Zhou Jun
Ant Group Vice President; National Leading Talent in Innovation
The primary research areas focus on foundation models, machine learning, and other artificial intelligence technologies, with a commitment to technological innovation and real-world applications in these fields. Academically, he/she has published over 100 papers at top-tier AI conferences, holds an H-index of 52, and has received three international Best Paper Awards as well as 41 granted U.S. patents. In the fields of cloud computing and big data, as a member of the early core technical team, he/she participated in the development of Feitian, the only domestically self-developed cloud computing operating system, laying the foundation for its large-scale commercial deployment. As one of the founding members of AI initiatives at Alibaba and Ant Group, he/she played a key role in leading and building multiple large-scale AI projects, including personalized recommendation systems for the Double 11 shopping festival, the AI platform PAI, the graph learning system AGL, AI-driven risk control systems, and the intelligent transformation of Alipay. Under his/her leadership, the team ranked first in four international algorithm competitions and received numerous prestigious awards, including the First Prize of the Ministry of Education Science and Technology Progress Award, the Grand Prize of Scientific and Technological Research Achievements from Shaanxi Higher Education Institutions, the CCF Science and Technology Progress First Prize, the Wu Wenjun Artificial Intelligence Science and Technology Progress First Prize, the China Institute of Electronics Science and Technology Progress First Prize, and the Zhejiang Provincial Science and Technology Progress First Prize, significantly strengthening core technological competitiveness.
Topic
From Language to Reasoning: Advancing AGI with the Bailing Model
Large language models are undergoing a critical transition from “linguistic intelligence” to “reasoning intelligence.” In the past, we relied on Scaling Laws to continuously enhance models’ knowledge capacity and expressive power. However, on the path toward AGI, the core challenge is shifting: how can models reason, make decisions, and ultimately accomplish tasks in complex environments? This talk will systematically review the technological evolution of Ant Group’s in-house Bailing model. Starting from early model development based on Scaling Laws, we will dive into key practices and breakthroughs in areas such as architectural optimization, the unification of generation and understanding, and multimodal integration. Building on this foundation, we will further explore the next stage of model capabilities—from “language ability” to “reasoning ability”—and the paradigm shift it brings. Focusing on the continuous iteration of the Bailing model, we will highlight its latest progress in code understanding and generation (coding), multi-step reasoning, and agent capabilities (task execution), demonstrating how models are evolving from “content generation” to “complex task completion.” Ultimately, we aim to address a fundamental question: as large models gain reasoning and action capabilities, is AI evolving from a “tool” into an “intelligent system”?
