Search
On March 26, Junyang Lin, who previously served as the technical lead for Alibaba's QianWen project, published a comprehensive essay following his resignation. In this essay, he argued that the trajectory of AI large-scale model development is currently experiencing a substantial shift, with the primary competitive edge moving from "reasoning-centric thinking" to "agent-centric thinking." Lin revisited the initial wave of reasoning models, exemplified by OpenAI's o1 and DeepSeek-R1, highlighting that this evolution signifies a transition within the industry—from merely scaling up pre-training efforts to broadening the scope of post-training through reinforcement learning. He emphasized that verifiable fields such as mathematics and coding have emerged as crucial benchmarks for assessing the accuracy and reliability of these models.
-
C114 Communication Network -
Communication Home
