AI资讯日报 - 2026/8/20

👨‍🔬 Bo Liu, Simon Yu, Yiding Jiang, Ao Qu, Andrew Zhao, Zichen Liu, Junsu Kim, Zijian Zhou, Seungone Kim, Tongzheng Ren, Mickel Liu, Hanfei Yu, Zhaorun Chen, Weiyan Shi, Paul Pu Liang, Luke Zettlemoyer, Yejin Choi, Natasha Jaques

Beyond Teacher Likelihood: Group-Calibrated On-Policy Distillation for Long-Context Reasoning

学术论文 ArXiv 重要度: 9

提出GC-OPD方法，通过群组校准蒸馏，解决长上下文推理中教师与学生不一致问题，显著提升Qwen3模型性能。

👨‍🔬 Zhu Zhang, Jixun Wang, Xiaoang Xu, Xiaorong Wang, Zihan Zhou, Zhiyuan Wang, Shuo Wang, Chaojun Xiao, Yuezhi Zhou

ADEPT: Accelerating Dexterity via Pre-Training and Post-Training using Reinforcement Learning

学术论文 ArXiv 重要度: 8

ADEPT框架通过预训练和稳定的后训练，实现高自由度机器人灵巧操作任务的模拟到真实迁移，解决长时程任务。

👨‍🔬 Jayjun Lee, Jessica Yin, Asif Rana, Nicholas Blauch, Sam Mady, Mohak Bhardwaj, Nima Fazeli, Nathan Ratliff, Karl Van Wyk, Ankur Handa

Beyond the Transcript: Detecting Covert Coordination in Latent Multi-Agent Communication

学术论文 ArXiv 重要度: 8

提出VLA框架监测和引导多智能体潜在通信，防止隐蔽协调，在拍卖基准上实现高AUROC和干预效果。

👨‍🔬 Ramneet Kaur, Pradyumna Chari, Ramesh Raskar, Jugad Singh, Sumit Kumar Jha, Anirban Roy

Interpretable AI predicts a 2026 summer dry anomaly in central China

学术论文 ArXiv 重要度: 7

利用深度学习模型预测2026年夏季中国中部干旱异常，并通过可解释性分析支持机制，提供气候预测证据。

👨‍🔬 Anran Wang, Wen Shi, Yong Luo, Jianbin Huang, Lijuan Chen, Junhu Zhao, Weixin Jin, Huihui Yuan

PGFS++: Molecular Property Improvement under Synthesis and Diversity Constraints

学术论文 ArXiv 重要度: 7

PGFS++框架在合成约束下改进分子性质，同时保持输出多样性，避免奖励黑客。

👨‍🔬 Boqiao Zhang, Godbless James, Sai Krishna Gottipati, Andrew Fitzgibbon

Pre-Compiled Pipeline Shards for Distributed LLM Inference on Intel AI PC Fleets

学术论文 ArXiv 重要度: 6

通过流水线分片和预编译，实现多台AI PC协同推理大型语言模型，提升吞吐量并支持70B模型。

👨‍🔬 Tate Berenbaum, Muthaiah Venkatachalam

Leaf Values as Coordinates: Exact Contrastive Explanation for Gradient-Boosted Ensembles

学术论文 ArXiv 重要度: 6

通过叶值坐标化，实现梯度提升树的精确对比解释，构建可验证的补救方法。

👨‍🔬 Emanuele Luzio

Discretizing Continuous Time Series for Imputation with Masked Diffusion Training

学术论文 ArXiv 重要度: 6

提出MDTIM模型，通过随机离散化和掩码扩散训练，提升时间序列插补的鲁棒性和可扩展性。

👨‍🔬 Dongbin Kim, Seungyun Lee, Geonwoo Shin, Jaewook Lee

Grouping the Stochastic Machine: Precision, Not Capability, as the Frontier Metric for AI Systems

学术论文 ArXiv 重要度: 5

提出以精度作为AI系统前沿度量，强调输出一致性，并提供测量方法以指导系统改进。

👨‍🔬 George Andrikopoulos

Tuning the Stochastic Machine: A Systems Engineer's Operating Model for Human-AI Engineering

学术论文 ArXiv 重要度: 5

从系统工程视角，提出人机协作的七项操作纪律，强调错误循环和测量框架。

👨‍🔬 George Andrikopoulos

Finetuning Strategies for Querying Sounds by Vocal Imitation

学术论文 ArXiv 重要度: 4

报告了在AES AIMLA 2025挑战中通过对比学习和联合对比三元组学习微调策略，实现声音查询的冠军方案。

👨‍🔬 Aditya Bhattacharjee, Christos Plachouras, Sungkyun Chang, Emmanouil Benetos

🤖 AI资讯日报

📊 今日趋势总结

The AI Crackpot Index

MIT Non-AI License

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

Why Boring Businesses Outlast AI Hype Cycles

Ask HN: Anyone concerned about NYC Local Law 144?

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

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

Show HN: Startup Raising capital through Book Sales

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

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

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

Bioinformatician

SPADE: Self-Play in Adaptive Synthetic Executable Environments

Beyond Teacher Likelihood: Group-Calibrated On-Policy Distillation for Long-Context Reasoning

ADEPT: Accelerating Dexterity via Pre-Training and Post-Training using Reinforcement Learning

Beyond the Transcript: Detecting Covert Coordination in Latent Multi-Agent Communication

Interpretable AI predicts a 2026 summer dry anomaly in central China

PGFS++: Molecular Property Improvement under Synthesis and Diversity Constraints

Pre-Compiled Pipeline Shards for Distributed LLM Inference on Intel AI PC Fleets

Leaf Values as Coordinates: Exact Contrastive Explanation for Gradient-Boosted Ensembles

Discretizing Continuous Time Series for Imputation with Masked Diffusion Training

Grouping the Stochastic Machine: Precision, Not Capability, as the Frontier Metric for AI Systems

Tuning the Stochastic Machine: A Systems Engineer's Operating Model for Human-AI Engineering

Finetuning Strategies for Querying Sounds by Vocal Imitation

📅 历史日报目录