RuleEdit: Failure-Guided Human-AI Model Editing with Prospective Impact Preview 事件

RuleEdit: Failure-Guided Human-AI Model Editing with Prospective Impact Preview arXiv:2606.00011v1 Announce Type: cross Abstract: Despite the promise of AI to assist complex decisions, practitioners still lack ways to detect likely failures and inspect the consequences of model edits before committing them. We present RuleEdit, an interactive, rule-guided human-AI model editing system that (i) surfaces likely failures through interpretable mismatch signals from rule tables and (ii) supports use