Pengxiang Li is a PhD student at the Hong Kong Polytechnic University and a research assistant whose work bridges GUI agents, multimodal large-language models, and diffusion-based world models. He focuses on cutting-edge research in Large Language Models (LLMs) and video generation. His efforts include developing novel techniques to improve LLM performance and efficiency, such as addressing oversmoothing issues and creating memory-efficient fine-tuning methods. Additionally, Pengxiang has made significant contributions to video generation, notably through the development of a tracklet-conditioned diffusion model. His research also extends to multimodal learning, where he has experience in constructing benchmarks and developing multimodal large-language models (MLLMs) for autonomous driving applications. His collaboration with leading academic institutions and contributions to open-source projects further demonstrate his commitment to advancing the field.