Reproduce “WinoGAViL: Gamified Association Benchmark to Challenge Vision-and-Language Models” 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
Yonatan Bitton, Nitzan Bitton Guetta, Ron Yosef, Yuval Elovici, Mohit Bansal, Gabriel Stanovsky, Roy Schwartz
WinoGAViL turns image–word association into a game where people choose images for a cue, while an AI opponent and human solvers help curate puzzles that challenge models but remain solvable by people. With 10–12 candidates, ViLT scores 52% mean Jaccard, versus 90% for humans; with 5–6, it scores 55% versus 92%. Dataset comparisons do not isolate why the tasks differ.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.