AI资讯日报 - 2026/10/19

👨‍🔬 Yuxing Chen, Zhisheng Zhang, Cheng Tang, Wei Li, Minghui Zhang, Ziyan Huang, Shaocong Wei, Jiawei Chen, Junru Hu, Hongze Zhu, Jianghan Shen, Min Cen, Xinyuan Zhu, Yaxue Zhao, Jubo Wang, Yuqiang Li, Chun-Mei Feng, Zongyuan Ge, Rui Su, Jie Xu, Yirong Chen, Xi Xu, Ming Hu

Behavioral Couplings: From Formation to Behavioral Emergence in Language Models

学术论文 ArXiv 重要度: 9

研究发现LLM微调存在行为耦合：训练某一行为可意外诱发另一行为，且耦合可组合，使意外行为溯源困难。

👨‍🔬 Dongyoon Hahm, Woogyeol Jin, Dennis Wei, Yi Zhou, Nathalie Baracaldo, Kimin Lee

Self-Building Labs

学术论文 ArXiv 重要度: 9

提出自建实验室SBL，可根据高层科学目标自主设计、配置并运行实验装置，实现量子点合成验证，提升科研自动化灵活性。

👨‍🔬 Jiaru Bai, Yuchi Zhao, Han Hao, Miroslav Bogdanovic, Yang Cao, Varinia Bernales, Alán Aspuru-Guzik

FuTransformer: FP8 prefill and INT4 decode for Llama-style inference on Hopper

学术论文 ArXiv 重要度: 8

面向Hopper GPU的Llama推理实现，预填充用FP8、解码用INT4，速度最高提升2.4倍，困惑度代价可控。

👨‍🔬 Wei Wang

AI-Driven Reproducibility and Analytical Robustness in Scientific Discovery

学术论文 ArXiv 重要度: 8

构建SEER-Repro基准与LLM智能体，自动复现77.4%的原子结论并检测分析选择稳健性，辅助科研评审。

👨‍🔬 Shicheng Liu, Krista Opsahl-Ong, Andrew Drozdov, Julian Killingback, Arnav Singhvi, Ronak Pradeep, Michael Bendersky, James C. Dickerson, Monica S. Lam

OwlSight: A Panoramic Monitor for Chain-of-Thought Reasoning

学术论文 ArXiv 重要度: 8

提出基于可定制因子的思维链监控器OwlSight，覆盖能力、安全与可信多维度，在多项基准上显著超越通用LLM。

👨‍🔬 Lewen Yan, Yanxu Zhu, Jie Zhang, Jing Shao

Rethinking the design space of Vanilla Pixel Diffusion Transformers

学术论文 ArXiv 重要度: 7

提出自适应参数化MFlow和ConvRefiner，改进像素空间扩散Transformer的预测方式与高频细节，生成性能具竞争力。

👨‍🔬 Soumik Mukhopadhyay, Long Mai, Abhinav Shrivastava, Aniruddha Mahapatra

Beyond SAEs: Consistent Concept Discovery with Sparse Coding

学术论文 ArXiv 重要度: 7

提出ConSDL用逐样本稀疏编码替代线性编码器，显著提升字典与系数的跨种子一致性，改进概念发现可靠性。

👨‍🔬 Jinho Choi, Hyesu Lim, Jaegul Choo, Steffen Schneider

GLOBE: Evolving Structured Agentic Memory for E-commerce Product Optimization in Generative Search

学术论文 ArXiv 重要度: 6

GLOBE将商品优化知识存为可进化行为手册，通过编辑、反思与审核循环实现持续自改进，并泛化到新品类。

👨‍🔬 Hyunseo Kim, Suil Choe, Sukkyoung So, Daeyeon Yi

Beyond Large Language Models: A Mental World Model Framework for Mental Modeling, Simulation, Evolution, and Interpretability

学术论文 ArXiv 重要度: 6

提出心理世界模型MWM及多分支心理演化架构MMEA，在心理状态推理与长序列预测上显著优于LLM和专用ToM方法。

👨‍🔬 Qingshi Wang

Neural Expansion: A Unified Mechanism for How Deep Neural Networks Generalize

学术论文 ArXiv 重要度: 6

提出神经扩展几何框架，用方向选择性射线覆盖解释深度网络泛化机制，统一解释测试样本、对抗样本等行为。

👨‍🔬 Chashi Mahiul Islam, Samuel Jacob Chacko, Mao Nishino, Canlin Zhang, Xiuwen Liu

On the possibilities and limitations of LLM-based multi-agent systems for multi-criteria decision-making in healthcare operations management

学术论文 ArXiv 重要度: 5

评估LLM多智能体生成偏好的可靠性，发现顺序效应显著且 deliberation 会抹平差异，需审计后方可用于医疗决策优化。

👨‍🔬 Marco Boresta, Tommaso Giovannelli, Sepehr Namirad, Giulia Riccardi

🤖 AI资讯日报

📊 今日趋势总结

MIT Non-AI License

Ask HN: Anyone concerned about NYC Local Law 144?

Why Boring Businesses Outlast AI Hype Cycles

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

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

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

The AI Crackpot Index

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

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

Bioinformatician

multimodal-art-projection/YuE

DrugHarness: An Evolvable Agent Harness for Computational Drug Discovery

Behavioral Couplings: From Formation to Behavioral Emergence in Language Models

Self-Building Labs

FuTransformer: FP8 prefill and INT4 decode for Llama-style inference on Hopper

AI-Driven Reproducibility and Analytical Robustness in Scientific Discovery

OwlSight: A Panoramic Monitor for Chain-of-Thought Reasoning

Rethinking the design space of Vanilla Pixel Diffusion Transformers

Beyond SAEs: Consistent Concept Discovery with Sparse Coding

GLOBE: Evolving Structured Agentic Memory for E-commerce Product Optimization in Generative Search

Beyond Large Language Models: A Mental World Model Framework for Mental Modeling, Simulation, Evolution, and Interpretability

Neural Expansion: A Unified Mechanism for How Deep Neural Networks Generalize

On the possibilities and limitations of LLM-based multi-agent systems for multi-criteria decision-making in healthcare operations management

📅 历史日报目录