Published : 10/02/2025
Artificial Intelligence (AI) development, long dominated by the United States, has recently seen the emergence of a rising star from China, DeepSeek.
DeepSeek's AI model has adopted a novel route with comparable performance and reportedly only one-thirtieth of the cost of the mainstream approach.
The rise of DeepSeek has also brought its founder Liang Wenfeng, born in the post-1980s generation, into the spotlight.
DeepSeek takes on OpenAI
In late January 2025, the relatively unknown AI company DeepSeek (深度求索), headquartered in Hangzhou, Zhejiang Province, released the reasoning model "DeepSeek-R1," which matched the performance of the "world's best" AI model by OpenAI, with development costs of only a few million US dollars and excellent cost-effectiveness.
It is known that the current development of mainstream AI models is akin to an "arms race," with growing demand for higher performance leading to increasing model sizes, exponential growth in parameter numbers, and remarkably high energy consumption and training costs, often totalling billions of dollars.
However, DeepSeek took an alternative path, not relying on cutting-edge hardware, but through ingenious engineering designs and efficient training methods utilising pruning, quantisation, knowledge distillation, and other innovative techniques to achieve "slimming down."
Who is DeepSeek's founder Liang Wenfeng?
The founder of DeepSeek is Liang Wenfeng (梁文峰), a member of the "post-80s" generation. He also holds another identity as the founder of the hedge fund High-Flyer (幻方量化).
From a young age, Liang showed extraordinary talent in mathematics and was already writing algorithms involving AI for stock selection during university.
In 2013, he founded an investment company; and two years later, he co-founded High-Flyer with his classmates. The latter quickly became one of China's largest quant funds, focusing on quantitative investment using AI.
The assets under management of High-Flyer surpassed the milestone of RMB 100 billion mark in 2021, positioning it as a leader in the domestic quantitative private equity sector.
In 2023, with the global AI boom, Liang announced a pivot to AI, and DeepSeek was officially established in July of the same year.
DeepSeek drives innovation with unique management style
Reportedly, DeepSeek has over a hundred R&D personnel with a high talent density, mostly nurtured domestically from top universities like Tsinghua University and Beijing University. They are described as "recruiting the top 1% of talent to do what 99% of Chinese companies cannot achieve."
In fact, the creation process of "DeepSeek-R1" is also very interesting. Its key cost-reducing framework, "MLA", originally stemmed from the personal interest of a young researcher.
According to Liang Wenfeng, the company structure is entirely bottom-up. There is no hierarchical relationships within teams, and each member is responsible for the part they are best at.
It is only when there is new progress that resources are deployed from the top down.
In addition, the company is highly flexible in deploying AI chips and manpower. If someone has an idea, they can access AI chips at any time without approval. As there is no hierarchy and cross-departmental boundaries, all manpower can also be flexibly allocated.
Liang Wenfeng stated that the reason why such a loose system can be implemented is that they have selected a group of people driven by passion for their work.
The company's recruitment criteria have always been passion and curiosity, so it has attracted a number of employees with unique experiences, many of whom have a strong desire for research that far exceeds their interest in money.