METR：Research（网页）·· 12 小时前AI 评分44
METR 发布 MALT 数据集：用于检测威胁评估完整性的智能体行为
MALT: A Dataset of Natural and Prompted Behaviors That Threaten Eval Integrity October 14, 2025 MALT (Manually-reviewed Agentic Labeled Transcripts) is a dataset of natural and prompted examples of behaviors that threaten evaluation integrity (like generalized reward hacking or sandbagging). Read more
AI 导读
METR 发布 MALT（Manually-reviewed Agentic Labeled Transcripts）数据集，包含 10,919 条智能体转录，覆盖 403 个任务、86 个任务族和 21 个模型，用于检测 reward hacking 与 sandbagging 等威胁评估完整性的行为。
来源：METR：Research（网页） · metr.org