论文勘误 & Gaia实验细节讨论

[Rooooyy]

这里的描述非常confusing, GAIA全集是466 questions，但是咱们测的是text-only，应该描述text-only的数量。实验使用的是validation还是test？考虑到可以筛出text-only那默认是validation了? 另，看起来text-only只报了Youtu-Agent的结果，没有和其他模型对比，很难support摘要的这句“state-of-the-art performance on WebWalkerQA (71.47%) and GAIA (72.8%) using open-weight models”

[Lightblues]

Thank you for your feedback. We would like to clarify the details regarding our experimental setup:

1. Text-Only Subset: Our evaluation setup follows the protocols established in WebSailor (arXiv:2507.02592), Cognitive Kernel-Pro (arXiv:2508.00414), etc. The exact sample size is n=103. We will update the paper in the next revision.
2. Performance Scope: Our experiments were conducted in September 2025. The performance comparisons were made against the aforementioned papers and other agent frameworks available at that time.