World Models Favor Learned Priors over Visual Evidencewith Zhiyong Bao, Xiaotong Liu, Yanhao Chen, Qingqiang Wu1 versionW_49m5fyr3v1 · current