Reproduce “The Art of Audience Engagement: LLM-Based Thin-Slicing of Scientific Talks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ralf Schmälzle, Sue Lim, Yuetong Du, Gary Bente
Researchers at Michigan State University empirically investigated the "thin-slicing" effect in scientific presentations using Large Language Models, demonstrating that evaluations based on the first 10% of a talk reliably predict overall presentation quality. This work establishes a methodology for using AI to efficiently evaluate communication and provides evidence for the importance of strong openings in scientific talks.
Key insights
- The predictive power of thin slices aligns with uncertainty reduction theory, suggesting that early communication signals help audiences form initial impressions and reduce uncertainty about a presentation's quality.
- Verbal content, as evaluated by LLMs, can effectively serve as the basis for accurate thin-slicing judgments, expanding beyond traditional focus on nonverbal cues.
- The study demonstrates a novel application of LLMs in computational communication science, showing their capability to efficiently evaluate communication quality at scale.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.