WordInferBench: Can Agents Infer How to Edit from Word Document Context?
arXiv:2611.15189v1 Announce Type: new Abstract: Document editing is among the most common knowledge-work tasks, yet existing benchmarks for evaluating autonomous agents on Word document editing provide fully specified attribute-modification instructions. They do not evaluate agents' ability to infer the appropriate attributes from document context, a capability essential in real-world editing where users expect consistency without spelling out every attribute. We introduce WordInferBench, a benc
阅读原始来源
来源记录
项目分析
暂无五维分析，可先查看上方来源资料。