On June 2nd, Gu Quanquan, a core member of the Science Intelligence (AI4S) team under ByteDance's Seed initiative, announced his departure, concluding his three-year research tenure at the company.

In a farewell message, Gu confirmed the date as his final day at ByteDance and emphasized his intention to continue exploring avenues for scaling technology in the future.

Market sources indicate Gu's next move may involve launching a startup in the fields of AI-driven drug discovery and protein design, with initial backing reportedly secured from a leading U.S. dollar-denominated fund.

Gu's exit and potential venture highlight the organizational constraints faced by major tech firms when pursuing cutting-edge technologies and reflect evolving patterns in talent mobility within the AI sector.

Over the past three years, Gu played a pivotal technical leadership role within the ByteDance Seed team.

His background showcases a blend of deep academic expertise and practical engineering capabilities.

He earned a Bachelor's degree in Automation and a Master's in Control Science and Engineering from Tsinghua University in 2007 and 2010, respectively, followed by a Ph.D. in Computer Science from the University of Illinois Urbana-Champaign in 2014.

Previously, he served as a tenured associate professor of computer science at UCLA.

After joining ByteDance Seed in 2023, Gu primarily led AI drug development efforts.

Key achievements under his leadership include the protein structure prediction model SeedFold, the protein binder design model SeedProteo, and the protein language model DPLM series.

By early 2025, his responsibilities expanded to include foundational infrastructure for general-purpose large models, forming a dedicated internal team to tackle challenges in scalable training of ultra-large parameter models.

Gu's career path clearly illustrates the development trajectory of AI4S within major tech companies: leveraging computing power and algorithms to rapidly establish foundational architectures in vertical domains.

Recently, the AI4S team under ByteDance Seed underwent significant organizational restructuring.

While previous market rumors suggested a potential spin-off, sources close to the company clarified that the team was not split but instead placed under the management oversight of ByteDance's Technology Vice President, Yang Zhenyuan.

This indicates ByteDance maintains a strategic, long-term commitment to frontier AI research, albeit with a more focused approach to business execution and organizational structure.

Such adjustments and the departure of key personnel are, in essence, an inevitable outcome as AI drug discovery evolves from "proof-of-concept" to "pipeline advancement."

Within internet giants' labs, developing open-source models like SeedFold aligns with their strengths; however, truly commercializing AI drug discovery requires tackling substantial real-world challenges such as target discovery, establishing wet lab experimental systems, and advancing clinical pipelines.

These tasks extend beyond the scope of pure algorithm teams, and their lengthy development cycles and high risks deviate from the internet industry's traditional business model emphasizing high frequency, rapid iteration, and standardized ROI.

Consequently, when technological progress reaches an inflection point, it becomes a highly rational business choice for core technical leaders to leave the corporate ecosystem, leveraging their accumulated expertise and capital support to enter more agile, vertical entrepreneurial fields.

Gu's departure serves as a case study in a significant paradigm shift in AI talent mobility, where talent is increasingly flowing from concentration in computing power giants towards dispersion into vertical industries.

During the initial explosion of large language models, the industry's core demand was for "centralization of computing power and data," leading top AI talent to cluster within large firms.

As underlying technologies mature and the competitive focus shifts to vertical domains like embodied AI and scientific AI, which require deep industry-specific know-how, talent is now seeking real-world scenarios where technology can be tangibly applied.

A parallel shift is occurring from algorithmic platform roles to industry-focused entrepreneurship.

Early-career AI scientists within large companies often worked in centralized technical platforms, but the AI4S field is characterized by high levels of non-consensus and industrial complexity.

This makes it difficult for scientists to secure sufficient long-term tolerance for trial and error within the existing internet KPI framework.

Thus, technical leaders are increasingly leaving large companies to become entrepreneurs directly addressing the core pain points of physical industries.

This trend is also being propelled by shifts in the investment landscape.

In the primary market, capital is placing greater emphasis on the commercial conversion rate of AI technology within tangible industries.

Compared to the seemingly bottomless investment required for general-purpose large models, venture capital is reallocating its bets towards niche sectors like AI drug discovery, which feature high industry barriers and clearer commercialization milestones.

The infrastructure-building phase for general-purpose large models is largely complete, and the battle for AI to transform the physical world through industrial breakthroughs is just beginning.

While major tech firms refine their organizational boundaries, the technical leaders entering the entrepreneurial arena must now navigate the complex realities of their chosen industries to achieve the final, complete integration of their technologies.