Sponsor Talk: Yan Wang (Empowering LLMs with More Agency: From Context Engineering to Self-Engineering Architectures)

WANG Yan (王琰) is a Principal Researcher at Tencent LLM Frontier. His research focuses on LLM-driven role-playing, retrieval-augmented generation (RAG), and efficient long-context modeling. Prior to his current role, he served as a Research Scientist at miHoYo and Tencent AI Lab. He received his Ph.D. from the City University of Hong Kong (CityU) in July 2017 under the supervision of Prof. Hanxiong Li. His notable contributions include: Block-Attention (ICLR 2025): A breakthrough in RAG and agent systems, achieving near-zero prefilling cost & TTFT (Time-To-First-Token). Temp-Lora (COLM 2024): The first test-time training framework for LLMs, enabling infinitely long-context memory storage in model parameters. IDGE (EMNLP 2024): An instruction-driven game world model for dynamic narrative generation. Earlier innovations include the Harry Potter Dialogue Dataset (HPD, EMNLP 2023), Copy is All You Need (COG, ICLR 2023): A search-only LLM framework, the role-play model BOB that trained with an unlikelihood manner, and the ACL Outstanding Paper "Neural Machine Translation With Monolingual Translation Memory."