Research Seminar by Huang Ting-Hao (Kenneth) | Help People Write Papers and Stories in the Era of Large Language Models

Help People Write Papers and Stories in the Era of Large Language Models
 

Writing stands at the center of human communication and creativity, and it is not easy: academics struggle to make their papers clear and effective, while novelists encounter writer's block. This talk covers two lines of research conducted at Penn State's Crowd-AI Lab, focusing on helping people write academic papers, particularly figure captions, and short stories. In our first set of projects, we introduced SciCap, the first large-scale dataset of real-world scientific figure captions from scholarly articles. Following this, we developed methods to automatically generate and assess captions for these scientific figures. The second set of projects explored how creative writers can choose and integrate suggestions from online crowd workers, story plot prediction models, and Large Language Models (LLMs) in realistic writing scenarios. In the era of LLMs, a key question is how we develop systems and technologies that enable humans to communicate and express themselves more easily, effectively, and authentically.