1
响应更快：多位大 V 推荐苹果 iOS 27 首个公测版，稳定性满足日常使用
IT之家 7 月 14 日消息，科技媒体 Cult of Mac 今天（7 月 14 日）发布博文，报道称苹果最新发布的 iOS 27 首个公测版已基本完善，其稳定性足以满足日常使用需求。 该媒体测试后认为 iOS 27 首个公测版已趋于完善，各方面的体验都非常稳定，完全满足日常使用需求，不过测试后…
2
Run Tokens Make Email CI Less Fragile
用Run Tokens替代"latest email"，告别CI中邮件测试的随机失败，让自动化更可靠
I keep seeing the same failure pattern in email automation: the test passes locally, gets weird in CI, and then somebody adds a sleep plus one more re…
3
Seismograph – open-source early warning for silent LLM API drift
开源监测大语言模型API无声漂移，提前预警保障输出稳定性。
Article URL: https://github.com/Tania-coder/SEISMOGRAPH Comments URL: https://news.ycombinator.com/item?id=48773957 Points: 1 # Comments: 0
4
Are Claude models broken with the Fable 5 update?
Fable 5更新后，Claude模型所有工具集体失效，赶紧看看你中招没！
Fable 5 is back (for a week) but I just noticed that all the Claude models suddenly are failing with all the Tools ( search, fetch, read, write). Anyo…
5
Core dump epidemiology: fixing an 18-year-old bug
通过大规模Core dump分析，OpenAI工程师揪出潜伏18年的基础设施罕见故障，展现系统性bug排查之道
OpenAI engineers used large-scale core dump analysis to debug rare infrastructure crashes, uncovering both a hardware fault and a long-standing softwa…
6
Claude Command Classifier: availability issues
Claude遭受稳定性危机，命令分类器与Opus模型双双短暂宕机，用户需注意避开高峰期。
Since this morning I am seeing several delays / stability issues with Claude: The command classifier is briefly unavailable. Retrying the PR inspectio…
7
Mechanism-Driven Monitors for Preemptive Detection of LLM Training Instability
提出机制驱动监控器，抢先检测LLM训练不稳定，提升大模型训练可靠性
arXiv:2606.28116v1 Announce Type: new Abstract: Frontier large language model training consumes massive accelerator fleets and long wall-clock computa…
8
Trust Region Masking for Long-Horizon LLM Reinforcement Learning
新方法Trust Region Masking，通过动态约束策略更新提升大模型长序列强化学习稳定性与收敛性。
arXiv:2512.23075v5 Announce Type: replace-cross Abstract: Policy gradient methods for Large Language Models optimize a policy $\pi_\theta$ via a surro…
9
Stabilizing black-box algorithms through task-oriented randomization
通过任务导向的随机化方法，有效提升黑盒算法的稳定性，为机器学习优化提供新思路。
arXiv:2606.25269v1 Announce Type: cross Abstract: As black-box models become foundational to modern research, ensuring their stability is paramount fo…
10
RetailBench: Evaluating Long-Horizon Autonomous Decision-Making and Strategy Stability of LLM Agents in Realistic Retail Environments
新基准RetailBench登场，专门考验大模型智能体在真实零售场景中的长期自主决策与策略稳定性。
arXiv:2603.16453v2 Announce Type: replace Abstract: Large language model (LLM) agents have made rapid progress on short-horizon, well-scoped tasks, ye…
11
The Geometry of Refusal: Linear Instability in Safety-Aligned LLMs
从几何角度揭示安全对齐LLM的拒绝机制实为线性特征，提出对比对数引导方法验证其不稳定性。
arXiv:2606.22686v1 Announce Type: cross Abstract: Modern Large Language Models (LLMs) rely on extensive safety alignment, yet the mechanistic basis of…
12
Home Assistant OS 18.0 发布：优化安装速度、提升系统稳定性
Home Assistant OS 18.0 带来更快的安装体验与更强的稳定性，内核与容器组件同步升级。
IT之家 6 月 23 日消息，科技媒体 Android Authority 昨日（6 月 22 日）发布博文，报道称 Home Assistant OS 18.0 系统上线发布， 重点优化安装速度、首次配置体验和系统底层稳定性。 IT之家注：这里需要区分 Home Assistant OS 和 H…
13
微软 Win11 多个通道更新预览版：提升文件管理器稳定性、修复虚拟机死机故障
微软Win11预览版更新：修复虚拟机蓝屏、增强搜索拼写容错，文件管理器稳定性提升。
IT之家 6 月 20 日消息，微软昨日（6 月 19 日）发布博文，邀请 Windows Insider 项目成员，测试 Windows 11 预览版， 明确更迭迈入 26H2 年度更新版本。 Windows 11 26H2 采用 25H2 相同服务分支，通过启用包（eKB）方式推送，用户只需重启…
14
Acquisition state behaves as a structured, measurable variable governing lung-nodule AI: kernel-driven measurement instability and noise-driven detection fragility, invisible to DICOM metadata
揭示医学影像AI在肺部结节检测中受采集状态影响的内在不稳定性，为AI治理提供定量依据
arXiv:2606.12824v1 Announce Type: cross Abstract: AI governance for medical imaging is formalizing: the 2026 ACR-SIIM Practice Parameter recommends lo…
15
Characterizing Software Aging in GPU-Based LLM Serving Systems
揭秘GPU上LLM服务系统的软件老化特征，为提升系统长期稳定性提供关键洞察。
arXiv:2606.11916v1 Announce Type: cross Abstract: This paper proposes an empirical methodology to study software aging in GPU-based LLM serving system…
16
We Used Agentic AI to Fix Kong Gateway's Flakiest Tests
用智能体AI自动化修复Kong Gateway最难搞定的不稳定测试，工程实践干货。
Article URL: https://konghq.com/blog/engineering/how-we-used-agentic-ai-to-fix-kong-gateways-flakiest-tests Comments URL: https://news.ycombinator.com…
17
AdaGC: Enhancing LLM Pretraining Stability via Adaptive Gradient Clipping
被ICML接收的自适应梯度裁剪方法，有效提升LLM预训练稳定性，AI训练优化的新突破
arXiv:2502.11034v3 Announce Type: replace Abstract: Loss spikes remain a persistent obstacle in large-scale language model pretraining. While previous…
18
Show HN: I applied Lyapunov stability theory to detect when LLM agents spiral
将控制论的李雅普诺夫稳定性应用于LLM代理，提前9步预判失控螺旋，成本节省10倍。
Article URL: https://github.com/vishal-dehurdle/state-harness Comments URL: https://news.ycombinator.com/item?id=48466381 Points: 1 # Comments: 0
19
Geometrically Averaged Hard Target Updates for Linear Q-Learning
提出几何平均硬目标更新方法，有效提升线性Q-learning的稳定性与性能。
arXiv:2606.10835v1 Announce Type: cross Abstract: Periodic hard target updates are among the most common stabilization devices in modern deep Q-learni…
20
Stabilizing On-Policy Distillation for MLLM Reasoning with Global Normalization
用全局归一化稳定多模态大模型基于策略的蒸馏，提升推理性能与训练效率的创新方法。
arXiv:2606.09091v1 Announce Type: new Abstract: On-policy distillation (OPD) has recently emerged as an important post-training paradigm. By using a s…