DeepSeek’s rise: Is the Chinese language the secret ingredient to its AI edge?

BEIJING, Feb 15 — DeepSeek, a Chinese artificial intelligence start-up, has recently gained global attention for its high-performing, cost-effective and open-source large language model (LLM).

Industry experts suggest that the model’s integration of Chinese characters during its pre-training phase has been a significant factor in its success, according to a report by South China Morning Post.

The use of Chinese characters, known for their high information density, is believed to enhance the model’s logical capabilities, enabling it to process complex concepts more efficiently.

“Chinese characters achieve maximum information transmission with minimal cost,” telecommunications industry analyst Xiang Ligang stated on social media.

As an efficient information encoding, Chinese has greatly improved efficiency and reduced costs in the processing of artificial intelligence.

Additionally, the multimodal nature of Chinese characters, which often combine visual elements with meanings, may provide rich learning material for AI models.

This characteristic could contribute to improved language comprehension and contextual understanding.

While DeepSeek has not publicly disclosed its training data sources, it is speculated that the model’s Chinese training data encompasses a diverse range of materials, including classical literature, internet slang, academic papers, government documents, and regional dialects.

This variety likely offers a comprehensive linguistic foundation, further enhancing the model’s performance.