 人，且没有随机、盲法和安慰剂控制，不能证明它直接提高了雅思成绩。

一、结论先行

茶氨酸与咖啡因组合可能在短时间内改善警觉性和注意控制，从而帮助我在雅思听力与阅读这类持续用脑任务中保持状态。已有随机对照研究也观察到该组合对部分注意任务的反应速度或准确率有小幅改善，但效果取决于任务和测试时间，并不稳定覆盖所有指标。

因此，更准确的表述是：该组合可能减少疲劳和注意波动，间接帮助发挥已有能力；它不是提分药，也不能替代语言基础、睡眠和训练。

二、组合与实验方法

1. 使用方案

• 茶氨酸：每次约 200 mg，选择常见正规品牌。
• 咖啡因：目标量约 100 mg，使茶氨酸与咖啡因约为 2:1。
• 咖啡因来源：Swanson 瓜拉纳胶囊，一种含天然咖啡因的植物提取物。
• 测试任务：雅思听力和阅读模拟训练。

需要注意，瓜拉纳原料的咖啡因含量可能存在批次差异，胶囊数量不能直接换算为准确的咖啡因剂量。如果要提高实验可重复性，应以产品标签标注的咖啡因毫克数为准，而不是只记录“几粒”。

2. 记录设计

实验分为日常测试和真实考试准备两个阶段。7 次记录中有 4 次服用组合、3 次不服用，后者构成对照。每次完成同类型的雅思听力和阅读任务，并记录成绩。

三、结果

7 次记录中，服用组的两科均值均高于未服用组。阅读的差异更明显，听力的波动相对较小。图中也保留了每次测试的原始分数，避免只看均值而忽略单次差异。

不过，数据不能排除以下解释：不同套题难度不同；训练本身带来进步；睡眠、情绪和测试时间没有控制；我知道自己是否服用，可能产生期待效应；样本量过小，单次高分会明显改变均值。因此，图表反映的是相关性和个人体验，不是因果结论。

四、为什么可能有效

咖啡因通过阻断腺苷受体提高警觉性，通常能减轻困倦，但也可能引起紧张、心率加快和注意过度集中。茶氨酸是茶叶中的氨基酸，部分研究认为它可能缓和咖啡因带来的主观紧张，并改善某些注意控制指标。两者组合的目标不是让人“更兴奋”，而是让清醒状态更平稳。

已有证据与这一解释基本一致：

这些研究测量的是实验室认知任务，不是雅思分数。因此，只能说明组合具备帮助注意表现的可能性，不能据此推断固定的提分幅度。

五、安全边界

1. 不要在正式考试当天第一次尝试。应先在同时间、同强度的模拟测试中确认身体反应。
2. 咖啡因可能引起失眠、焦虑、心悸、胃部不适和手抖；对咖啡因敏感的人应降低剂量或避免使用。
3. 美国 FDA 提到，多数健康成年人每日 400 mg 咖啡因通常不与负面影响相关，但个体差异很大。这是安全参考上限，不是建议剂量。
4. 避免纯咖啡因粉或高浓度液体。它们难以准确计量，过量可造成严重伤害。
5. 孕期、心血管疾病、焦虑或睡眠障碍患者，以及正在服药的人，应先咨询医生。补充剂也不能视为完全无风险。

六、如何做出更可靠的个人验证

如果要继续测试，应尽量固定题目来源、测试时间、睡眠时长和饮食，并提前随机安排“服用日”和“对照日”。记录实际毫克数、主观专注度、不良反应和成绩，至少积累更多重复数据。条件允许时可使用外观一致的安慰剂并让第三方编码，减少期待效应。

对我而言，这 7 次记录支持继续谨慎观察，但不足以宣称组合一定提分。真正稳定的收益仍来自语言能力、限时训练和睡眠；茶氨酸与咖啡因最多只是帮助状态接近正常上限。

参考资料

1. Haskell CF, et al. The effects of L-theanine, caffeine and their combination on cognition and mood. Biological Psychology. 2008.
2. Giesbrecht T, et al. The combination of L-theanine and caffeine improves cognitive performance and increases subjective alertness. Nutritional Neuroscience. 2010.
3. Einöther SJL, et al. Effects of L-theanine and caffeine on attention and task switching. Appetite. 2010.
4. Williams JL, et al. The Cognitive Effects of L-Theanine and Caffeine: A Systematic Review and Meta-Analysis. 2025.
5. U.S. Food and Drug Administration. Spilling the Beans: How Much Caffeine is Too Much?

I taught myself IELTS and scored 7.0 on my first attempt, spending nothing in total. This guide collects the study methods, resources, and score references I used. The views here are not official standards, and I have not independently checked whether every website is still available or how current the recalled-question banks are.

Study Tools

Computer-based mock tests: ieltscat.xdf.cn
Writing feedback: essay.art (free) and ielts9.me (paid)
Practice beyond official materials: Xiahua Listening and ZYZ Reading. There is a small chance of seeing a recalled question in the real exam, but I recommend finishing the official materials first.
Xiahua and ZYZ are ji jing: collections of real test questions recalled and compiled by people who have taken the exam. Cambridge IELTS books contain official practice tests published each year, and their questions will not appear again in the real exam.

1. The Overall Study Plan

1. Build a foundation: vocabulary and grammar.
2. Learn the computer-based format and choose your materials: use a mock-test website and Cambridge IELTS practice tests.
3. Work on each section: listening, reading, writing, and speaking.
4. Prepare for the final stretch: use strict time limits, review exactly why answers were wrong, and work through the speaking question bank.

The foundation stage matters. Formal test preparation becomes much more efficient once the basics are solid. Depending on your current level, I suggest allowing two to six months for this stage.
At its core, IELTS is about recognizing paraphrases.

2. Building the Foundation

2.1 Vocabulary

These are useful reference points for the minimum vocabulary needed:

My suggestions:

• Set a daily target that is realistic for you. I reviewed about 300 words a day and eventually reached a vocabulary of more than 9,000 words.
• Any mainstream vocabulary app will do. Shanbay Vocabulary and Bubei Vocabulary are two examples.
• The Daily English Listening app can be used to estimate your vocabulary size.
• Keep checking in every day until one week before the exam.

2.2 Grammar

I suggest building your grammar in this order:

1. Learn the five basic sentence patterns.
2. Learn conjunctions and use them to connect simple sentences into longer ones.
3. Learn clauses.
4. Move on to less central topics such as the subjunctive mood.

The point of studying grammar is not to memorize rules. It is to train yourself to identify the main structure of a long, difficult sentence quickly.

One way to test yourself is to ask AI to generate long sentences, mark the core structure on your own, then ask AI to check and correct your analysis.

Free course recommendation: the Bilibili creator “Larry 想做技术大佬” has a course called 《半个月搭建你的语法体系》, roughly “Build Your Grammar System in Half a Month.”

3. Computer-Based Practice

New Oriental IELTS Computer-Based Test Website

• Website: ieltscat.xdf.cn
• Use it for listening, reading, and writing mock tests.
• It is useful for getting familiar with the computer-based interface and the complete test process.

4. Preparing for Listening

Daily Practice

• Choose material at the right difficulty level in a listening app. I recommend CNN 10.
• After completing a Cambridge IELTS listening test, try intensive listening: play one sentence and write it down, then move to the next. I never tried this method myself, but from what I have heard, many people practise this way.
• Build faster grammatical reactions through repeated practice. When speech is hard to follow, speed is not always the only problem; sometimes the meaning of the sentence simply has not registered quickly enough.

What Different Scores Tend to Mean

• Band 6 listening relies more heavily on catching keywords.
• Band 7 listening means being able to understand the meaning of a whole passage.

Using Past Papers

• I recommend starting with Cambridge IELTS 10 and the books that follow it.
• The older Cambridge IELTS 4–7 books can be used for extra practice. Some questions differ in quality from the current exam, so do not worry too much if your results are poor.

5. Preparing for Reading

Using Past Papers

• Again, I recommend starting with Cambridge IELTS 10 and later books.
• If you have enough time, you can also work through the earlier books.

Three Main Reasons for Wrong Answers

1. Misreading the question: overlooking a word or misunderstanding the question because your grammar is not strong enough.
2. Misreading the passage: missing part of the information in a paragraph, or locating the right sentence without understanding it.
3. Insufficient vocabulary: failing to recognize a paraphrase between the question and the passage.

6. Strategies Shared by Listening and Reading

• Build your vocabulary and grammar before focusing on test-taking techniques.
• Recommended online course: the Bilibili creator “天狗吃月亮.” Focus on how to mark keywords and recognize paraphrases.
• You can practise reading without a time limit at first. If your accuracy is high when time is unrestricted, your foundation is probably not the main problem.
• Two months before the exam, you must begin practising under the official time limits.
• Whether you should use a timer from the very beginning depends on your own situation.
• Consider Xiahua Listening and ZYZ Reading only after finishing the Cambridge IELTS tests. The Cambridge books alone are enough preparation for the real exam.

7. Preparing for Writing

Study Method

• Search for Simon on Bilibili and learn his four-paragraph structure, along with the different question types for Task 1 and Task 2.
• Opinions about Simon's system vary online. I used his templates and received 6.5 in writing, which is already enough for most candidates.

Essay Feedback Tools

• Website: essay.art
• It is completely free.
• My essays usually received 6.0 on the site, while my actual IELTS writing score was 6.5.
• Website: ielts9.me
• The first essay is free; later submissions are paid.
• Its scores are relatively more accurate.

8. Preparing for Speaking

• Once you have built a foundation in grammar and vocabulary, find people from other countries to talk with on an online platform and practise real conversation.
• If you feel nervous at first, ask AI to prepare a topic, then record yourself speaking on your phone.
• Start working through the speaking question bank one month before the exam. I used “IELTS Bro” (雅思哥).
• Never memorize someone else's answers. Base your responses on your own experiences whenever possible.
• The speaking test has three parts. Part 1 is everyday conversation. In Part 2, you have one minute to make notes and then speak about one topic for two minutes. Part 3 explores broader questions in more depth. In my view, Part 2 is the most important. Aim to speak for at least one minute and forty seconds, and preferably for the full two minutes.
• Collect speaking questions and send them to AI, asking it to suggest useful material and expressions based on your own experiences. Your chat history can help it find a Part 2 story that suits you, including experiences you may not remember when you first see the question.
• If possible, avoid taking the speaking test during a question-change season. These usually fall in January, May, and September.

9. Preparation Timeline

10. Core Principles

1. Foundation first: techniques will not work consistently without solid vocabulary and grammar.
2. Execution matters: finishing your first IELTS paper or speaking your first sentence in English is already an important step.
3. Build skill with official practice tests: finish the Cambridge IELTS materials before turning to extra resources.
4. Identify the exact cause of every mistake: decide whether it came from vocabulary, grammar, comprehension, locating information, or time management.
5. Speak from real experience: especially in the speaking test, avoid memorized model answers and build responses around your own life.

The most important part of IELTS preparation is execution. Starting now and completing each day's work is usually more effective than endlessly looking for new resources and techniques.

我自学雅思首考7.0，总花费0元。本文保留了我的备考经验、资源推荐和分数参考。相关观点并非官方标准，网站可用性与题库时效性也未作额外核验。

备考工具

机考模拟网站：ieltscat.xdf.cn
写作批改网站：essay.art（免费）、ielts9.me（付费）
官方题以外的练习：虾滑听力、ZYZ阅读；虽然有小概率遇到原题，但是建议官方做完再做这些。
虾滑和ZYZ是机经，通过参加考试的考生总结出来的考场原题，有几率遇到。剑雅官方是每年出的官方题目，不会遇到原题。

一、备考路线总览

1. 打基础：词汇 + 语法。
2. 熟悉机考与备考资料：选择模拟网站和剑桥雅思真题。
3. 分项练习：听力、阅读、写作、口语。
4. 考前冲刺：严格限时、复盘错因、过口语题库。

基础阶段非常重要。基础打牢后再进入正式备考，效率会更高。建议根据个人水平安排 2～6 个月的基础期。
雅思的核心是同义词替换

二、基础阶段

1. 词汇

词汇量最低标准参考如下：

学习建议：

• 每天背多少单词应量力而行。我每天复习约 300 个单词，最终词汇量达到 9,000+。
• 单词 App 选择主流产品即可，例如扇贝单词、不背单词等。
• 可使用“每日英语听力”测词汇量。
• 建议每天坚持打卡，直到考试前一周。

2. 语法

建议按以下顺序建立语法体系：

1. 学会五种简单句。
2. 学习连词，把简单句连接成长句。
3. 学习从句。
4. 再了解虚拟语气等相对次要的语法点。

语法学习的主要目的不是死记规则，而是训练自己快速抓住长难句主干的能力。

检验方法：让 AI 生成长句，自己标记句子主干，再让 AI 检查和批改。

免费课程推荐：B 站 UP 主“Larry 想做技术大佬”的课程《半个月搭建你的语法体系》。

三、机考练习工具

新东方雅思机考网站

• 网址：ieltscat.xdf.cn
• 用途：练习听力、阅读和写作模拟考试。
• 适合用于熟悉机考界面及完整考试流程。

四、听力备考

日常训练

• 在听力 App 中选择难度合适的材料，推荐 CNN 10。
• 完成剑桥雅思听力题后进行精听：听一句，写一句。我没试过这个方法，但听说别人都是这么做的。
• 通过大量练习建立快速的语法反应。听不懂不一定只是因为语速快，也可能是没有及时理解句意。

能力差异

• 听力 6 分：更多依赖抓关键词。
• 听力 7 分：能够理解整段内容的语义。

真题使用建议

• 建议从《剑桥雅思 10》及之后的真题开始练习。
• 较早的《剑桥雅思 4～7》可作为辅助练习；部分题目质量与当前考试存在差异，做得不好不必过度在意。

五、阅读备考

真题使用建议

• 同样建议从《剑桥雅思 10》及之后的真题开始。
• 如果时间充裕，也可以练习更早的题目。

阅读错题的三个主要原因

1. 题目理解错误：漏看词，或因语法不扎实而误解题意。
2. 文章理解错误：段落信息没有看全，或虽然定位到答案句，却没有理解句意。
3. 词汇不足：没有识别出题目与原文之间的同义替换。

六、听力与阅读的共同策略

• 先打牢词汇语法基础，再学习做题技巧。
• 网课推荐：B 站 UP 主“天狗吃月亮”，重点学习划关键词和识别同义替换。
• 阅读训练初期可以不限时。如果不限时状态下正确率较高，通常说明基础问题不大。
• 考前两个月必须开始严格按照正式考试时间做题。
• 是否从一开始就限时应结合个人情况决定。
• “虾滑听力”和“ZYZ 阅读”可在做完剑桥雅思真题后再考虑；剑桥雅思已经足以应对真实考试。

七、写作备考

学习方法

• 在 B 站搜索 Simon，学习他的四段式写作结构，以及大作文、小作文的各类题型。
• 网络上对 Simon 的写作体系存在不同评价，至少我使用其模板后取得了写作 6.5 分，对大部分考生已经够用。

作文批改工具

• 网址：essay.art
• 完全免费。
• 我在该网站上的作文评分通常为 6.0，实际考试写作成绩为 6.5。
• 网址：ielts9.me
• 第一篇免费，后续需要付费。
• 其评分相对更准确。

八、口语备考

• 完成基础语法和词汇学习后，可以在线上平台找外国人聊天，训练真实交流能力。
• 如果刚开始容易紧张，可以让 AI 准备话题，然后对着手机录屏练习表达。
• 考前一个月开始过口语题库，我用的是“雅思哥”。
• 千万不要背诵别人的答案。尽量结合自己的真实经历回答问题。
• 口语分为三个部分，part1是日常对话、part2是围绕一个话题展开2分钟的描述，有一分钟做笔记时间、part3是深入挖掘其它话题。个人认为part2最重要，至少要说满1分钟40秒，最好说满2分钟。
• 可以把口语问题收集之后发给 AI，让它根据你的个人经历提供相关语料和表达方式。AI会根据你的历史对话记录给你推荐适合自己的part2语料，因为有时候你自己都不记得发生过什么和题目相符的故事。
• 如果条件允许，尽量避开换题季参加口语考试。换题季大致为每年 1 月、5 月和 9 月。

九、备考时间节点

十、核心原则

1. 基础优先：词汇和语法不牢，技巧很难稳定发挥作用。
2. 重视执行力：完成第一套雅思试卷、说出第一句英语，就已经迈出了关键一步。
3. 用真题建立能力：优先完成剑桥雅思真题，再考虑额外训练资料。
4. 复盘具体错因：区分是词汇、语法、理解、定位还是时间管理问题。
5. 坚持真实表达：尤其是口语，不背模板化答案，围绕自己的经历组织内容。

备考雅思最重要的是执行力。开始行动，并持续完成每天的学习任务，往往比不断寻找新的资料和技巧更有效。