No matter how powerful AI becomes, it cannot replace the judgment that comes from hands-on practice and experience.
The other day, Tsinghua University started its new semester, and I took a look at the report on the university president's "first lesson of the semester", which I found quite interesting, so I'd like to discuss it with you today.
To start with, this sounds like it's being said to graduate students at Tsinghua University, but the final point of this speech, I think, is relevant to each and every one of us, especially if you're using AI in your work every day, and there's quite a bit of inspiration to be found.
On August 27, after the opening ceremony for the 2026 graduate class at Tsinghua University, President Li Qiming gave a lecture titled "Stand at the Forefront, Open and Inclusive, and Cultivate Exceptional Abilities through Unremitting Efforts" in the Comprehensive Gymnasium.
The structure of this lesson itself is what I want to talk about today.
Resources have been maximized, and the requirements have been brought back to basics
He first laid out the resources, explaining Tsinghua's forward-looking and comprehensive layout in the field of artificial intelligence, and the latest developments in how AI empowers teaching, scientific research, and discipline construction.
A freshman remarked after class that the school has made giving play to the enabling role of artificial intelligence a key task in its 15th Five-Year Plan.
The resources provided are extremely abundant, it's as if the entire school is putting all its assets on display to tell you: you've caught the best time.
However, I do have some reservations about this. A few years ago, I attended an EMBA program at Tsinghua, and I found that many of the internal information systems were actually not very user-friendly. But as an educational institution, it's probably unrealistic to expect too much in this regard.
After laying out his vision, he began to discuss the requirements for students, which can be summarized in four key points: enhancing awareness of problems, enhancing evidence-based capabilities, enhancing judgment capabilities, and enhancing a sense of responsibility. Additionally, he emphasized the importance of adhering to long-termism and working diligently with a down-to-earth approach.
The discrepancy between the two extremes is what left the deepest impression on me. The school provides top-notch resources, yet the expectations for students are lowered to the level of the weakest link.
My impression of these requirements can be summed up in four words: work diligently and honestly. I'd like to emphasize that this is my own condensed interpretation, not the principal's exact words.
But think about it, isn't it just about honestly using facts and data to speak, and then being able to take responsibility.
Large models can program, so why do students still need to learn?
However, just having this one sharing session is merely a gesture. What actually caught my attention was another speech given by Li Luming on June 27 this year at the undergraduate graduation ceremony. A full transcript of that speech is available, and it revolves around AI throughout.
He said that since the start of this year, the university has convened a series of seminars on talent development in the age of intelligence. On one question, faculty opinion was unanimous: even though large language models are already highly capable programmers, Tsinghua's science and engineering students must still learn to code.
What you see is not just the principal's personal sentiment, but a consensus formed by teachers during a round of discussions.
I was quite surprised when I read this, because in this era where AI has become a dominant field, it actually seems more trendy not to learn programming, or to just muddle through with a few vague phrases.
However, teachers unanimously believe that AI, no matter how powerful, still requires learning to program.
And it's not just a figure of speech. The official figure of speech is to make a speech without paying a price, you say on stage that you want to stay passionate, to look up at the stars, and everyone applauds, but in fact, they forget after a night's sleep. But behind this phrase, is a group of AI experts who have held a round of meetings.
The attendees were those who best understand large models, are closest to the forefront, and have the most resources in the country.
It must be acknowledged that, whether it's data, algorithms, or computing power, from hardware to software, Tsinghua's technological accumulation in this regard is absolutely formidable. Consider that the country's top large models, such as Zhiyun, actually originated from Tsinghua's laboratory in the early days.
They are fully aware of the level of code that current models can produce, and they are also fully aware of how much more time it takes for students to continue writing code by hand. However, experts overwhelmingly agree: it is still worth learning.
You could ask anyone in the AI industry, and they'd tell you it doesn't add up. Machines can already do the work faster and better than you can—so why would you want someone in their early twenties to retrace that entire path from scratch? What's the point?
But I understand that what the school is after is for the students to actually walk that road themselves.
The AI's response was that it couldn't go down that path.
Li Luoming was more direct in his speech at the graduation ceremony, saying that completely outsourcing thinking to large models, although seemingly convenient and efficient, may actually lead to weakened cognitive abilities.
He also quoted Mr. Yao Qizhi, saying that as artificial intelligence can replace more and more cognitive labor, many abilities still need to be formed through personal practice and going through a difficult process.
Understanding must come after experiencing chaos
In May 2025, I came across a study on brain neurons that mentioned the principle of brain insight. It said that you are able to remember something not because you saw the answer, but because your brain had previously experienced being stuck.
You have to start with a tangled mess, and only then can that mess be reorganized. In the instant that reorganization occurs, the original pattern of neuronal activation in the brain undergoes a structural shift, as if switching to a different channel, and it is at that moment that knowledge is engraved into long-term memory.
So, if you directly jump to letting AI give you the answers, what you're skipping over is not just the process of "thinking", but possibly the entire pathway by which the brain forms long-term memories.
AI responses may create false epiphanies
I once wrote down a sentence in my notes, but to be honest, I really don't remember whether I came up with it myself or read it somewhere. It mentioned that AI responses have three characteristics: logical consistency, fluent wording, and clear structure.
These three combined are likely to be an effective engine for manufacturing false epiphanies.
You may feel suddenly enlightened after reading it, but that feeling is likely an illusion. You've simply been persuaded by AI, not truly understood it, because you haven't gone through the process of piecing together the puzzle from chaos to clarity.
This is why, after all these years, primary and secondary education still places great emphasis on correcting mistakes. Because only by making mistakes first can one identify problems in their thinking; and only by identifying problems can one solve specific issues.
Being persuaded and truly understanding are fundamentally not the same thing in the brain. One is information that passes by and then leaves, while the other is memory that is carved in.
Looking back at Tsinghua's decision, it's different. Having students continue to write code by hand is not about preserving the code itself, but about preserving the mental struggle that occurs during the coding process.
Code can be generated by models, but the path to success cannot be traversed by models on your behalf.
AI is a magnifying glass, not a balancer
Li Luming also shared at the graduation ceremony a remark a student had made to him: AI is a magnifying glass — it amplifies ability, and it also amplifies laziness. The line came from a student, and he thought it was well put, so he brought it to the stage.
Judgment will also be magnified several times over
I particularly like the term "magnifying glass". AI is not an equalizer in many cases, it won't simply bring insiders and outsiders to the same starting line.
If you have discernment, it helps you apply discernment to ten projects. If you don't have discernment, it helps you apply the lack of discernment to ten projects.
This is no longer just a matter within the school.
I previously reviewed a case in which Deloitte Australia was commissioned by a government department to prepare a compliance review report under a A$440,000 contract. After submission, the report was found to contain fabricated court rulings and several academic papers that did not exist. Deloitte ultimately refunded part of the fee.
The original task of this report was to assess whether a system was compliant. However, a report on compliance acceptance itself failed to pass AI verification.
Domestic government departments have also openly written to remind people of the same thing, saying that AI has increased the speed and quantity of materials, making the data appear detailed, but in reality, it cannot withstand scrutiny.
You see, the people writing the material aren't lacking in tools, but rather someone who can review it before it's submitted and spot the problems at a glance.
Evidence-based capabilities are of utmost importance
So in the first lesson of the semester, among the four words mentioned by Li Luoming, from my own perspective, the most important one is actually the ability to trace evidence, or in other words, whether you can trace back the origin of a statement.
Everyone is talking about being aware of the problem, and judgment ability sounds like a slogan, but the two words "evidence-based" are very specific: it means tracing back to the point where it stands up to scrutiny.
This capability is not about knowledge, but rather a habit, and it can only be developed by going through the process oneself.
When you see the perfect results given by AI, you should also ask it to provide the data or case sources, and then spend some time to verify whether these sources are reliable.
However, if these words are only applied to students, they are wasted.
If you strip away the identity of being a student, the situation is the same as ours: tools are dozens of times more powerful than they used to be, yet the work we submit is easier to see through than ever before.
Only by doing it personally can one have discernment and a sense of feel.
Speaking of which, I'd like to share with you a small incident that happened in my life last weekend. My partner and I went to a nearby building materials market, where we actually just wanted to replace a small cabinet.
We had AI prepare a very detailed specification document in advance, with various indicators written in great detail. For example, the mirror on the surface, AI told me that I must ask my boss to provide me with ultra-white copper-free silver mirror.
I told my boss the requirements and had him write them into the order. He smiled at me and said, "Was it Dou Bao who told you this?" Then he pulled me to a wall where many mirrors were hung.
He asked, do you know which one is the mirror you want? Should I give you the ultra-white or the regular white one, how do you tell them apart?
I was stuck right there, I had the parameter report given by the AI, but I couldn't make a judgment at all.
I mentioned this because it's actually similar to today's topic: why do students still need to learn programming in the AI era?
The key point is not to force students to endure hardship, but rather to have them personally experience things, which will allow them to develop different judgment and tactile senses regarding the outcome.
Of course, this applies far beyond programming. Many fields work the same way. For instance, no matter how impressive AI-generated images become, they can't replace the hand-drawn work of designers—because what's cultivated in that process is a sense of aesthetics.
Related Links
President Li Luoming delivered the "First Lecture of the New Semester" to the 2026 graduate students of Tsinghua University
Tsinghua University recently held a ceremony to commemorate the establishment of the Tsinghua University - Huawei Joint Innovation Laboratory, with the goal of further deepening cooperation between the two parties in the fields of artificial intelligence, cloud computing, and 5G. The laboratory will focus on cutting-edge technological research and development, as well as talent cultivation, and will provide a platform for the exchange of ideas and collaboration between Tsinghua University and Huawei. The ceremony was attended by Yang Bin, Vice President of Tsinghua University, and William Xu, Director of the Board and Chief Strategy Marketing Officer of Huawei. In his speech, Yang Bin emphasized the importance of cooperation between universities and enterprises in driving technological innovation and economic development. He expressed his hope that the joint laboratory would become a model for cooperation between universities and enterprises, and would make significant contributions to the development of China's high-tech industry. William Xu stated that Huawei is committed to investing in technological research and development, and is willing to work with Tsinghua University to cultivate top talent and drive innovation. He believes that the joint laboratory will play a key role in promoting the development of China's artificial intelligence, cloud computing, and 5G industries. The establishment of the Tsinghua University - Huawei
President Li Luoming's speech at Tsinghua University's 2026 undergraduate commencement ceremony
Tsinghua University and Huawei Technologies Co., Ltd. jointly established the Tsinghua-Huawei Innovation Research Institute, aiming to promote cutting-edge technological innovation and achieve breakthroughs in key areas. The institute will focus on research and development of emerging technologies such as artificial intelligence, 5G, and the Internet of Things. The establishment of the Tsinghua-Huawei Innovation Research Institute marks a new level of cooperation between Tsinghua University and Huawei, with the goal of cultivating innovative talents, advancing scientific research, and driving industrial development. The institute will provide a platform for researchers and engineers from both parties to collaborate and innovate, promoting the transformation of scientific and technological achievements into real-world applications. Tsinghua University President Qiu Yong and Huawei Rotating Chairman Xu Zhijun attended the unveiling ceremony of the Tsinghua-Huawei Innovation Research Institute. In his speech, Qiu Yong emphasized the importance of cooperation between universities and enterprises in driving technological innovation and economic development. Xu Zhijun expressed his expectation that the institute would become a hub for innovation and talent cultivation, contributing to China's technological advancement and global competitiveness. The Tsinghua-Huawei Innovation Research Institute will be located in the Tsinghua University Science Park, with an initial investment of