Reproduce “Can Visual Generation Expand What Unified Multimodal Models Know?” 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
Wenkai Li, Wenyi Hong, Zhuoyi Yang, Jie Tang
The authors test whether facts learned through image generation become available to text-only question answering, using controlled fictional entities across seven categories. They find direct transfer is incomplete: SenseNova-U1 reaches 88% visual accessibility but 0% text accuracy on number questions. Interleaved image–text alignment raises GLM-Omni’s average from 12.57% to 26.86%, but results concern tested models and fictional facts.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.