Lisan al Gaib· @scaling01 · X·· 4 小时前AI 评分52
AI 导读
AISecurityInst 发布开源工具 Transect，可将智能体运行轨迹转成交互式时间线，用于观察智能体写代码、跑实验和协作时的过程。该机构指出，单一评测分数会掩盖智能体达成结果的具体路径。转发者表示，这与 LisanBench 让每一步自动可解释的目标一致，使模型在分数相同时仍能按失败模式和行为加以区分。
正文
finally someone did it
this was one of my goals with LisanBench to have each step automatically interpretable, so that models could get the same scores but still be differentiated by failure modes / behavior
As AI agents write code, run experiments, and work together, complex evaluations get harder to interpret. A single eval score obscures the ways agents reached it. Today, we're releasing Transect: an open-source tool that turns agent transcripts into interactive timelines. 🧵在 X 查看被引用的帖子
来源：Lisan al Gaib · x.com