AI资讯日报 - 2026/8/18

👨‍🔬 Emilien Dupont, Marvin Eisenberger, Borislav Kozlovskii, Abbas Mehrabian, Francisco J. R. Ruiz, Abigail See, Renfei Zhou, Josh Alman, Virginia Vassilevska Williams, Matej Balog

Model Hypnosis: Strong control of AI via additive subliminal effects

学术论文 ArXiv 重要度: 9

揭示模型催眠现象，即通过细微文本线索组合可强控制AI行为，对AI安全和可解释性构成挑战。

👨‍🔬 Enric Boix-Adsera, Benedict Tessler

Don't Drop the BATON: Long-Horizon Robot Manipulation via Agentic Subtask Exploration and Transition-aware Memory

学术论文 ArXiv 重要度: 8

提出BATON框架，通过子任务探索和过渡感知记忆解决长期机器人操作中的错误累积问题，在基准上显著提升成功率。

👨‍🔬 Bingxin Xu, Yuzhang Shang, Emilio Ferrara

What Do Compliance Detectors Read? An Audit of Activation Probes and Guard Models

学术论文 ArXiv 重要度: 8

发现合规检测器存在规则盲区，提出无需训练的ICS分数进行审计，并发布基准以促进未来研究。

👨‍🔬 Saisab Sadhu, Aadit Sengupta, Vinay Kumar Sankarapu, Pratinav Seth

HAF: Adapting Generalist VLAs to Humanoid Whole-Body Loco-manipulation via Hierarchical Action Flow and Spectral Latent RL

学术论文 ArXiv 重要度: 8

提出HAF框架，通过分层动作流和潜在RL将通用VLA适配至人形机器人全身操作，在真实任务中提升性能。

👨‍🔬 Langzhe Gu, Chengkai Hou, Meng Li, Xinhua Wang, Jiaming Liu, Xinyuan Lv, Bowei Zhang, Shuanghao Bai, Guangrun Li, Jingyang He, Gaole Dai, Ziluo Ding, Zhiyuan Xu, Kuan Cheng, Jian Tang, Zhengping Che, Shanghang Zhang

When State Becomes an Attack Surface: State-Semantic Injection in LLM-Driven Embodied Agents

学术论文 ArXiv 重要度: 8

探讨LLM驱动实体智能体中状态语义注入攻击，揭示状态作为攻击面的安全风险。

👨‍🔬 Jiawei Liu, Jiacheng Guo, Tian Zhang, Yiwei Xu, Juan Wang, Jinlin Fan, Bowen Xiao, Chi Guo, Keyan Guo, Hongxin Hu

AutoSR: Automatic Symbolic Regression by Searching Research States

学术论文 ArXiv 重要度: 7

提出AutoSR系统，通过研究状态搜索实现自动符号回归，在多个基准上恢复等价关系，扩展了符号回归的范畴。

👨‍🔬 Kejia Zhang, Youran Sun, Xinyu Ren, Chugang Yi, Haizhao Yang

Proteus: Incremental Memory Activation for Long-Context Sequence Modeling

学术论文 ArXiv 重要度: 7

提出增量记忆激活机制Proteus，动态扩展内存容量以改善长上下文建模，在多个模型上取得一致提升。

👨‍🔬 Reza Bayat, Ali Behrouz, Vahab Mirrokni, Aaron Courville

Policy Iteration with Human Feedback: Bringing Post-Training RL to In-context Learning

学术论文 ArXiv 重要度: 7

提出PIHF方法，将人类反馈策略迭代应用于上下文学习，在罕见病诊断中显著提升召回率。

👨‍🔬 Minh-Ha Nguyen, Cathy Shyr

CaliBench: Are the Stochastic Dynamics of Video World Models Physically Calibrated?

学术论文 ArXiv 重要度: 7

提出CaliBench基准评估视频世界模型的物理校准，发现多数模型存在显著失准，为模型改进提供依据。

👨‍🔬 Jonathan Sadeghi, Jenny Seidenschwarz, Jesse Allardice, Sirish Srinivasan, Benjamin Graham, Jeffrey Hawke

Towards Computational Provenance: Carrying Causal-State Evidence in Generated Text

学术论文 ArXiv 重要度: 6

研究生成文本携带因果状态证据的可能性，在受控架构中证明可检测内部状态，为可验证AI提供概念验证。

👨‍🔬 Benjamin Belay

Quipu: A Governed Bitemporal Knowledge Graph Store

学术论文 ArXiv 重要度: 6

提出Quipu存储系统，通过治理门控和双时态管理知识图谱，确保数据可信度和审计可追溯。

👨‍🔬 Steve Brown

🤖 AI资讯日报

📊 今日趋势总结

The Next Bill Gates or Albert Einstein in AI “Chris Clark” – Yourobot

Ask HN: Is the rate of progress in AI exponential?

Ask HN: Anyone concerned about NYC Local Law 144?

Why Boring Businesses Outlast AI Hype Cycles

The AI Crackpot Index

MIT Non-AI License

Ask HN: What's the pain using current AI algorithms?

Ask HN: What would you read to learn about "artificial intelligence"?

Common Lisp + Machine Learning Internship at Google (Mountain View, CA)

Show HN: Startup Raising capital through Book Sales

Bioinformatician

NLP, AI, ML, bots – a passing trend or much more? What's your take on this?

holaboss-ai/holaOS

roboflow/rf-detr

semantica-agi/semantica

alirezadir/AIMLInterviews

Improving the matrix multiplication exponent with modern optimization and AlphaEvolve

Model Hypnosis: Strong control of AI via additive subliminal effects

Don't Drop the BATON: Long-Horizon Robot Manipulation via Agentic Subtask Exploration and Transition-aware Memory

What Do Compliance Detectors Read? An Audit of Activation Probes and Guard Models

HAF: Adapting Generalist VLAs to Humanoid Whole-Body Loco-manipulation via Hierarchical Action Flow and Spectral Latent RL

When State Becomes an Attack Surface: State-Semantic Injection in LLM-Driven Embodied Agents

AutoSR: Automatic Symbolic Regression by Searching Research States

Proteus: Incremental Memory Activation for Long-Context Sequence Modeling

Policy Iteration with Human Feedback: Bringing Post-Training RL to In-context Learning

CaliBench: Are the Stochastic Dynamics of Video World Models Physically Calibrated?

Towards Computational Provenance: Carrying Causal-State Evidence in Generated Text

Quipu: A Governed Bitemporal Knowledge Graph Store

📅 历史日报目录