BrowseComp
收藏资源简介:
BrowseComp是由OpenAI创建的一个简单但具有挑战性的基准数据集,用于衡量智能体浏览互联网的能力。该数据集包含1266个需要持续浏览互联网以寻找难以找到的、相互关联的信息的问题。这些问题虽然难度较大,但预测答案简短且易于验证。数据集的创建过程完全由人工完成,问题旨在挑战现有模型的能力。该数据集可应用于评估智能体在浏览互联网时的持久性和创造力,以解决信息检索问题。
BrowseComp is a simple yet challenging benchmark dataset created by OpenAI, designed to evaluate the web browsing capabilities of AI Agents. It contains 1,266 questions that require sustained internet browsing to locate hard-to-find and interconnected information. Although these questions are highly challenging, their corresponding answers are short and easy to verify. The entire creation process of the dataset was fully completed manually, and these questions are intended to challenge the capabilities of existing models. This dataset can be utilized to assess the persistence and creativity of AI Agents when browsing the internet to solve information retrieval tasks.

- 1BrowseComp: A Simple Yet Challenging Benchmark for Browsing AgentsOpenAI · 2025年



