When Even the Top Experts Are Anxious: AI Isn't Taking Away Your Ceiling, It's Raising the Floor
Hello everyone, I'm your financial journalist and friend, also an economist.
Recently, the tech world has been in an uproar after Liu Shengyu, an operator engineer at DeepSeek, published a lengthy article that quickly made it to the trending topics. Many people, upon seeing headlines like "Even Top Engineers Are Anxious About AI" and "They Must Change Careers," thought, "Oh no, are people from Peking University's Turing Class and international supercomputing champions going to lose their jobs? Are we going to survive?"
If you only read about "job loss" and "anxiety," you might have misunderstood the core of the article. What Liu Shengyu really wants to convey is something much deeper and more sobering than just worrying about being replaced. He's not complaining about losing his job; he's warning us that AI hasn't improved the average level of coding skills. Instead, it has dramatically increased the production of "junk code"—code that is functionally useless but still appears to be well-written.
Today, we'll break down this article into five key points to understand what this technological transformation means and what it implies for us ordinary people.
---
1. First, let's clarify: He's not giving up on "working," but on the identity of a "skilled craftsman"
To understand Liu Shengyu's anxiety, we need to know who he is. He's not just an ordinary programmer who copies and patches code. He graduated from Peking University's Turing Class and has competed in international supercomputing competitions. At DeepSeek, he's in charge of optimizing the most fundamental parts of the systems.
Imagine an AI model as a supercar: ordinary programmers might just change the tires or paint the car, while engineers like Liu Shengyu are refining every piston and valve inside the engine, even studying how gasoline molecules can explode most efficiently within the cylinders. This requires knowledge of hardware, assembly language, and underlying logic—typical skills of a skilled craftsman, acquired through years of experience and a high level of intelligence.
In the past, such skills were rare and provided a competitive advantage. But now, Liu Shengyu has realized that AI outperforms humans in terms of thinking speed and parallel processing. AI can execute 300 steps in a second and write a line of code in half a second, something humans can't match.
The key point is this: He's not worried about not being able to write code anymore; he's worried that even if he writes the best code, AI could catch up with him or even surpass him in just half a year. It's like a top chef whose skills used to rely on knife skills and cooking techniques, but now AI robots can cook ten thousand dishes at the same time. His uniqueness is disappearing because his core skill—writing high-quality code—is becoming something that machines can replicate easily.
So, when he says he needs to change careers, he means that the era of making a living by writing code manually is over, and a new era has begun, where we manage AI to write code for us.
---
2. The most stark insight: AI hasn't eliminated junk code; it's just made it look better
This is the most chilling part of the article and a truth that most people overlook. In the past, a poor programmer's code was easily recognizable: messy variable names, chaotic logic, and lack of comments. Now, AI can generate code with well-defined variables and clear comments, making it look professional.
The result is: The code looks good, but the underlying errors and risks remain. It's like turning a shoddy street vendor's product into a high-end garment with a fancy label. Liu Shengyu says, "A person with poor engineering skills, when paired with AI, can produce junk code at an even higher rate."
What does this mean? In the past, low-quality code was quickly eliminated because it was obvious. Now, it can slip into systems unnoticed. AI raises the "output volume," not the quality. When the market is filled with code that appears competent but actually hides problems, the stability of the entire system is at risk. This is what he means by saying the world is becoming more precarious.
---
3. The shift in roles: From "driver" to "mechanic pilot"
If writing code by hand is no longer effective, what can top engineers do? Liu Shengyu proposes a new role: the "mechanic pilot" of AI systems.
In the past, programmers were like drivers, controlling every aspect of the system. Now, they're more like pilots of a robotic mech. They no longer write the code themselves but give instructions to AI: "Scan that area on the left," "Avoid that obstacle," "Attack with maximum power."
The key skills for this new role are:
- Global control: Knowing where the system is vulnerable and what needs improvement.
- Boundary setting: Understanding what AI can and can't do, and setting clear limits.
- Evaluation and judgment: Identifying the best solution from multiple options.
Liu Shengyu says, "I have more tools at my disposal, but I feel less in control." Writing code used to be a satisfying experience, like breaking speed records in a game. Now, we just describe the requirements, and the machine does the work. The sense of creation and control is gone.
For ordinary people: If you're still debating whether to learn Python syntax, you might be late. The valuable skills of the future are "how to ask AI the right questions" and "how to judge the accuracy of AI's answers." You need to transition from being an executor to an evaluator and architect.
---
4. The gap in education: We're still assessing "brainpower" by "speed"
Liu Shengyu raises a critical sociological issue: Our current education and assessment systems are out of step with the AI era. For example:
- Tests often ask if you can write a sorting algorithm by hand in 45 minutes.
- Interviews require you to write a linked list on the spot.
- Performance is measured by the number of lines of code you submit.
These metrics are worthless in an AI world where AI can write perfect algorithms or generate thousands of lines of code instantly.
The truly valuable skill is: The ability to make informed decisions. In other words, among the countless options provided by AI, who gets to say, "This doesn't work; start over?" Who decides which version to release and who is responsible for the system's outcome?
This is the return of "power." In the past, ability was a filter—without it, you couldn't enter certain fields. Now, everyone can produce something that seems competent, so the new filter is "position" and "responsibility." Our education, hiring, and promotion systems reward speed, not judgment.
This leads to a gap: top engineers like Liu Shengyu have the judgment needed, but they're losing their uniqueness. Many ordinary engineers, with AI's help, produce a lot of code, but they lack the judgment to identify and fix problems.
Who fills the gap in judgment? No one. There's no training for rejecting AI's wrong suggestions, and no one is held accountable for the long-term risks of the systems they create.
---
5. The ultimate warning: When the ability to say "no" becomes the most scarce resource
Liu Shengyu asks a philosophical question: "In the future society, will power be more important than technology or intelligence?"
The logic is this:
- AI makes it extremely easy to produce something.
- Therefore, producing something alone no longer determines its value.
- Value lies in "deciding what to produce" and "what not to produce."
- Whoever has the power to say, "This plan won't work because it has long-term risks," holds the real power.
This is what Liu Shengyu is warning us about: He's not just worried about losing his job. He's saying that when everyone can produce a lot, the ability to say "no" will become the most valuable and crucial skill.
We're collectively skipping the step of developing this ability. We're using AI to speed up tasks but weakening our thinking and judgment.
For ordinary people:
- Don't just chase speed and quantity; they're less valuable in an AI world.
- Practice judgment regularly. Ask yourself why a solution is good, where its limits lie, and what the consequences of its mistakes are.
- Allow yourself to make mistakes. Judgment comes from trial and error. Turn off AI and write code from scratch occasionally to maintain your intuition and control over the system.
- Focus on responsibility, not just output. In the future, your value will lie in your ability to oversee the results.
In summary, Liu Shengyu sees himself as being replaced by the technology he created, but his main concern is the deterioration of the entire system. He stays at DeepSeek because he believes that cutting-edge intelligence should be accessible to everyone, to prevent a future dominated by a few. This is the sober view of an idealist.
We need to shift from worrying about being replaced by AI to worrying about being overwhelmed by mediocrity. Don't fear that AI is faster than you; fear that you're not yet learning how to say "no" when necessary.
---
This analysis reflects the complex realities of the AI revolution and the challenges it poses for individuals and society. As we embrace AI, we must also develop the skills to navigate its impact wisely.