视频选题榜 2026-10-03

  • https://arxiv.org/abs/2610.01428 · Generalization Is Stability, Not Accuracy: Multi-Axis Evaluation of LLMs · 把"泛化"从单点准确率换到多轴稳定性,4 轴独立采集 + 拒绝汇总分数 + 跨数据集变体可反转模型排名 · round=R1