<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/b7edb66a-bef0-4686-a820-412064ae9540.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/b7edb66a-bef0-4686-a820-412064ae9540.png"/> The competition and fusion in the AI industry are also reflected in the flow of talent.
On March 12, following the departure of Lin Junyang, the technical leader of the Qwen large model at the former Alibaba Tongyi Laboratory, the whereabouts of another core member of his team have finally come to light.
Industry sources indicate that Yu Bowen, the former head of Post-training for Qwen, has officially joined ByteDance, taking on the role of head of Post-training for the Seed team's visual model and multimodal interaction team.
Sources close to ByteDance have confirmed this personnel change to Wall Street Insight.
This personnel change occurs at a time when the Qwen team at Alibaba has just completed an organizational restructuring, with several core technical talents leaving in quick succession, raising widespread attention in the industry regarding talent flow and technological competition in the domestic large model field.
Yu Bowen's academic and technical background is solid within the industry. Public information shows that he graduated with a bachelor's degree from Central South University and later pursued graduate studies at the Institute of Information Engineering, Chinese Academy of Sciences, where he obtained his Ph.D. from the University of Chinese Academy of Sciences in 2022.
During his doctoral studies, he focused on research in natural language processing and information extraction, publishing multiple papers at top international academic conferences such as ACL and EMNLP. He innovatively proposed the idea of transforming information extraction tasks into graph structure problems, effectively addressing recognition challenges in complex scenarios such as entity overlap and nesting, and was awarded the Chinese Academy of Sciences President's Award for his outstanding academic performance.
After graduating with his Ph.D. in 2022, Yu Bowen joined Alibaba's DAMO Academy as an algorithm expert (P7) through Alibaba Group's highest-level campus recruitment program, &#34;Ali Star.&#34; In the early stages of his employment, he was deeply involved in the early training and research and development of the Tongyi Qwen large model, quickly becoming a core member of the Qwen team and ultimately serving as the head of Post-training.
Yu Bowen's departure is closely related to the recent organizational restructuring of Alibaba's Tongyi Laboratory.
In March, Alibaba's Tongyi Laboratory initiated an organizational restructuring, planning to split the originally vertically integrated Qwen team into several parallel modules for pre-training, post-training, text, and multimodal tasks. This adjustment directly led to a significant reduction in Yu Bowen's management scope and created a clear conflict with his long-held technical belief that &#34;pre-training and post-training must be deeply coupled.&#34;
Additionally, the commercialization pressure imposed by Alibaba's senior management on the Qwen team has also exacerbated internal differences within the team.
On March 3, Yu Bowen submitted his resignation, and the next day he officially left. His work was subsequently taken over by Zhou Hao, a former senior researcher at Google DeepMind.
Yu Bowen's next choice also reflects the new focus of competition in the large model arena.
ByteDance's Seed team has been continuously investing resources in the large model and multimodal fields in recent years. With Yu Bowen joining as the head of Post-training for the visual model and multimodal interaction team, it signifies that ByteDance is strengthening its &#34;post-training&#34; capabilities in the multimodal direction
Post-training is a key step for large models transitioning from a general foundation to productization and scenario application, directly determining the model's performance in actual interactions.
The experience accumulated by Yu Bowen during the Qianwen period in optimizing dialogue models, multimodal alignment, and knowledge distillation aligns closely with the current technical layout of the Seed team. Especially in the field of visual and multimodal interaction, how to make the model &#34;understand&#34; users better through efficient post-training fine-tuning and reinforcement learning has become a key factor for differentiation among major manufacturers.
Yu Bowen's transition from Alibaba to ByteDance is one of the core talent movements in this round of AI competition. In January, Qwen Code leader Hui Bin had already left Alibaba to join Meta. Even earlier, international giants like OpenAI, xAI, and Meta also experienced core talent movements internally.
This round of talent movement reflects several deep changes in the development of the large model industry:
First, the era of strong technical talent is reshaping the relationship between talent and platforms. Currently, large model technology is still in a rapid evolution stage, and the personal judgment and vision of top technical talents have a far greater impact on the technical path than before. When company strategies diverge from personal technical philosophies, talents tend to seek platforms that can better realize their technical ideals.
Second, computing resources and organizational collaboration have become key variables for talent retention. Simple salary incentives are no longer sufficient to lock in top talent; whether companies can provide adequate computing support and build organizational structures that align with the technical philosophies of talents is becoming a more important factor in retaining core personnel.
Third, multimodal and post-training are becoming the main battleground for talent competition in the next phase. As the capabilities of foundational large models gradually converge, how to achieve differentiation through post-training technology and how to deeply integrate visual and language capabilities have become the directions that leading manufacturers are competing to layout. Yu Bowen's joining of ByteDance's multimodal team is a reflection of this trend.
For the industry, the movement of core talent is both a challenge and a catalyst. It forces companies to rethink their collaboration methods with top talents and accelerates the cross-platform dissemination and collision of technical ideas.
As large models have not yet reached their conclusion, the flow of talent is, to some extent, mapping out the future landscape of technological competition

KWEB

512720

BYTED

159998

The battle for AI talent continues

Wallstreetcn

Krne Csi China Internet

Guotai CSI Computer Theme ETF

ByteDance

Tianhong CSI Computer Theme ETF

ByteDance Seeds Recruit Qianwen General