1
Klarna's AI assistant does the work of 700 full-time agents
Klarna用AI助手替代700名全职客服，效率惊人，看AI如何重塑客户服务。
Klarna is using AI to revolutionize personal shopping, customer service, and employee productivity.
2
Introducing the GPT Store
OpenAI正式上线GPT应用商店，用户可发现、分享自定义GPT，AI生态迎来里程碑。
3
Using GPT-4 for content moderation
OpenAI官方分享如何用GPT-4实现内容审核自动化，更一致、更高效、减少人工介入
We use GPT-4 for content policy development and content moderation decisions, enabling more consistent labeling, a faster feedback loop for policy ref…
4
New and improved content moderation tooling
OpenAI推出全新免费内容审核API，更精准高效地过滤不当内容，开发者立即可用。
We are introducing a new and improved content moderation tool. The Moderation endpoint improves upon our previous content filter, and is available for…
5
Efficient training of language models to fill in the middle
OpenAI 详解高效训练语言模型完成中间填充（FIM）的新方法，提升代码补全与文本生成能力
6
Learning to play Minecraft with Video PreTraining
OpenAI 通过视频预训练让神经网络学会玩 Minecraft，仅需少量标注数据就能制作钻石工具，迈向通用计算机智能体。
We trained a neural network to play Minecraft by Video PreTraining (VPT) on a massive unlabeled video dataset of human Minecraft play, while using onl…
7
Proximal Policy Optimization
OpenAI发布新强化学习算法PPO，简单易调优且性能卓越，已成为默认算法。
We’re releasing a new class of reinforcement learning algorithms, Proximal Policy Optimization (PPO), which perform comparably or better than state-of…
8
The History of the Web
讲述web发展史的宝藏newsletter，从Mosaic的img标签到整个网络的演变，令人着迷
I’m not sure if you’ve heard about this but The History of the Web newsletter is simply outstanding and you ought to sign up immediately if publishing…
9
Enhanced and Efficient Reasoning in Large Learning Models
图灵奖得主Valiant提出高效原则性推理方法，破解大模型可信度难题
arXiv:2605.14036v1 Announce Type: new Abstract: In current Large Language Models we can trust the production of smoothly flowing prose on the basis of…
10
ASH: Agents that Self-Hone via Embodied Learning
无需人工标注，ASH从网络视频自我学习具身策略，突破长时域任务瓶颈。
arXiv:2605.14211v1 Announce Type: new Abstract: Long-horizon embodied tasks remain a fundamental challenge in AI, as current methods rely on hand-engi…
11
Measuring Google AI Overviews: Activation, Source Quality, Claim Fidelity, and Publisher Impact
谷歌AI概览（AIO）覆盖20亿用户，首篇系统测评其生成答案的质量、忠实度及对出版商的影响。
arXiv:2605.14021v1 Announce Type: cross Abstract: Google AI Overviews (AIOs) are arguably the most widely encountered deployment of generative AI, rea…
12
CRANE: Constrained Reasoning Injection for Code Agents via Nullspace Editing
提出CRANE方法，通过零空间编辑向代码代理注入约束推理，巧妙平衡指令模型的简洁纪律与思考模型的规划韧性，为智能体模型对齐提供新思路。
arXiv:2605.14084v1 Announce Type: cross Abstract: Code agents must both reason over long-horizon repository state and obey strict tool-use protocols. …
13
Generalized Priority-Aware Shapley Value
突破二元无环限制，提出支持任意有向加权优先级图的广义Shapley值，解决真实数据估值难题。
arXiv:2605.15018v1 Announce Type: cross Abstract: Shapley value and its priority-aware extensions are widely used for valuation in machine learning, b…
14
Quotient-Space Diffusion Models
商空间扩散模型：用商空间思想巧妙处理系统对称性，实现更高效、保对称性的分子3D结构生成。
arXiv:2604.21809v2 Announce Type: replace-cross Abstract: Diffusion-based generative models have reformed generative AI, and also enabled new capabili…
15
Behind the Scenes Hardening Firefox with Claude Mythos Preview
Mozilla利用Claude Mythos预览版深挖Firefox漏洞，AI安全报告从垃圾变利器，颠覆开源安全模式
Behind the Scenes Hardening Firefox with Claude Mythos Preview Fascinating, in-depth details on how Mozilla used their access to the Claude Mythos pre…
16
Tailwind CSS v4.1: Text shadows, masks, and tons more
Tailwind CSS v4.1 终于原生支持文本阴影和遮罩，还有大量新实用工具与变体，大幅提升交互体验。
I wasn't sure it would ever happen but we did it — we released a version of Tailwind CSS that includes text shadow utilities. Tailwind CSS v4.1 is her…
17
Swift at scale: building the TelemetryDeck analytics service
用Swift构建服务百万用户的隐私分析平台，尽显Swift在服务端的大规模应用能力。
TelemetryDeck is an app analytics service specifically for developers, designed to manage usage analytics that are anonymized, privacy-focused, and re…
18
Vercel Flags now supports JSON values
Vercel Flags新增JSON值支持，将多个相关标志合并为一个特性标志，简化A/B测试配置。
You can now store JSON values in Vercel Flags , extending the existing support for boolean, string, and number values. This allows you to collapse wha…
19
A new programming model for durable execution
Vercel Workflows 推出全新持久执行编程模型，零心智负担构建高可靠、可观测的长运行后端，已承载超1亿次运行验证。
The gap between prototypes and production-ready systems is huge. Code that's trivial to run locally falls apart the moment it needs to handle failures…
20
Introducing new token formats and secret scanning
Vercel 自动撤销泄露的 API 凭证，并更新 Token 格式增加前缀标识，提升账户安全。
When Vercel API credentials are accidentally committed to public GitHub repositories, gists and npm packages, Vercel now automatically revokes them to…