Ex-Qwen Core Member Yu Bowen Joins ByteDance's Seed Team
Reports indicate that following the departure of Lin Junyang, head of large model technology at Alibaba's Tongyi Lab, Yu Bowen, former head of post-training for the Qwen (Qianwen) model, has recently officially joined ByteDance. He will lead the post-training efforts for the Seed team's visual model and multimodal interaction initiatives. While ByteDance has not issued an official statement, multiple sources confirm this move is part of the broader organizational restructuring at Alibaba's Tongyi Lab that began in early March.

This restructuring dismantled the previously vertically integrated Qwen team, reorganizing it into parallel modules focused on pre-training, post-training, and multimodal processing. This shift led to the departure of several key technical staff, including Yu Bowen, Kaixin Li (a core contributor to Qwen3.5/VL), and Hui Bingyuan, who had previously joined Meta. Yu Bowen, a recognized technical expert and "Alibaba Star," played a leading role in developing the Qwen series of chat models and possesses deep expertise in areas like multimodal alignment.
The Seed team at ByteDance, now under the leadership of Dr. Wu Yonghui—former Vice President of Research at Google DeepMind, who joined in February 2025—reports directly to CEO Liang Rubo. The team has already developed the Seed2.0 series of base models and the Seedance2.0 video generation model. As global competition in large models intensifies, with a critical focus on multimodal and visual understanding capabilities, ByteDance's recruitment of core talent from Tongyi underscores the strategic priority leading companies place on post-training technology. This move also signals that talent mobility within China's large model industry is accelerating a fundamental reshaping of the underlying technology landscape.
Related article
Marc Benioff-backed startup aims to solve AI deployment challenges
Big enterprises face such steep challenges in deploying reliable AI tools that entire teams of forward-deployed engineers—specialists embedded within client companies to operationalize AI systems—are emerging to bridge the gap.“Paradoxically, AI driv
AliVideo Unveils Wan3.0, Delivering Ultra-Long Video Generation and Multi-Dimensional Consistency
On August 24, Alibaba unveiled Wan3.0, its latest video generation large model. This release represents a major leap forward for Alibaba’s video technology, delivering significant improvements in duration, universal creation, reference handling, and
MiniMax Large Model Mispronounces Names - Xiyu Technology: Insufficient Training After Specific Tokens
MiniMax recently published a technical report detailing why its M2 series models struggled to accurately generate specific names like "Ma Jiaqi." This apparent error highlights a systemic flaw in current large language model training.Token Offset: Th
Related Special Topic Recommendations
Comments (0)
0/500
Reports indicate that following the departure of Lin Junyang, head of large model technology at Alibaba's Tongyi Lab, Yu Bowen, former head of post-training for the Qwen (Qianwen) model, has recently officially joined ByteDance. He will lead the post-training efforts for the Seed team's visual model and multimodal interaction initiatives. While ByteDance has not issued an official statement, multiple sources confirm this move is part of the broader organizational restructuring at Alibaba's Tongyi Lab that began in early March.

This restructuring dismantled the previously vertically integrated Qwen team, reorganizing it into parallel modules focused on pre-training, post-training, and multimodal processing. This shift led to the departure of several key technical staff, including Yu Bowen, Kaixin Li (a core contributor to Qwen3.5/VL), and Hui Bingyuan, who had previously joined Meta. Yu Bowen, a recognized technical expert and "Alibaba Star," played a leading role in developing the Qwen series of chat models and possesses deep expertise in areas like multimodal alignment.
The Seed team at ByteDance, now under the leadership of Dr. Wu Yonghui—former Vice President of Research at Google DeepMind, who joined in February 2025—reports directly to CEO Liang Rubo. The team has already developed the Seed2.0 series of base models and the Seedance2.0 video generation model. As global competition in large models intensifies, with a critical focus on multimodal and visual understanding capabilities, ByteDance's recruitment of core talent from Tongyi underscores the strategic priority leading companies place on post-training technology. This move also signals that talent mobility within China's large model industry is accelerating a fundamental reshaping of the underlying technology landscape.
Marc Benioff-backed startup aims to solve AI deployment challenges
Big enterprises face such steep challenges in deploying reliable AI tools that entire teams of forward-deployed engineers—specialists embedded within client companies to operationalize AI systems—are emerging to bridge the gap.“Paradoxically, AI driv
AliVideo Unveils Wan3.0, Delivering Ultra-Long Video Generation and Multi-Dimensional Consistency
On August 24, Alibaba unveiled Wan3.0, its latest video generation large model. This release represents a major leap forward for Alibaba’s video technology, delivering significant improvements in duration, universal creation, reference handling, and
MiniMax Large Model Mispronounces Names - Xiyu Technology: Insufficient Training After Specific Tokens
MiniMax recently published a technical report detailing why its M2 series models struggled to accurately generate specific names like "Ma Jiaqi." This apparent error highlights a systemic flaw in current large language model training.Token Offset: Th





Home






