​​Speaker: Qiang Zhu​​

Professor and PhD Supervisor, College of Computer Science and Technology, Zhejiang University

•  Laureate of China's National Millions of Talent Project
•  Distinguished Professor of Zhejiang Province
• Recipient of Zhejiang University's "Qiushi Engineering Professorship"
•  Vice Chair, Technical Committee of Human-Machine Integrated Intelligence, Chinese Association for Artificial Intelligence (CAAI)
• R&D and Industry-Academia Collaboration Director, Provincial-Ministerial Collaborative Innovation Center for Artificial Intelligence (Zhejiang University)

​​        Abstract​​

This report traces the technical lineage of DeepSeek while exploring its evolutionary trajectory. Language models, centered on language comprehension and generation, have evolved from traditional statistical approaches to Transformer-based architectures, with the latter revolutionizing language modeling through self-attention mechanisms. Milestones like ChatGPT have propelled generative AI toward practicality via large-scale pretraining and alignment with human feedback.

        Currently, DeepSeek pioneers a transformative wave by leveraging large-scale reinforcement learning to simulate human-like reasoning and decision-making processes. This breakthrough significantly advances the paradigm shift from System 1 (generative models) to System 2 (reasoning models), laying technical foundations for next-generation intelligent agents. Such technological evolution not only drives intelligent upgrades across future industries but will also profoundly reshape human-machine interaction, innovative applications, and societal frameworks.

Chinese website redirect link:https://www.bj-first.org.cn/single?id=338