PAWBench：概率对齐世界建模的现状评估
原标题：Paper page - PAWBench: How Far Are We from Probabilistically Aligned World Modeling?
AI 摘要
该研究正式定义了世界模型的概率对齐概念，并引入PAWBench基准和PAWEval评估协议，用于评估视频生成器作为随机采样器的能力。研究发现，在50个场景和11个当前系统中，没有模型能一致匹配参考概率分布，同时覆盖有效行为范围。研究还测试了语言提示、初始噪声采样和模型训练是否能重塑模型的预测分布，为未来概率对齐世界建模奠定基础。
正文节选
PAWBench: How Far Are We from Probabilistically Aligned World Modeling? Abstract The study formalizes probabilistic alignment for world models, introduces PAWBench and PAWEval to evaluate video generators as stochastic samplers, and finds current models fail to match reference behavior distributions. Recent video generation models are increasingly framed as world models. Many physical processes can unfold in more than one valid way. Therefore, a world model should reproduce not only a plausible