1
芬兰大学说新型AI聊天机器人有助抵抗错误健康信息
芬兰大学最新AI聊天机器人专治健康谣言，用技术对抗虚假信息传播。
芬兰奥卢大学日前发布新闻公报说，该校研究人员与国际同行合作开发出一款人工智能（AI）聊天机器人，通过使用“认知接种”方法，可帮助人们抵抗健康领域的错误信息。
2
Linux Sound Subsystem Also Seeing Many Fixes Driven by AI/LLMs
AI/LLM正深入Linux内核开发，驱动声音子系统大量底层bug修复，技术趋势值得关注。
Article URL: https://www.phoronix.com/news/Linux-7.1-Sound-Many-Fixes Comments URL: https://news.ycombinator.com/item?id=48240454 Points: 10 # Comment…
3
微软承认在 Word、Excel 和 PowerPoint 中加入 Copilot 悬浮按钮是个错误，遭大量投诉后回调
IT之家 5 月 22 日消息，当地时间 5 月 21 日，微软发布了一篇博客，阐述了其为 Word、Excel 和 PowerPoint 引入 Copilot 的美好愿景。 微软官方承认，近期在 Word、Excel 和 PowerPoint 中添加的浮动 Copilot 按钮确实影响了用户的工作…
4
Insights Generator: Systematic Corpus-Level Trace Diagnostics for LLM Agents
系统级诊断LLM代理执行失败，从单条追踪到全语料，规模化发现隐藏错误模式。
arXiv:2605.21347v1 Announce Type: cross Abstract: Diagnosing failures in LLM agents remains largely manual. Practitioners inspect a small subset of ex…
5
When the API literally burned your database after a typo
一个typo就能烧掉数据库？看如何设计能容忍人类愚蠢的staging环境。
The Problem We Were Actually Solving We needed a staging environment that could tolerate the stupidity of humans. Not a toy cluster that looked like p…
6
2.5-D Decomposition for LLM-Based Spatial Construction
用2.5D分解破解LLM空间构建中的坐标错误，神经符号方法让三维结构理解更可靠
arXiv:2605.07066v2 Announce Type: replace Abstract: Autonomous systems that build structures from natural-language instructions need reliable spatial …
7
Incident Report: Railway Blocked by Google Cloud (Resolved)
深入复盘GCP误封账户导致Railway全线瘫痪的真实事故，揭示云服务依赖的致命风险与应急恢复细节。
https://status.railway.com/incident/I23M92U0 Comments URL: https://news.ycombinator.com/item?id=48201484 Points: 512 # Comments: 330
8
KASER: Knowledge-Aligned Student Error Simulator for Open-Ended Coding Tasks
ACL 2026 论文：知识对齐的学生错误模拟器，精准生成开放式编程任务中的学生典型错误，助力智能教育评估
arXiv:2601.06633v2 Announce Type: replace Abstract: Open-ended tasks, such as coding problems that are common in computer science education, provide d…
9
Nuxt 4.2
Nuxt 4.2 官方发布，带来实验性 TypeScript 插件、开发错误处理优化及数据请求 AbortController 支持，提升前端开发体验。
Nuxt 4.2 is out - with experimental TypeScript plugin support, better error handling in development, abort control for data fetching, and more!
10
AI Makes Mistakes; Process Design Matters More
AI犯错不可避免，流程设计比技术本身更关键，值得每位产品经理思考。
Article URL: https://medium.com/@olowu.marydan/ai-makes-mistakes-that-means-process-design-matters-more-than-ever-980097b68f2a Comments URL: https://n…
11
Corti's new Symphony for Speech-to-Text model beats OpenAI at medical terminology accuracy, highlighting the value of specialized AI
专攻医学场景的语音模型精准度碾压通用方案，Corti以1.4%词错误率证明垂直AI的价值。
Today, Copenhagen-based healthcare AI Corti is launching Symphony for Speech-to-Text, a new generation of clinical-grade speech recognition models eng…
12
Weak-to-Strong Elicitation via Mismatched Wrong Drafts
通过注入弱模型的错误草稿，激发强模型在on-policy RL中难以达到的能力，研究视角独特
arXiv:2605.17314v1 Announce Type: cross Abstract: We consider whether off-policy experience from a smaller, weaker model can elicit capability in a st…
13
LoopQ: Quantization for Recursive Transformers
循环语言模型量化面临三大挑战，首次系统性研究揭秘其脆弱性根源
arXiv:2605.16343v1 Announce Type: new Abstract: Looped language models (LoopLMs) improve parameter efficiency by recursively reusing Transformer block…
14
Analyzing Error Propagation in Korean Spoken QA with ASR-LLM Cascades
研究韩语口语问答中ASR与LLM级联系统的错误传播机制，为改进多语言语音交互提供关键参考。
arXiv:2605.17443v1 Announce Type: new Abstract: We analyze how automatic speech recognition (ASR) errors propagate through ASR-LLM cascades in Korean …
15
Take your broken heart and blog
将心碎化作文字，在记录错误中找到治愈
It’s an evening of clear skies, big hearts, and extremely good Internet. I’m listening to Cabin and working on a project that is suspiciously shaped l…
16
A股三大指数集体低开，湖南白银跌超3%
提供的页面仅为网站导航栏，缺少文章正文内容，无法生成推荐语和评分。
36氪获悉，A股三大指数集体低开，沪指低开0.37%，深成指低开0.62%，创业板指低开0.87%；贵金属、稀土、海运板块跌幅居前，招商轮船跌超4%，湖南白银跌超3%，中国稀土跌超2%；HBM、氟化工、存储器概念领涨，中欣氟材涨超9%，雅克科技涨超9%，兆易创新涨超8%。
17
How confessions can keep language models honest
OpenAI秘密武器：让大模型学会“认错”，诚实度飙升的秘密在这篇研究里！
OpenAI researchers are testing “confessions,” a method that trains models to admit when they make mistakes or act undesirably, helping improve AI hone…
18
I broke a working PR because an LLM convinced me there was a bug
程序员被LLM误导，亲手破坏了完美PR，反思人与AI竟有相同“bug”
Article URL: https://www.droppedasbaby.com/posts/2602-02/ Comments URL: https://news.ycombinator.com/item?id=48178732 Points: 1 # Comments: 0
19
Automatic build fix suggestions with Vercel Agent
Vercel Agent 自动诊断构建失败，在 GitHub PR 或 Dashboard 中直接给出代码修复建议，省时省力。
You can now get automatic code-fix suggestions for broken builds from the Vercel Agent, directly in GitHub pull request reviews or in the Vercel Dashb…
20
AI code looks clean. That's the Trap
AI 生成的代码表面光鲜、格式整洁，但隐藏的 off-by-one 和边界处理错误却能让整个系统无声崩溃。别被“干净”的外表骗了，真正的陷阱往往在看似合理的细节里。
Article URL: https://thinkpol.ca/2026/05/15/ai-code-looks-clean-thats-the-trap/ Comments URL: https://news.ycombinator.com/item?id=48161224 Points: 1 …