@MaxForAI: A Japanese who just completed a CS master's at Zhejiang University wrote an article "Why China's AI Is Developing" that is closer to reality than many self-congratulatory analyses on the Chinese internet. Its core is not to praise China or criticize Japan, but to point out a simple fact: China's AI talent did not appear out of thin air; they were trained by a whole set of harsh but effective systems. ...

A Japanese who just finished a CS master’s at Zhejiang University wrote an article on “Why China’s AI is developing.” It’s closer to reality than many self-congratulatory analyses on the Chinese internet.

Its core isn’t to praise China or criticize Japan, but to point out a very simple fact:

China’s AI talent didn’t appear out of nowhere—they were trained by a brutal but effective system.

The Gaokao determines your school and major. Once you enter university, it’s not easy either, because good jobs are limited. Many people start preparing for the graduate entrance exam as early as their freshman year.

The author argues that Chinese students essentially go through two rounds of intense screening: one is the Gaokao, the other is the graduate entrance exam. As a result, their foundations in math, English, and learning endurance are all pushed to a very high level.

Then at the graduate level, the pressure continues to increase.

Taking his own experience in the CS master’s program at Zhejiang University as an example, to graduate you need to pass internal review, external expert review, a defense committee, and also have a paper accepted at a CCF-recognized conference or journal, or hold a Chinese patent.

The author therefore believes that the large number of AI-related patents in China is not just because companies file many, but also related to the graduation mechanism in universities.

He also mentions that PhD requirements are even stricter. CCF conferences are classified as A, B, or C, with Class A including top conferences like CVPR, NeurIPS, AAAI, etc.

According to him, a Chinese PhD needs 2 Class A papers and 1 Class B paper to graduate. This mechanism steadily produces PhDs with proven capabilities.

He also offers several interesting judgments:

First, he believes China’s AI can develop because CS really can change your destiny. In a country with average low income, top CS fresh graduates can earn far above the average. So both parents and students will pour resources into CS.

Second, he thinks China is good at catch-up innovation but finds disruptive innovation more difficult. The reason is that the education system is too systematized—people are better at optimizing within the system, while the US is better at exploration.

Third, he believes GPU export controls are effective. Without GPUs, universities struggle to do influential AI research, so top students go to big companies with GPUs for internships. Schools without compute power can only do smaller-scale research.

Fourth, his view on Huawei’s chips is relatively tepid. He says they may have value in the long run, but for now they are not yet at a practical level, and he’s never met a Chinese AI student who is genuinely optimistic about Huawei’s chips.

The strength of China’s AI is not a myth, nor a miracle. It’s the product of countless people who started competing fiercely from their teens.

This is very well written—it’s a brutally honest reality.

Goshi Aoki (@goshi_aoki): I completed a master’s in Computer Science at Zhejiang University in China one year ago. Based on my interactions with top Chinese AI talent and my experience in graduate school, I’ve summarized China’s AI talent cultivation system below.