遇见数据集

ScaleAI/mrt

收藏
Hugging Face2026-02-23 更新2025-09-13 收录
官方服务:

资源简介:

Monitor Red Teaming (MRT) 数据集包含AI代理的轨迹和监控结果,用于评估对抗性LLM代理试图规避监督时的监控系统可靠性。数据集分为SHADE-Arena任务和CUA-SHADE-Arena任务,涵盖工具使用和计算机使用场景。数据集旨在促进AI安全领域的研究,特别是自主代理的监控。

The Monitor Red Teaming (MRT) dataset contains agent trajectories and monitoring results for evaluating the reliability of monitoring systems against adversarial LLM agents attempting to evade oversight. It includes SHADE-Arena and CUA-SHADE-Arena tasks covering tool-use and computer-use scenarios. The dataset is designed to facilitate research in AI safety, particularly in the monitoring of autonomous agents.

提供机构:
ScaleAI
二维码
社区交流群
二维码
科研交流群
商业服务