视频选题榜 2026-10-14

• https://arxiv.org/abs/2610.13166 · Scaling Agentic Data for Long-Horizon Terminal Intelligence · 反合成偷懒+阶段化延展·1.5K环境→12/12 benchmark·round=R1