The advantages of Chinese characters can be reflected from the APP list, and now they come to the big model

Chen Jing
Director of Research and Development, Visual Technology, Asia

1. Chinese apps can occupy six positions in the top ten in the US list. In the past, there were also a lot of Chinese apps on the Indian list, which was later banned, and recently 36 quietly returned. In various countries around the world, Chinese APPs are suddenly on fire. What is the reason?

2. The French TV program discussed that a sister was familiar with China and had a Chinese boyfriend. She explained that this is because the Chinese use Chinese characters, which is two-dimensional ideograph, while Westerners are one-dimensional linear characters. This leads to fundamentally different ways of thinking, with Chinese people paying more attention to the text and the picture as a whole, while Westerners pay more attention to local and focus. For example, the Western APP interface needs to highlight a button and a text prompt; and the Chinese APP first gives people a sense of the whole, the layout will be more far-reaching consideration, and the layer by layer.

3. Sister said that this thinking makes the Chinese APP obviously have design advantages, there are many excellent designs, can better meet customer requirements, and quickly meet common needs. If all Chinese apps can quickly complete the payment, the West needs to set up a bunch of things, sometimes tens of minutes.

4. Recently, the big model has made the world more aware of the advantages of Chinese characters, and some Western large models actually think with Chinese characters. Many Westerners are surprised that under severe restrictions, China has quickly made breakthroughs in large models. Some Western users commented that deepseek performed extremely well in English “creative writing”. This is another manifestation of the advantages of the Chinese character system, and it is more essential.

5. Chinese characters not only have the overall characteristics of two-dimensional, but also have a very different topology in the correlation between “word” tokens and linear text systems such as English. The biggest task of the big model is to establish the relevant weight for the characters of various words, the Chinese characters are obviously superior, the association is more logical, such as pork, beef, chicken, January, February, March, the association is very natural, but in English, it is all seemingly unrelated words. And Chinese characters can easily expand to include the integration into new concepts, no need to invent new words, and even reduce many words. The Chinese character system is “hard and easy”, and it is easy to learn to see what level of knowledge continues to progress, while Western society is prone to people with very poor knowledge, and many adults have the ability to read and understand.

6. Deepseek’s performance progress, in addition to large model data structure, training algorithm, code optimization, the news did not say, is the Chinese training data collation, evaluation. The industry said that Deepseek attaches great importance to data labeling, Liang Wenfeng himself to mark, please Peking University Chinese Ph.D. to mark. High-quality data, like strict teachers, has a very strong Chinese language ability. It seems to have the Chinese thinking as the base, with a different flexible thinking, the general response of the masses is different from the previous big model. Even if it is exported in English, it can also show ability in innovative writing.

7. Therefore, Deepseek has made a performance breakthrough and is likely to benefit from the Chinese system. The simple fact is that only Chinese and English can build a complete understanding of the human knowledge system. Many languages do not even have scientific vocabulary, and it is not enough to make new words, and it is better to learn English or Chinese.

https://user.guancha.cn/main/content?id=1383625