UCLA Professor Quanquan Gu Leaves ByteDance Seed, Led AI Drug Discovery and LLM Pre-Training

Quanquan Gu, a UCLA computer science associate professor and co-head of large model pre-training and scaling at ByteDance's Seed lab, has officially announced his departure on X. Reflecting on his three years at Seed, Gu said he was lucky to work on two major AI frontiers — "curing disease (AI drug discovery)" and "building frontier intelligence (LLM pre-training)" — and signed off with: "The best model is yet to come. Scaling continues."

During his tenure at ByteDance, Gu led the AI for Science (AI4S) drug discovery and scientific computing direction, developing three major biocomputing models. His team built SeedFold, a biomacromolecule structure prediction model that comprehensively surpassed DeepMind's AlphaFold 3 on broad benchmarks; SeedProteo, a protein binder design model that outperformed AlphaProteo, RFdiffusion, Chai-2, BinderCraft, and BoltzGen; and the high-performance protein language model DPLM series. In early 2025, Gu shifted to LLM pre-training, founding an optimization and scaling team that built a highly scalable pre-training tech stack, laying the foundation for Seed 2.0 and subsequent frontier models.

Gu's departure confirms rumors reported by this channel on May 29. Sources close to ByteDance say the AI4S team under Seed is discussing reorganization or a spin-off, and several key members — including Gu, Protenix (an open-source replication of AlphaFold 3) lead and computational biology head Xiao Wenzhi — have left. The departing scientists plan to focus on AI drug discovery, protein design, and drug discovery platforms, backed by top-tier investors. Unlike traditional internet ventures, AI drug discovery requires navigating wet labs and clinical applications, a long cycle that drives scientists to spin out on their own.