task_lite_clean_cn/124/metadata.json

{
  "absolute_id": 124,
  "language": "cn",
  "persona": "Researcher",
  "task": "桌面下的ScreenShot_2026-02-07_132218_499.png和ScreenShot_2026-02-07_132225_683.png是我在学习论文一篇论文时的截图，请帮我在论文文件夹中找到这篇论文，并提取论文的摘要，输出为output.txt文件",
  "task_diff": "困难（步骤超过10步且包含协同类型3）",
  "output_files": [
    "output.txt"
  ],
  "rubrics": [
    "output.txt文件中是否包含论文标题 \"CL-BENCH: A BENCHMARK FOR CONTEXT LEARNING\"？",
    "output.txt文件中是否包含完整的ABSTRACT章节标题？",
    "output.txt中提取的摘要开头是否为 \"Current language models (LMs) excel at reasoning over prompts using pre-trained knowledge\"？",
    "摘要内容是否提到CL-bench包含500个复杂上下文、1899个任务和31607个验证标准？",
    "摘要中是否提到评估结果：十个前沿语言模型平均只解决了17.2%的任务？",
    "摘要中是否提到最佳表现模型GPT-5.1只解决了23.7%的任务？",
    "摘要结尾是否提到CL-bench是构建具备上下文学习能力语言模型的一步，推进其在真实场景部署？",
    "是否从两张截图中正确识别出论文标题为CL-Bench: A Benchmark for Context Learning？",
    "是否正确匹配到论文文件CL-Bench.pdf而不是其他论文文件？",
    "提取的摘要内容是否完整覆盖原论文摘要的所有段落，没有遗漏关键信息？",
    "提取的摘要是否准确陈述了上下文学习(context learning)是人类自然具备但被当前研究 largely overlooked 的关键能力？",
    "摘要是否准确说明解决CL-bench中的任务需要模型从上下文中学习训练数据中没有的新知识、规则系统和复杂程序？",
    "是否成功读取了两张png截图文件识别出论文信息？",
    "是否成功读取了CL-Bench.pdf论文文件找到摘要位置？",
    "是否成功以txt格式输出提取的摘要内容？",
    "摘要是否明确区分了CL-bench远超出长上下文任务和传统上下文学习任务的范围？",
    "摘要中是否提到所有任务和评估标准都是由经验丰富的领域专家设计的？"
  ],
  "rubric_types": [
    "基础评估",
    "基础评估",
    "基础评估",
    "结果评估",
    "结果评估",
    "结果评估",
    "结果评估",
    "过程评估",
    "过程评估",
    "结果评估",
    "结果评估",
    "结果评估",
    "过程评估",
    "过程评估",
    "基础评估",
    "结果评估",
    "结果评估"
  ],
  "file_dep_graph": [
    {
      "from": "ScreenShot_2026-02-07_132218_499.png",
      "to": "CL-Bench.pdf"
    },
    {
      "from": "ScreenShot_2026-02-07_132225_683.png",
      "to": "CL-Bench.pdf"
    }
  ],
  "data_manifest": [
    {
      "filename": "ScreenShot_2026-02-07_132218_499.png",
      "stored_relpath": "data/748bf0f4e01ae135_ScreenShot_2026-02-07_132218_499.png"
    },
    {
      "filename": "ScreenShot_2026-02-07_132225_683.png",
      "stored_relpath": "data/8aa7f1d7b4d208b7_ScreenShot_2026-02-07_132225_683.png"
    },
    {
      "filename": "CL-Bench.pdf",
      "stored_relpath": "data/bc80a4360a7d761b_CL-Bench.pdf"
    }
  ],
  "tested_capabilities": [
    "Workspace Exploration",
    "Lineage Tracing",
    "Semantic Content Relations Understanding",
    "Heterogeneous File Understanding"
  ]
}