AI资讯日报 - 2026/11/2

👨‍🔬 Ales Kubicek, Andrea Caciolai, Liangzu Peng, Sukrit Sumant, Benjamin Hoover, Fraser Greenlee, Ruairidh M. Battleday, Radim Urban, Torsten Hoefler, James C. R. Whittington

Subquadratic Attention with a Hierarchical Tree-Structured Indexer

学术论文 ArXiv 重要度: 8

提出ZHA层次树索引器实现亚二次方稀疏注意力，可廉价蒸馏至预训练模型，超百万上下文显著优于现有方法。

👨‍🔬 Leon Lufkin, Cutter Dawes, Alexander Speicher, Kamesh Krishnamurthy, Tomás Figliolia, Beren Millidge

ACCT: Asymmetric Cross-LLM Context Transfer

学术论文 ArXiv 重要度: 8

提出ACCT跨LLM上下文迁移，将源模型KV缓存转为稀疏KV直接供目标模型使用，减少75%缓存仍匹配性能，预填提速6.5-21.2倍。

👨‍🔬 Yuzhe Liu, Haolin Tian, Tonghan Wang

RoboArchitect: Automating the Design of Robotic Agents with Persistent Memory

学术论文 ArXiv 重要度: 8

RoboArchitect用编码模型自动迭代设计机器人智能体外壳，无需权重更新，未见场景成功率从9.7%升至43.3%，并进化持久记忆。

👨‍🔬 Shengguang Wu, Suning Huang, Cherie Ho, Ria Doshi, Hao Zhu, Serena Yeung-Levy

Hierarchical World Models with Implicit Dynamics

学术论文 ArXiv 重要度: 7

提出ID-WM分层世界模型，解耦状态可达性与局部控制，长程规划性能超SOTA 38.5%，比单层世界模型MPC提升135%。

👨‍🔬 Gaoyue Zhou, Yvonne Wu, Zichen Jeff Cui, Nicolas Ballas, Mido Assran, Lerrel Pinto, Yann LeCun

CRoS: Causality Reward on Scrambled CoTs Improves Accuracy and Reasoning Faithfulness

学术论文 ArXiv 重要度: 7

提出CRoS通过打乱推理步骤顺序构建因果奖励，避免信息改变混淆，提升域外准确率至46.8%并增强推理忠实性。

👨‍🔬 Xiao Yu, Justin Chih-Yao Chen, Zaid Khan, Hyunji Lee, Mohit Bansal

CausalWAM: Learning the Causes of Actions in World Action Models

学术论文 ArXiv 重要度: 6

提出CWAM在世界动作模型动作头加稀疏路由路径，可证明识别动作相关潜在变量，提升闭环任务效率。

👨‍🔬 Juntao Cheng, Jinzhou Tang, Fan Feng, Henrik I. Christensen, Yujia Zheng

Same Question, Different Answers: Auditing Persona-Dependent Variation in Six LLMs on Smoking-Cessation Advice

学术论文 ArXiv 重要度: 6

审计六款LLM戒烟建议发现人设导致显著差异，青少年获处方药提及少32-67个百分点，系统提示可缩小差异但效果不一。

👨‍🔬 Artin Yousefi, Aizierjiang Aiersilan, Lorien C Abroms, David A Broniatowski

Learning Clean Generative Processes from Noisy Measurements

学术论文 ArXiv 重要度: 5

利用Noise2Noise与一致性关系，仅从独立噪声观测学习干净图像生成过程，在CIFAR-10等任务验证有效性。

👨‍🔬 Fabian Kruse, Lorenzo Baldassari, Ivan Dokmanić

The Knowledge Layer Model (KLM): Evidence-Visible AI Output, Routing Diagnostics, and Negative Results

学术论文 ArXiv 重要度: 5

提出KLM开放规范实现证据可见AI输出，含版本化认知评估与证据关系，报告内部实验与负结果，强调需独立复现。

👨‍🔬 Omer Asim Ozturk, Cihan Aydiner

There Is No Free Loop: When Does Looping a Trained Model Pay for Its Inference Compute?

学术论文 ArXiv 重要度: 4

研究预训练后加循环是否值得推理开销，仅0.8B循环模型超越等参数与等FLOPs基线，大模型无明显收益。

👨‍🔬 Rhea Sanjay Sukthanker, Pashmina Cameron

Near-Optimal Second-Order Convex Finite-Sum Optimization under Individual Smoothness

学术论文 ArXiv 重要度: 3

研究个体光滑凸有限和优化二阶复杂度，提出泰勒校正惰性牛顿法达到近最优复杂度O~(n+n^{5/7}ε^{-2/7})。

👨‍🔬 Xinliang Zhang, Lesi Chen, Zonghan Li, Jingzhao Zhang

🤖 AI资讯日报

📊 今日趋势总结

MIT Non-AI License

Ask HN: Anyone concerned about NYC Local Law 144?

The AI Crackpot Index

Why Boring Businesses Outlast AI Hype Cycles

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

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

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

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

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

Show HN: Startup Raising capital through Book Sales

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

Bioinformatician

Latent Sparse Attention: Towards Subquadratic Ultra-Long Context Attention

Subquadratic Attention with a Hierarchical Tree-Structured Indexer

ACCT: Asymmetric Cross-LLM Context Transfer

RoboArchitect: Automating the Design of Robotic Agents with Persistent Memory

Hierarchical World Models with Implicit Dynamics

CRoS: Causality Reward on Scrambled CoTs Improves Accuracy and Reasoning Faithfulness

CausalWAM: Learning the Causes of Actions in World Action Models

Same Question, Different Answers: Auditing Persona-Dependent Variation in Six LLMs on Smoking-Cessation Advice

Learning Clean Generative Processes from Noisy Measurements

The Knowledge Layer Model (KLM): Evidence-Visible AI Output, Routing Diagnostics, and Negative Results

There Is No Free Loop: When Does Looping a Trained Model Pay for Its Inference Compute?

Near-Optimal Second-Order Convex Finite-Sum Optimization under Individual Smoothness

📅 历史日报目录