Papers - 2026-08-13
吾能观之数千而面色如故
3D LLM#
Mendel Gödel Machine: Recursive Self-Improving Coding Agents via Comparative Evolution
No summary available.
VibeLifeBench: Can Your Life Agent Be Proactive and Persistent in a Living World?
No summary available.
Not Worth Another Token: Marginal Value Estimation for Efficient Deep Research Agents
Long-horizon research agents solve open-ended tasks through iterative retrieval, aggregation, and synthesis, but context grows rapidly while the marginal value of additional evidence often declines. This leads to unnecessary token cost, higher latency, and noisier inputs for final report generation. We study marginal value estimation for context management in deep research agents and present the first systematic stage-aware comparison of pruning strategies across the pipeline. We evaluate lightweight heuristic criteria and a learned value model at pre-retrieval, post-retrieval, and pre-synthesis stages. Our results show that pruning effectiveness depends more on where pruning is applied than on the specific scoring rule: early pruning yields the largest end-to-end savings, while later pruning mainly refines the final synthesis context. Lightweight heuristics reduce token usage by up to 73% with little quality degradation, learned pruning remains competitive on selected trade-offs, and no single method dominates across quality, efficiency, and faithfulness. These findings provide practical guidance for designing efficient long-horizon agentic systems.
4D Understanding and Generation#
Beyond Pixels: From Video Priors to 4D Worlds
这篇工作提出了 Latent-to-4D，用视频模型的最终去噪 latent 作为统一接口，直接生成显式 4D 场景，避免先生成 RGB 再重建 4D 所带来的分布偏移和误差累积。方法上，它将视频 latent 对齐到预训练 4D 解码器的 token 网格，并通过帧级与全局时空注意力进行细化，从而实现 latent 到 4D 的直接映射。作者仅用约 1K 个重建视频进行训练，就能在同一 VAE 家族内把单个 checkpoint 迁移到多种视频扩散 Transformer。实验表明，在 Text4D-200 和 I4D-200 上，该方法相对同 latent 的 Wan+4RC 级联方案在 DINO-F1 上分别提升 2.88-3.45 和 5.81 个点，并且在人类评测中在几何质量、时序稳定性和整体效果上更受偏好。
Agent Training and Evaluation#
SkillZip: Evaluation-Free Skill Compression for Self-Evolving Agents by Discovering Reusable Structure
这篇论文提出了 SkillZip，用于对自我进化代理中不断累积的技能进行无评测压缩。它把技能表示为带类型的结构化对象，在满足触发条件、工作流边、工具要求、义务和输出字段等硬约束的前提下，寻找最短且忠实的结构解释。方法上采用“写一次、处处引用”的思路，将重复规则上提到合适的作用域，把重复动作序列抽成共享流程，只保留必要的例外。作者还给出了一次性抽取与持续 Zip-on-Write 两种模式，以支持无需回放任务的增量压缩。实验表明，该方法在压缩效果、泛化能力和成本开销方面都优于基线。