SLAI Seminar 40th Session will be discussing the topic on “When Research Enters Reality: Systems, Organizations, and Talent in the AI Industry”,  from 2:30pm-4pm, Juny 17 (Wednesday) at Room B301, online participation is welcome. (Tencent Meeting ID:  453-565-893)

报告主题： 当研究落地现实：人工智能行业的体系、组织与人才 

时间：2026年6月17日（周三）下午2:30-4:00

地点: 深圳河套学院B301教室

线上参与：腾讯会议号 453-565-893

主讲嘉宾：肖桐博士

主持人：欧阳万里教授

讲者简介About the Speaker:

肖桐，资深人工智能与计算机视觉方向研究科学家，深耕大语言模型、多模态系统、AI 智能体领域。

曾任职 Meta，全程参与 Llama 系列大模型研发，负责模型训练底层架构、多阶段微调、多模态能力等核心模块；后加入 AI 初创公司悠途智能（Yutori），投身 AI 网页智能体方向的科研与创业工作，牵头负责浏览器交互模型的强化学习研发。他搭建了 Navi-Bench 评测基准，用于推动模型在真实网页场景下的训练与测评；其团队自研的悠途 n1 模型在效果、性能、成本控制三方面均处于行业顶尖水平，至今稳居 Mind2Web 官方线上榜单第一位。

肖博士本科毕业于清华大学计算机科学与技术系，博士就读于香港中文大学多媒体实验室（MMLab），师从王晓刚教授。

Tong Xiao is a research scientist with long-standing experience in artificial intelligence (AI) and computer vision, focusing on large language models (LLMs), multimodal systems, and AI agents. Previously, he worked at Meta on the research and development of the Llama family of LLMs, where he contributed to core areas including model training infrastructure, multi-stage post-training, and multimodal capabilities. He later joined an AI startup, Yutori, to pursue research and entrepreneurship in AI Web Agents, leading the reinforcement learning (RL) efforts for Browser-Use models. He proposed Navi-Bench to advance the training and evaluation of models on real-world websites. The Yutori n1 model developed by his team is industry-leading across quality, performance, and cost-effectiveness, which remains ranked No. 1 on the official Online Mind2Web leaderboard.

Dr. Xiao received his Ph.D. from MMLab at The Chinese University of Hong Kong, advised by Professor Xiaogang Wang, and his bachelor’s degree from the Department of Computer Science and Technology at Tsinghua University.

报告摘要：

短短数年间，人工智能已从前沿科研构想迅速转化为实实在在的生产力。AI 不再局限于论文与实验室，全面融入产品、企业与各行各业，并开始重塑大众日常生活。本次分享将聚焦一个核心问题：当前沿科研走向产业落地，真正关键的核心要素究竟是什么。

结合本人在 AI 行业的实操经历与行业观察，我将分享认知层面的转变：从最初专注单点技术创新，到逐步意识到决定项目成败的核心往往是体系化能力、规模放大效应、组织协同效率，而更深层的关键，则是人的审美判断力与决策眼光。围绕这一主线，本次演讲将剖析 AI 行业面临的核心挑战、落地方法论以及不断迭代变化的人才需求，并结合智能体强化学习、混合专家模型、缩放定律等技术实例，对比大厂与初创企业两种截然不同的实践模式。

In just a few years, AI has rapidly evolved from a frontier research possibility into a real-world source of productivity. It has moved beyond papers and labs into products, organizations, and industries, and is beginning to reshape people’s everyday lives. This talk focuses on what becomes important once research enters reality.

Drawing on my own experiences and observations in the AI industry, I will share a shift in perspective: from initially focusing on isolated innovations to gradually realizing that what often determines success or failure is system-level capability, scaling effects, organizational coordination, and, more fundamentally, human taste and judgment. Around this central theme, the talk will discuss the key challenges, methodologies, and evolving talent needs in the AI industry, with examples from Agentic RL, MoE, Scaling Laws, and the different practices between large companies and startups.