DeepSeek Hires Tsinghua PhD Scholar as V4 Model Launch Nears

Strategic Talent Acquisition

DeepSeek has brought on Gu Yuxian, a PhD candidate at Tsinghua University and one of China's most cited young AI researchers, according to the company's V4 technical paper. The hire comes as the startup expands its algorithm research, engineering, and data teams in preparation for the official release of DeepSeek V4 in mid-July.

Gu received the 2025 Tsinghua Graduate Special Scholarship, the university's highest recognition for doctoral students, along with the 2025 Apple PhD Scholarship and the Ant Group In-Tech Scholarship. He completed both his undergraduate and doctoral work in the Department of Computer Science at Tsinghua, conducting research under Professor Huang Minlie in the Conversational AI group.

His Google Scholar profile lists nearly 5,000 citations. Two of his papers have crossed the 1,000-citation threshold: "Pre-trained Models: Past, Present and Future" and "MiniLLM: Knowledge Distillation of Large Language Models." He has published first-author work at NeurIPS, ICLR, ACL, and other top-tier AI conferences.

Efficiency-Driven Research Profile

Gu's work centers on improving computational efficiency across the full lifecycle of large language models. His research spans three areas: pre-training data selection, where he develops algorithms to optimize training datasets; model compression through knowledge distillation, which transfers capabilities from large models to smaller, deployable versions; and efficient architecture design that reduces computational cost while maintaining performance.

One recent project, Jet-Nemotron, demonstrates this approach. The 2B-parameter hybrid architecture model outperformed Qwen3, Qwen2.5, Gemma3, and Llama3.2 on MMLU and MMLU-Pro benchmarks. On H100 GPUs at 256K context length, it delivered up to 53.6x generation throughput acceleration compared to larger mixture-of-experts models including DeepSeek-V3-Small and Moonlight.

Gu has stated that algorithmic innovation becomes critical when hardware resources are constrained, a philosophy that aligns with DeepSeek's development strategy. The company has emphasized efficiency and cost reduction in its model design, positioning itself as an alternative to compute-intensive approaches favored by larger labs.

Hiring Push Ahead of V4

DeepSeek has been recruiting across multiple functions as it prepares for the V4 launch. The company is filling roles in algorithm research, engineering, product development, operations, and data engineering. Gu's name appears among the authors listed in the DeepSeek V4 technical paper, indicating direct involvement in the model's development.

The V4 release represents a major milestone for the Hangzhou-based startup, which has attracted attention in Chinese AI circles for its focus on open-weight models and efficient training techniques. The company's previous releases, including DeepSeek-V3, have been noted for achieving competitive performance at lower computational cost than comparable models from established players.

Asia's AI Talent Competition

Gu's move to DeepSeek reflects broader dynamics in Asia's AI talent market. Chinese startups are competing aggressively for researchers with expertise in model efficiency, a capability that has become strategically important as compute costs rise and export controls limit access to advanced chips.

Tsinghua University remains a primary source of AI talent in China. The institution's computer science department has produced researchers now working at ByteDance, Alibaba, Baidu, and multiple AI-focused startups. The 2025 Tsinghua Graduate Special Scholarship, which Gu received, is awarded to fewer than 10 doctoral students annually based on academic achievement and research impact.

DeepSeek's ability to attract top-tier academic talent suggests the company has positioned itself as a credible alternative to established tech giants for researchers focused on algorithmic innovation. The V4 launch will test whether that talent translates into market traction as the company scales beyond research releases into production deployments.