Where DeepSeek, Qwen's AI engineers really come from

Asia Times

17-02-2025

The recent rise of DeepSeek and Alibaba's Qwen artificial intelligence platforms is a hot topic among technology experts and investors in China and the United States. Yet specific information about the Chinese engineering teams remains scarce.
An Asia Times analysis using public information on the backgrounds of the two chatbots' developers may help draw a clearer picture of how DeepSeek and Qwen emerged.
Based on publicly available research papers and media reports, DeepSeek and Qwen's engineering teams do not collaborate or overlap.
The only linkage between the duo is that DeepSeek's researchers said in a paper on January 22 this year that they had 'distilled' Qwen2.5, and also Meta's Llama, to develop DeepSeek-R1. The launch of DeepSeek-R1 caused a slump in the US stock market in late January.
Some analysts believe that DeepSeek, an open-source AI, may also have used 'knowledge distillation' to extract data from OpenAI's ChatGPT and train its AI models. However, no conclusive evidence has been made public that it did so.
According to the January 22 paper, DeepSeek-R1 has 16 core contributors, some of whom have direct connections with Microsoft Research Asia (MSRA) and the China Computer Federation (CCF).
Core contributor Yu Wu was supervised by MSRA's Ming Zhou in his PhD program at Beihang University from 2014 to 2019. He was a full-time intern at MSRA from 2013 to 2019 before joining it as an associate researcher in 2019.
Core contributor Daya Guo was also supervised by Ming Zhou during his PhD program at Guangzhou's Sun Yat-sen University from 2018 to 2023. He was mentored by Nan Duan in 2020-2023 and by Duyu Tang in 2017-2020 in MSRA's Natural Language Computing Group.
Core contributors Zhibin Gou and Zhihong Shao, both from Tsinghua University, co-wrote papers with MSRA's Nan Duan.
Zhenda Xie, another contributor to DeepSeek, was advised by MSRA's distinguished scientist Baining Guo at Tsinghua University from 2018 to 2023. He also worked as a research intern at MSRA during the same period.
Zhou Ming and Nan Duan are still working for MSRA. In 2016 and 2018, they jointly led the CCF's Computer Terminology Approval Working Committee. Zhou is now a vice president at CCF.
The connection between DeepSeek's team and MSRA researchers, of course, does not mean that MSRA has any stake in the Hangzhou-based company. However, if the US were to exert more pressure on China's tech sector, MSRA may have to stop its work and internship programs in China.
In early 2023, MSRA reportedly stopped recruiting interns from seven Chinese universities and the Beijing University of Posts and Telecommunications due to concerns about their linkages with the People's Liberation Army.
The US-sanctioned Northwestern Polytechnical University (NPU) and the Harbin Institute of Technology (HIT) are among these schools.
The Hangzhou-based Alibaba, founded by Chinese tycoon Jack Ma, developed its chatbot differently.
Alibaba, mobilizing cash flow from its e-commerce businesses, established the DAMO Academy in 2017 to undertake AI research. DAMO stands for Discovery, Adventure, Momentum and Outlook.
The academy established an advisory board of 10 renowned educators and researchers, six from the US and four from Chinese universities.
Alibaba assigned Aliyun's Chief Technology Officer Jingren Zhou to run the DAMO Academy. Zhou earned his PhD degree in computer science from Columbia University in the United States. He was a Microsoft R&D partner for four years before joining Alibaba in July 2016.
Chang Zhou, an algorithm engineer who was responsible for Qwen's data processing, joined DAMO Academy in 2017. He graduated from Fudan University in 2012 and finished his PhD in Peking University in 2017.
Before joining Alibaba, he had already worked with Alibaba's software engineers on some projects and co-wrote two papers with them in 2017. Peking University professor Jun Gao was among the co-writers.
Gao received his PhD from Peking University in 2003 and has published more than 30 research papers. He has projects funded by China's 863 Program and the National Natural Science Foundation of China (NSFC).
In March 1986, 200 top Chinese scientists proposed the 863 Program to then-Chinese leader Deng Xiaoping. The government set up the program in November of the same year.
In 2022, Chang Zhou co-wrote a research paper with a group of academics, including Peking University's Bin Cui, who also led some projects funded by the 863 Program.
Cui is now the vice dean of Peking University's School of Computer Science and the deputy director of CCF's Technical Committee on Databases.
Last July, Chang Zhou decided to take along a team of about 10 engineers to join ByteDance. Alibaba reportedly filed a lawsuit against Zhou, claiming he wasn't entitled to join a competitor.
Alibaba's research team appears to have a stronger advisory board than DeepSeek. It also has a more extended history in data management research.
This may be why Alibaba could build Qwen2.5, an original AI model, while DeepSeek has only mustered distilled versions of other AI models. This is also likely why Apple Inc has recently partnered with Alibaba to launch AI-powered iPhones.
IT columnist Amanda Caswell wrote in a recent article that Qwen2.5 beats Deep-R1 in all seven tests she set out. She says Qwen2.5 offers more structured and readable answers, while DeepSeek-R1's responses lack depth and originality.
From Beijing's perspective, both Qwen and DeepSeek are equally essential as they help China upgrade its industry and hedge against US decoupling and sanctions.
Media reports said Chinese President Xi Jinping will soon chair a symposium to boost private sector sentiment. Alibaba co-founder Jack Ma and DeepSeek founder Liang Wenfeng will attend.
Yong Jian is a contributor to the Asia Times. He is a Chinese journalist who specializes in Chinese technology, economy and politics.
Read: Apple partners with Alibaba to sell AI-powered iPhones in China