Reproduce “ThumbnailTruth: A Multi-Modal LLM Approach for Detecting Misleading YouTube Thumbnails Across Diverse Cultural Settings” 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
Wajiha Naveed, Zartash Afzal Uzmi, Zafar Ayyub Qazi
ThumbnailTruth tests whether YouTube thumbnails accurately preview videos by asking language models to compare each image with subtitles and a generated scene summary. On a curated, class-balanced dataset, Claude 3.5 Sonnet reached 93.8% accuracy, while GPT-4o Mini reached 84.8% at its best prompt settings. The dataset is not representative of YouTube, and the study does not establish production performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.