1
峰飞 V2000CG 获印尼型号认可证，成全球首款获海外适航认证 eVTOL
IT之家 6 月 11 日消息，峰飞航空 11 日（今天）通过公众号宣布，V2000CG 凯瑞鸥正式获得印度尼西亚民航局颁发的型号认可证， 成为全球首款获得海外型号认可的 eVTOL 机型 。 根据介绍，作为目前唯一获中国民航局颁发“适航三证”的吨级以上 eVTOL，V2000CG 获得印尼 VTC…
2
Provably Auditable and Safe LLM Agents from Human-Authored Ontologies
基于人类编写本体论，实现LLM代理的可证明安全与可审计性，为智能体可靠性提供新路径。
arXiv:2606.04903v1 Announce Type: cross Abstract: We introduce the LLM agent architecture Agentic Redux, intended for use with nontrivial problem doma…
3
iStat Menus 7.3 更新 – macOS 必备，优雅的系统监控工具
macOS 必装系统监控利器，家庭版每台仅15元还享额外优惠，七年无替代品。
iStat Menus 7.3 更新 iStat Menus 7.3 是自2025 年 11 月 11 日以来（7.2）的一次大更新。新增了以下内容： Apple 在 M5 系列上将原来的 Performance Cores（性能核，P 核）重命名为了 Super Core。而此前的 Efficie
4
Model Collapse as Cultural Evolution
从文化演化理论解释大模型自我训练导致的模型崩溃，提出五个可证伪预测，填补语言学空白。
arXiv:2605.23054v1 Announce Type: cross Abstract: Model collapse, the progressive degradation of LLMs trained on their own outputs, has been character…
5
Provably Protecting Fine-Tuned LLMs from Training Data Extraction while Preserving Utility
新研究用可证明方式保护微调大模型免遭训练数据窃取，同时维持模型效能，隐私与实用兼得。
arXiv:2602.00688v2 Announce Type: replace Abstract: Fine-tuning large language models (LLMs) on sensitive datasets raises privacy concerns, as trainin…
6
Learning to Hand Off: Provably Convergent Workflow Learning under Interface Constraints
首次证明带接口约束的工作流学习能保证收敛，理论突破为自动化流程生成奠基。
arXiv:2605.19140v1 Announce Type: new Abstract: We study workflow learning in a setting where specialized agents hand off control through a shared art…
7
Provable Fairness Repair for Deep Neural Networks
深度神经网络公平性修复新突破，提供可证明保证，已被ASE 2025录用。
arXiv:2605.19549v1 Announce Type: cross Abstract: Deep neural networks (DNNs) are suffering from ethical issues such as individual discrimination. In …
8
MAGIQ: A Post-Quantum Multi-Agentic AI Governance System with Provable Security
后量子时代多智能体AI治理系统，提出可证明安全的MAGIQ架构，解决新兴计算范式下的安全挑战。
arXiv:2605.06933v2 Announce Type: replace Abstract: Our computing ecosystem is being transformed by two emerging paradigms: the increased deployment o…