Someone around me had mentioned Chen Changwen before. Lately people have been forwarding his clips on WeChat Moments. I opened his page and felt physically ill. I still sat through a few, because I wanted to understand why the traffic is that big.

His move is seventy-five videos a day. The thesis is simple: if you are not successful, you did not post enough and you did not grind enough. His life philosophy is: do not deal with people, everyone around you is bad, making money is the only point, then put the money in a bank.

Then I hit a video where he described his own life, and the world model clicked. He said his father never stopped mocking him—worse before he made money, still mocking after. Ugly. Crooked mouth. Every angle he could find. Then Chen said rural people are probably all like that.

Street-smart cynicism

Chen is not alone. Short-video platforms are full of copy like this:

• “A friend will not give you another path. He will give you another danger. If someone cannot give you an advantage, he will give you trouble.”
• “Turn off Moments. Fall in love with being alone. Eat alone. Read alone. Run alone.”
• “The kinder and more sincere you are, the more bad people you meet.”
• “Be shamelessly transactional. If there is a payoff, play. If there is none, get lost.”
• “Go back to the mountains. No social life. No job. One person fed, the whole house fine.”

Call it closed egoism. That is modern cynicism: doubt every kindness without looking, treat self-interest as the only possible motive. In that picture there are two kinds of people: the openly bad, and the fake-good. The fake-good are worse.

As kids we inherit our parents’ traditional story. School also sells you a bright future. We build a Chinese traditional model almost without noticing: effort gets rewarded; eating bitterness is a virtue; money is saved, not made; favors are given, then returned. It barely has a model. It is linear cause and effect. You put in diligence and virtue; the other side is supposed to spit out the matching result. That line works at home and in school because it was designed for obedience and exams.

Then you grow up. Society is messy. People are messy. You get burned. You get hurt. You find out you have no leverage. The linear model cracks. Effort does not pay. Good people do not get good outcomes. Savings lose to inflation. A favor you lent comes back as a WeChat block. When people realize the path is wrong, the cheapest move is not to build a better thinking frame. It is to flip to the opposite. That is when the cynical gospel shows up. Work is just capitalists squeezing you. People are dangerous, so stay away. Money is the only goal. Not getting played is winning.

Traditional thinking and this street-smart cynicism are the same kind of machine: linear cause and effect. They just face opposite directions.

Feedback loops

People who actually get good at something build a feedback system and adjust. They do not ride one story into a wall.

Linear thinking has a hidden assumption: push the cause once, then accept the result. I worked hard, so I should be rewarded. I stayed guarded, so I should stop getting hurt. In the real world, cause and effect are not a line. They are a loop. You do A. A leads to B. B leads to C. C comes back and acts on A. That is a feedback loop. A factory is the boring version: buy materials, make a product, sell it for cash, use the cash to buy materials, make the next product.

There are two kinds.

1. Positive feedback

Positive feedback means the return strengthens the first move and amplifies the change.

Compound interest is one. Money makes interest. Interest makes interest. The snowball gets bigger. Cognition works the same way. The more you already understand, the easier new knowledge sticks, the more connections you get, the faster the next round goes.

2. Negative feedback

Negative feedback means correct the drift and hold a balance. If you wander off the set point, it pulls you back.

A home air conditioner is the textbook case. You set 25°C. The room goes above that, it cools. It gets close again, it stops.

“Positive” and “negative” sound like good and bad, so the cleaner names are reinforcing loop and balancing loop.

A reinforcing loop amplifies the first move. You write something good. People share it. Sharing brings new readers. New readers bring more shares. That is reinforcement. Reinforcement does not care about morals. The more you fear talking to people, the less you practice, the more you fear. That is reinforcement too. It is an amplifier. What it amplifies depends on what you put in.

A balancing loop pulls you back. You run a fever, you sweat, the temperature drops, the sweating stops. It does not want more. It wants the set point. It is a corrector. It keeps the system from running away.

Want growth? Start a reinforcing loop. Want stability? Build a balancing loop. That is the whole pair.

Traditional thinking and cynical success gospel fail in the same place: they are open loop. Open loop means you issue the command and you are done. You do not look at what the world sent back.

The traditional story says “try harder,” then never asks: is the direction right? Did the payoff structure change? The cynical story says “guard harder,” then never asks: did I block a real threat, or did I also block people I could have worked with? Both assume they were right on the first try. The rest is just turning up the volume. In a complex system, nobody is right on the first try.

A closed loop is different. A coach can talk swimming theory for an hour. One messy lap that makes you swallow water will teach you more. The water tells you what is wrong. Tasting the soup beats a hundred-line recipe.

There is a trap most people miss.

A system usually has more than one loop, some fast, some slow. You cannot sleep. A sleeping pill works tonight. That is the fast loop. Changing your schedule, dropping coffee, exercising, dealing with the anxiety—that is the slow loop. Fast loops show results now. Slow loops fix the actual problem. Under pressure, people grab the fast loop and squeeze the slow one out. Then you need the pill to sleep, and needing the pill makes the insomnia worse, until the body only runs on the drug.

Relationships work the same way. You get burned once. You block, shut down, stop trusting. Fast loop. It stops the bleeding tonight. The judgment you never practiced, the weak ties you never built, the cooperation credit you never saved—that is the slow loop. The fast loop can save this evening. The slow loop decides where you are in five years.

Cynical success gospel is poison not because it teaches caution. Caution is right. It is poison because it makes caution the only loop, then cuts every slow one. You spin faster and tighter inside the fast loop, and the world gets smaller.

How to build a feedback loop

What if you do not have one yet?

Explore first.

You cannot close a loop before you have something to close. After 1949, the path China took was a history of exploring. After reform and opening, nobody really knew how to run a market economy. Deng Xiaoping’s move was to build a closed loop. If you do not know, run a pilot. A special economic zone here. A new policy there. If something works, put a reinforcing loop on it and spread it. If it does not, use a balancing loop and pull it back.

If you just graduated and do not know which industry fits, explore. Do not hide from the job market inside another exam. Send the résumé. Take a job. After a year, be honest: did I grow? Do I like this? Does this company have a future? If yes, stay and do it well. If not, change. That has to be a slow loop. Quit after a few days and you will not get a real signal.

For founders, a closed loop means a growth flywheel. If you are going to start something, the first job is to build the loop: a smallest product, real users, real feedback, then another pass. Without that, it is easy to build something nobody asked for.

How do you get the first users? That is the hard part. The usual path is you already know the customer—old business friends who need this, so you can sell it when it exists. The second path is raise money, build, then pay for distribution. The third is build a founder personal brand. The last two almost always fail.

The human-AI loop

In the AI era you can put the model inside the loop. I call that the human-AI loop.

Codex, Doubao, or anything else—you can use it to raise your own operating level.

Inside that loop, AI plays two roles. First, a private advisor: it turns a fuzzy wish into a loop you can actually run. Second, a mirror: after a round, it reads the data, shows you where you drifted, and what to change.

Say you want to learn English.

Step one: set a goal.

“I want to learn English” is a wish, not a goal. AI should force it into something measurable. Learn it for what? Client calls? IELTS? Travel? Different goals, different loops. After a few rounds it should pin you to one sentence: “In three months I can run a thirty-minute client call in English without a translator.” That sentence is the output standard.

Step two: break the moves apart.

Once the goal is set, have AI split it into four pieces: input, output, feedback point, adjustment point.

• Input: what you do every day. Fifteen minutes of English with the AI, on topics you actually hit at work.
• Output: what you have to produce. After each session, retell three things from the talk in English.
• Feedback point: where it corrects you. Grammar, word choice, pronunciation, right then.
• Adjustment point: what changes next round. It keeps the errors you repeat and drills those.

This is where the model is worth the most. Most people break a goal into “vocab, listening, exercises” and stop. That is a line, not a loop. What you want is a ring: practice → output → get corrected → log the misses → drill the misses tomorrow → output again. Each round connects. Each round leaves something for the next one.

Step three: run one round.

Then run it. Do not wait until you are “ready.” The first round will look bad. You will stumble. Two sentences out of three will be wrong. Fine. The point of a loop is to read signal inside the mess. After one round you have data: what you missed, where you froze, which errors keep coming back. You do not get that data from solo practice.

Step four: read the feedback.

Hand the data to the AI. Its job is to show you what you cannot see. You think you are “okay.” It tells you sixty percent of last week’s errors were tense, mostly past tense. You think your listening is weak. It tells you that you understood; you were just too slow to build the answer.

The key here is honesty. The model will not soothe you and it will not sneer. It gives you the kind of feedback almost nobody around you has the patience—or the nerve—to give: specific, unemotional, pointed at the next move.

Step five: change one parameter.

After you read the feedback, change one thing, not five. If tense is the problem, the next round is past tense. Park the rest. Then go back to step three and run again.

That is the full ring: set a goal → break the moves apart → run one round → read the feedback → change a parameter → run again. Once it is turning, each round is more accurate than the last, because the change is based on real data, not a feeling.

Building a human-AI loop may be the biggest gift AI has given people.