Junlin Zhang
Chief Scientist and Head of AI R&D Department at Sina Weibo
Junlin Zhang is a director of the China Society for Computational Linguistics and holds a Ph.D. from the Institute of Software, Chinese Academy of Sciences. He currently serves as the Chief Scientist and Head of AI R&D at Sina Weibo. Previously, he was a senior technical expert at Alibaba, leading a new technology team. He is also the author of the technical books *This is a Search Engine: A Detailed Explanation of Core Technologies* and *Big Data Daily Knowledge: Architecture and Algorithms.
Topic
The Future of Deep Thinking Models from the Replication of DeepSeek R1
After DeepSeek R1 was open-sourced, a large number of replication studies have emerged in the academic community, covering the lightweight adaptation of the SFT phase (e.g., S1) and the innovative practice of the RL phase. This sharing will systematically sort out its technical lineage, focusing on the analysis of the two-stage training paradigm: cold-start fine-tuning combined with multi-domain data optimization, and leapfrogging the capability through GRPO reinforcement learning and full-scene alignment. This sharing tries to answer the key technical questions, such as where is the boundary of RL Scaling Law? What are the key factors affecting the effectiveness of SFT stage distillation methods? How to explain the Aha Moment mentioned by DeepSeek, etc.
