ByteDance Seed recruits Qianwen's top general

The competition and fusion in the AI industry are also reflected in talent mobility.

On March 12, following the departure of Lin Junyang, the head of the Qwen large model technology at Alibaba’s Tongyi Laboratory, the whereabouts of another core team member finally surfaced.

Industry sources say that Yu Bowen, the former post-training lead of Qwen, has officially joined ByteDance as the head of post-training for the Seed team’s vision models and multimodal interaction team.

An insider close to ByteDance confirmed the personnel change to Wall Street Journal.

This personnel shift occurred shortly after Alibaba’s Qwen team completed an organizational restructuring and several key technical talents left en masse, sparking widespread industry attention on talent flow and technological competition within the domestic large model field.

Yu Bowen’s academic and technical background is considered solid within the industry. Public information shows he graduated with a bachelor’s degree from Central South University, then pursued graduate studies at the Institute of Information Engineering, Chinese Academy of Sciences, earning his Ph.D. from the University of the Chinese Academy of Sciences in 2022.

During his doctoral studies, he focused on natural language processing and information extraction, publishing multiple papers at top international conferences such as ACL and EMNLP. He innovatively proposed transforming information extraction tasks into graph-structured problems, effectively solving recognition challenges in complex scenarios like entity overlap and nesting, earning the Chinese Academy of Sciences President’s Award for outstanding academic performance.

After earning his Ph.D. in 2022, Yu Bowen joined Alibaba DAMO Academy through Alibaba’s top-tier campus recruitment program “Alibaba Star,” as an algorithm expert (P7). Early in his tenure, he deeply participated in the early training and development of the Tongyi Qwen large model, quickly becoming a core team member and eventually taking on the role of post-training lead.

Yu Bowen’s departure is closely related to recent organizational adjustments at Alibaba’s Tongyi Laboratory.

In March, Alibaba Tongyi Laboratory initiated a restructuring plan to split the originally vertically integrated Qwen team into several parallel modules, including pre-training, post-training, text, and multimodal divisions. This change significantly reduced Yu Bowen’s management scope and created a clear conflict with his long-held belief that “pre-training and post-training must be deeply coupled.”

Additionally, the commercial performance pressures imposed by Alibaba’s senior management on the Qwen team have also intensified internal disagreements.

On March 3, Yu Bowen submitted his resignation, leaving the next day. His work was subsequently taken over by Zhou Hao, a former senior researcher at Google DeepMind.

Yu Bowen’s next move also reflects a new focus in the current large model competition.

ByteDance’s Seed team has been continuously investing in large models and multimodal research in recent years. His joining as head of vision models and multimodal interaction post-training indicates ByteDance is strengthening its “post-training” capabilities in multimodal development.

Post-training is a critical stage for transforming large models from general-purpose foundations into products and scenarios, directly affecting their performance in real-world interactions.

Yu Bowen’s experience in dialogue model optimization, multimodal alignment, and knowledge distillation aligns highly with the team’s current technical layout. Especially in visual and multimodal interaction, how to use efficient fine-tuning and reinforcement learning to make models more “understanding” of users has become a key differentiator among major companies.

Yu Bowen’s move from Alibaba to ByteDance exemplifies a core talent flow in this round of AI competition. In January, Hui Bin, head of Qwen Code, left Alibaba to join Meta. Earlier, international giants like OpenAI, xAI, and Meta also experienced significant talent movements.

This wave of talent mobility reveals several deep changes in the industry’s development:

First, the era of technical elites is reshaping the relationship between talent and platforms. As large model technology continues to evolve rapidly, the personal judgment and vision of top technical talent have a greater influence on technological directions than ever before. When company strategies conflict with individual technical ideals, talents tend to seek platforms that better realize their visions.

Second, computing resources and organizational collaboration have become key variables in talent retention. Purely financial incentives are no longer sufficient to lock in top talent; providing adequate computing support and building organizational structures aligned with their technical philosophies are now more critical.

Third, multimodal and post-training techniques are becoming the main battleground for talent competition. As foundational large models’ capabilities converge, how to differentiate through post-training and deeply integrate visual and language abilities has become the strategic focus for leading companies. Yu Bowen’s move to ByteDance’s multimodal team exemplifies this trend.

For the industry, the movement of core talents is both a challenge and a catalyst. It forces companies to rethink how to collaborate with top talents and accelerates the cross-platform dissemination and collision of technological ideas.

As the development of large models is still ongoing, the flow of talent is, to some extent, shaping the future landscape of technological competition.

Risk Warning and Disclaimer

Market risks exist; investments should be cautious. This article does not constitute personal investment advice and does not consider individual users’ specific investment goals, financial situations, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article are suitable for their particular circumstances. Investment is at your own risk.