Primary sourcearXiv cs.AI
View sourceEvent ·Single source
Annotrace 发布:面向 Agent 轨迹的步骤级标注工作台
arXiv 论文 Annotrace 专注于解决 LLM Agent 失败分析中的人工标注难题,支持 OpenTelemetry GenAI、LangSmith、Langfuse 等 6 种轨迹格式,把每次运行转换为有序嵌套的步骤列表。研究者可自定义标注 codebook,由前沿模型基于样本轨迹提出初始标注集,并回答与人类完全相同的问题,形成人机协同标注流程。在多智能体问答语料上,它精确复现了此前人工标注的 235 个标注单元,两个模型对同一批轨迹的标注一致率达 91%,κ=0.82。
Current read
做 Agent 调试的人可以看看,这个工具能读 LangSmith、Langfuse 等六种格式,自动和人工一起做步骤级标注,复现了 235 个标注单元。
2 sources33° AI heatLast updated 2026/10/16 10:55:26
Evidence
2 sourcesRelated source 1AI Engineer
View sourceConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.