视频选题榜 2026-09-15
- https://arxiv.org/abs/2609.11115 · Benchmark Radar · A Living Database and Search Engine for AI Benchmarks and Evaluation · 1283 个 benchmark 终于能搜了 · 37 sources / 13 connectors / 24 feeds / 1.29 万数字观察 · cs.AI+cs.IR 交叉列入 · round=R1
- https://arxiv.org/abs/2609.11977 · Occamy-1.0 · Open Pareto-frontier 35B Intelligence for Co-work · Qwen3.6-35B-A3B post-trained · co-work 场景下 cost-efficiency Pareto 拐点 · cs.AI 单分类 · Apache 2.0 模型权重开源 · round=R2
- https://arxiv.org/abs/2609.10895 · ReactHuman · A Physics-Grounded Benchmark for Human-Like Reactive Decision-Making in Embodied Multimodal LLMs · 首个物理可执行的 reactive decision-making 基准 · 17 event families + 1,054 scenes + 240Hz 刚体仿真 · 1/3 危险翻车 + scale 不收敛 · cs.RO+cs.AI 2 类 cross-listed · round=R3