It is rumored that ByteDance's Seed AI4S team is considering a separate spin-off, with core members Xiao Wenzhi and Gu Quanquan reportedly leaving to found an AI pharmaceutical company.

According to ME AI , Beating, a monitoring agency, reports that Seed, ByteDance's AI R&D department, is discussing a new round of organizational restructuring, even considering a spin-off from ByteDance. One proposed solution is for the AI4S team to be transferred from the Machine Learning Systems (AML) team led by Xiang Liang to Yang Zhenyuan's team. However, this plan is still under discussion, and sources close to ByteDance indicate that Yang Zhenyuan is not enthusiastic about the transfer. Since Wu Yonghui took over Seed, the three major teams under AI Lab—Seed Robotics, AI4S, and Responsible AI—have been successively merged into Seed. The reporting lines of the AI4S team have changed several times. Xiao Wenzhi's team, originally part of AML, merged into the AI for Science team led by Li Hang last year, with Xiao Wenzhi reporting to Li Hang instead of Xiang Liang. After Li Hang's retirement, the team returned to Xiang Liang's system. More concerning than the organizational structure changes is the loss of core technical backbone. Several core members of the team, including Quan-Quan Gu, Associate Professor of Computer Science at UCLA and co-leader of the Seed large-scale model pre-training and expansion team, and Wen-Zhi Xiao, the head of computational biology at the Protenix project, have left UCLA to start their own businesses. Their ventures focus on AI-driven drug development, protein design, and drug discovery platforms, and have already secured multiple rounds of funding from leading US dollar institutions. The team's core achievements are Protenix, an open-source replication of AlphaFold 3, and PXDesign, which surpasses AlphaProteo in protein binder design. PXDesign achieved nanomolar binding success rates of 20% to 73% on five out of six different protein targets. However, the logic of AI-driven drug development differs from that of internet businesses. Model predictions must undergo multiple validation processes, including wet experiments, animal experiments, Investigational New Drug (IND) applications, and Business Development (BD), resulting in a very long feedback and monetization chain. This is the underlying motivation driving the scientist team towards asset generation and independent entrepreneurship. (Source: ME)