TechFeed

安全的代理并不保证互联代理生态系统的安全。Microsoft Research 探讨了 AI 代理交互时出现的问题以及为何网络层面的风险需要新方法。文章《对代理网络进行红队测试：了解 AI 代理大规模交互时会出现哪些问题》…

Safe agents don’t guarantee a safe ecosystem of interconnected agents. Microsoft Research examines what breaks when AI agents interact and why network-level risks require new approaches. The post Red-teaming a network of agents: Understanding what breaks when AI agents interact a…

Cloudflare IPsec 现已通过混合 ML-KEM 正式支持后量子加密。我们已确认其与 Cisco 和 Fortinet 的互操作性。

Cloudflare IPsec now has generally available support for post-quantum encryption via hybrid ML-KEM. We’ve confirmed interoperability with Cisco and Fortinet.

从今天起，代理现在可以成为 Cloudflare 客户。他们可以创建 Cloudflare 账户、开通付费订阅、注册域名，并立即获得用于部署代码的 API 令牌。人类可以介入授予权限，但无需进入仪表板…

Starting today, agents can now be Cloudflare customers. They can create a Cloudflare account, start a paid subscription, register a domain, and get back an API token to deploy code right away. Humans can be in the loop to grant permission, but there’s no need to go to the dashboa…

研究通往 AI 增强护理的路径并开发 AI 共临床医生

Researching the path to AI-augmented care and development of an AI co-clinician.

数据挖掘与建模

Data Mining & Modeling

我们如何复现了三种从 AI 模型中提取私有训练数据的攻击以及阻止它们的密码学防御

How we reproduced three attacks that extract private training data from AI models and the cryptographic defenses that stop them.

Link 为代理提供的钱包让代理能够以编程方式访问 Link，包括生成一次性卡或由钱包中已有的卡片和银行账户支持的共享支付令牌（SPT）的功能。它基于 Stripe 的全新 Issuing for agents 构建。

Link’s wallet for agents gives agents programmatic access to Link, including the ability to generate a one-time-use card or Shared Payment Token (SPT) backed by the cards and bank accounts already in your wallet. It’s built on Stripe’s new Issuing for agents.

我们让 Stripe 更加可编程；凭借 Stripe 网络的强大力量保护并推动您的业务；并为 AI 构建经济基础设施。

We’re making Stripe even more programmable; protecting and propelling your business with the strength of the Stripe network; and building economic infrastructure for AI.

在 2026 年 “What’s Next with AWS” 活动中，AWS 推出了 Amazon Quick——一款具备桌面应用和扩展集成的工作 AI 助手，并将 Amazon Connect 拓展为面向供应链、招聘、客户体验和医疗保健的四种代理式 AI 解决方案。AWS 还进一步扩展了它…

At the "What's Next with AWS" 2026 event, AWS launched Amazon Quick—an AI assistant for work with a desktop app and expanded integrations—and expanded Amazon Connect into four agentic AI solutions for supply chain, hiring, customer experience, and healthcare. AWS also expended it…

Airbnb 如何构建轻量级工作流引擎以解决持久执行问题。作者：Ricardo Gamba、Andriy Sergiyenko 引言：持久执行问题 设想这样一个流程：房东向 Airbnb 提交其房源的保险理赔。系统需要验证…

How Airbnb built a lightweight workflow engine to solve durable execution. By : Ricardo Gamba , Andriy Sergiyenko Introduction: The durable execution problem Picture this hypothetical flow: A host submits an insurance claim about their listing to Airbnb. The system needs to valid…

2026 年第一季度，互联网中断激增，涵盖乌干达和伊朗的全国性断网以及前所未有的针对云基础设施的无人机袭击。我们使用 Cloudflare Radar 探索这些事件背后的数据。

The first quarter of 2026 saw a surge in Internet disruptions, from nationwide shutdowns in Uganda and Iran to unprecedented drone strikes on cloud infrastructure. We explore the data behind these events using Cloudflare Radar.

新框架提供了一种统计方法，用于估计大型语言模型在对抗性对话中发生灾难性故障的概率。

A new framework provides a statistical method for estimating the likelihood of catastrophic failures in large language models in adversarial conversations.

作者：Richard Huang | Machine Learning Engineer II；Yu Liu | Senior Machine Learning Engineer；Ziwei Guo | Senior Machine Learning Engineer；Andy Mao | Staff Machine Learning Engineer；Supeng Ge | Sr. Staff Machine Learning Engineer 引言 在Pinterest，转化广告……

Authors: Richard Huang | Machine Learning Engineer II; Yu Liu | Senior Machine Learning Engineer; Ziwei Guo | Senior Machine Learning Engineer; Andy Mao | Staff Machine Learning Engineer; Supeng Ge | Sr. Staff Machine Learning Engineer Introduction At Pinterest, conversion ads ar…

三月下旬，我前往西雅图参加了 Specialist Tech Conference，这是全球 AWS 专家的最具活力的聚会之一。这是一次与同行交流、分享经验、深入探讨生成式 AI 最新进展的绝佳机会……

Late March took me to Seattle for the Specialist Tech Conference, one of the most energizing gatherings of AWS specialists from around the world. It was an incredible opportunity to connect with peers, exchange experiences, and go deep on the latest advancements in Generative AI…

Google DeepMind 与韩国合作，加速利用前沿 AI 模型实现科学突破

Google DeepMind and Korea partner to accelerate scientific breakthroughs using frontier AI models

LARYBench （Latent Action Representation Yielding Benchmark），一个指引从大规模的视觉数据学习到通用的隐式动作表征的系统化评测基准。实验结果表明：在动作泛化和控制精度上，通用视觉模型的表现均显著优于专门为具身智能设计的动作专家模型，具身动作表征可以从大规模人类视频数据中涌现。

LARYBench (Latent Action Representation Yielding Benchmark) is a systematic evaluation benchmark that guides the learning of universal implicit action representations from large-scale visual data. Experimental results show that in terms of action generalization and control precision, general visual models outperform specialized action expert models designed for embodied intelligence, and embodied action representations can emerge from massive human video data.

通过战略协作编排媒体工作流 作者: Eric Reinecke , Bhanu Srikanth 内容中心媒体制作套件简介 在 Netflix，我们希望为电影制作人提供所需工具，以在全球规模上快速交付内容，快速周转…

Orchestrating Media Workflows Through Strategic Collaboration Authors: Eric Reinecke , Bhanu Srikanth Introduction to Content Hub’s Media Production Suite At Netflix, we want to provide filmmakers with the tools they need to produce content at a global scale, with quick turnaroun…

我们如何使用 Honk、Backstage 和 Fleet Management 来缓解迁移数千个数据集的痛点。文章《后台编码代理：加速下游消费者数据集迁移（Honk，第 4 部分）》首次发表于 Spotify Engineering

How we used Honk, Backstage, and Fleet Management to ease the pain of migrating thousands of datasets. The post Background Coding Agents: Supercharging Downstream Consumer Dataset Migrations (Honk, Part 4) appeared first on Spotify Engineering.

生成式 AI

Generative AI

在真实世界的高风险场景中部署大语言模型（LLMs）比预期更困难。在法律、医学和云事件响应等高风险场景中，性能和可靠性可能迅速恶化，因为将模型适配到特定领域的需求…

Deploying large language models (LLMs) in real-world, high-stakes settings is harder than it should be. In high-stakes settings like law, medicine, and cloud incident response, performance and reliability can quickly break down because adapting models to domain-specific requireme…

Rust Workers 中的 panic 过去是致命的，会污染整个实例。通过在 wasm‑bindgen 项目上游合作，Rust Workers 现在支持弹性关键错误恢复，包括使用 WebAssembly 异常处理的 panic 展开。

Panics in Rust Workers were historically fatal, poisoning the entire instance. By collaborating upstream on the wasm‑bindgen project, Rust Workers now support resilient critical error recovery, including panic unwinding using WebAssembly Exception Handling.

我们如何构建一个每秒摄取 5000 万样本、存储 2.5 PB 逻辑时间序列数据的存储系统。作者：Rishabh Kumar 现代可观测性实践鼓励对每条有意义的代码路径进行仪表化。在过去的 15 年里，开源可观测性 S…

How we built a storage system that ingests 50 million samples per second and stores 2.5 petabytes of logical time series data. By : Rishabh Kumar Modern observability practice encourages instrumenting every meaningful code path. Over the past 15 years, open-source observability S…

生成式 AI

Generative AI

我们从根本上改造了 Facebook 群组搜索，帮助用户更可靠地发现、筛选并验证最相关的社区内容。我们采用了全新的混合检索架构，并实现了自动化的基于模型的评估，以应对 t…

We’ve fundamentally transformed Facebook Groups Search to help people more reliably discover, sort through, and validate community content that’s most relevant to them. We’ve adopted a new hybrid retrieval architecture and implemented automated model-based evaluation to address t…

Google DeepMind 与全球咨询公司合作，将前沿 AI 的力量带给全球各组织。

Google DeepMind partners with global consultancies to bring the power of frontier AI to organizations around the world.

随着 AI 助手和隐私代理挑战传统机器人检测的能力，网络需要新的问责模型。我们认为控制权应当留给客户端，开放的匿名凭证生态系统是保护用户隐私的关键…

As AI assistants and privacy proxies challenge the capabilities of traditional bot detection, the Web needs new models for accountability. We believe that control should remain with the client, and that an open ecosystem of anonymous credentials is key to preserving user privacy…

Doug Burger、可持续发展专家 Amy Luers 和优化研究员 Ishai Menache 探讨数据中心运营、效率提升以及 AI 在电气化、材料和食品系统中的潜力对全球排放的影响。文章《Can we AI our way to a mo…》

Doug Burger, sustainability expert Amy Luers, and optimization researcher Ishai Menache examine the global emissions implications of datacenter operations, efficiency gains, and AI's potential across electrification, materials, and food systems. The post Can we AI our way to a mo…

廖山海 | 内容获取与媒体平台高级软件工程师；阮迪 | 内容获取与媒体平台资深员工软件工程师；李峻 | 内容获取与媒体平台高级工程经理 介绍 准确的内容理解…

Shanhai Liao | Senior Software Engineer, Content Acquisition and Media Platform; Di Ruan, | Senior Staff Software Engineer, Content Acquisition and Media Platform; Evan Li, | Senior Engineering Manager, Content Acquisition and Media Platform Introduction Accurate content understa…

Claude Opus 4.7 在 Amazon Bedrock 上线，具备改进的代理式编码和 100 万 token 上下文窗口。AWS Interconnect 达到正式可用，提供多云私有连接和全新最后一公里选项。此外，Secrets Manager 支持后量子 TLS，新的 C8in/C8ib EC2 实例…

Claude Opus 4.7 arrives in Amazon Bedrock with improved agentic coding and a 1M token context window. AWS Interconnect reaches general availability with multicloud private connectivity and a new last-mile option. Plus, post-quantum TLS for Secrets Manager, new C8in/C8ib EC2 insta…

Agents Week 2026 已圆满结束。让我们回顾一下我们宣布的所有内容，从计算和安全到代理工具箱、平台工具以及新兴的代理网络。我们为代理云交付的全部成果。

Agents Week 2026 is a wrap. Let’s take a look at everything we announced, from compute and security to the agent toolbox, platform tools, and the emerging agentic web. Everything we shipped for the agentic cloud.

我们在同样的产品上构建了内部 AI 工程栈。这意味着有 2000 万请求通过 AI Gateway 路由，处理了 2410 亿 token，推理在 Workers AI 上运行，服务超过 3,683 名内部用户。以下是我们的实现方式。

We built our internal AI engineering stack on the same products we ship. That means 20 million requests routed through AI Gateway, 241 billion tokens processed, and inference running on Workers AI, serving more than 3,683 internal users. Here's how we did it.

了解我们如何使用 OpenCode 构建原生 CI 的 AI 代码审查工具，帮助工程师交付更好、更安全的代码。

Learn about how we built a CI-native AI code reviewer using OpenCode that helps our engineers ship better, safer code.

能不能让 AI 直接学会声音本身的规律，跳过中间环节？为破解这一技术瓶颈，美团 LongCat 团队正式发布 LongCat-AudioDiT。在该模型中，彻底抛弃梅尔谱等中间表示，直接在波形潜空间进行基于扩散模型的文本转语音（Text-to-Speech, TTS），从根源阻断数据转换的级联误差。

Can AI learn the inherent patterns of sound directly, bypassing intermediate steps? To overcome this technical bottleneck, Meituan's LongCat team officially released LongCat-AudioDiT. In this model, mel-spectrograms and other intermediate representations are completely discarded, and diffusion-model-based text-to-speech (TTS) operates directly in the waveform latent space, fundamentally eliminating cascading errors from data conversion.

作者：Brett Axler、Casper Choffat 和 Alo Lowry 自我们首个 Live 节目《Chris Rock: Selective Outrage》以来的三年里，我们见证了直播内容阵容和支撑其运作的直播系统的惊人扩张。从最初仅流播一个节目…

By: Brett Axler , Casper Choffat , and Alo Lowry In the three years since our first Live show, Chris Rock: Selective Outrage , we have witnessed an incredible expansion of our live content slate and the live operations that support it. From modest beginnings of streaming just one…

Agent Readiness 评分可以帮助站点所有者了解其网站对 AI 代理的支持程度。在这里我们将探讨新标准，分享 Radar 数据，并详细说明我们如何让 Cloudflare 的文档成为网络上最友好的代理文档。

The Agent Readiness score can help site owners understand how well their websites support AI agents. Here we explore new standards, share Radar data, and detail how we made Cloudflare’s docs the most agent-friendly on the web.

今天，我们很高兴为您抢先预览我们对共享压缩字典的支持，展示它如何提升页面加载速度，并透露您何时可以自行试用 beta 版。

Today, we’re excited to give you a sneak peek of our support for shared compression dictionaries, show you how it improves page load times, and reveal when you’ll be able to try the beta yourself.

Isabelle/HOL 在表达力、自动化和可扩展性之间的平衡，使得全球首个形式化验证的云级别虚拟机监控器成为可能。

Isabelle/HOL's balance of expressiveness, automation, and scalability enabled the world's first formally verified cloud hypervisor.

作者 Emily Gill 每年，我们将分析工程社区聚集在一起，举办分析峰会——为期多天的内部会议，分享 Netflix 各部门的分析成果，讨论分析实践，并在社区内建立联系。本文是其中之一…

by Emily Gill Each year, we bring the Analytics Engineering community together for an Analytics Summit — a multi-day internal conference to share analytical deliverables across Netflix, discuss analytic practice, and build relationships within the community. This post is one of s…

我们分享 Meta 容量效率计划的洞见，该计划构建了一个 AI 代理平台，帮助自动化定位并修复整个基础设施的性能问题。通过在统一、标准化的工具界面中利用编码的领域专长…

We’re sharing insights into Meta’s Capacity Efficiency Program, where we’ve built an AI agent platform that helps automate finding and fixing performance issues throughout our infrastructure. By leveraging encoded domain expertise across a unified, standardized tool interface the…

我们分享 Meta 在后量子密码（PQC）迁移中的经验教训，帮助其他组织在行业向后量子密码标准转变时提升韧性。我们提出 PQC 迁移等级的概念，以帮助组织内部的团队…

We’re sharing lessons learned from Meta’s post-quantum cryptography (PQC) migration to help other organizations strengthen their resilience as industry transitions to post-quantum cryptography standards. We’re proposing the idea of PQC Migration Levels to help teams within organi…

AWS 在 Amazon Bedrock 上推出 Claude Opus 4.7，Anthropic 最智能的 Opus 模型，可提升编码、长时运行代理和专业工作等方面的性能。Claude Opus 4.7 由 Amazon Bedrock 的下一代推理引擎提供动力，该引擎专为生成式任务而构建…

AWS launches Claude Opus 4.7 in Amazon Bedrock, Anthropic's most intelligent Opus model for advancing performance across coding, long-running agents, and professional work. Claude Opus 4.7 is powered by Amazon Bedrock's next generation inference engine, purpose-built for generati…

生成式 AI

Generative AI

通用科学

General Science

单一、优化后的 LLM 整合了此前需要多个模型的功能，并可作为医药化学家的推理伙伴

A single, optimized LLM unifies what previously required multiple models and can serve as a reasoning partner for medical chemists.

我们的最新音频模型引入了细粒度音频标签，让您能够精确控制 AI 语音，以生成富有表现力的音频。

Our newest audio model introduces granular audio tags that give you precise control to direct AI speech for expressive audio generation.

Vaibhav Shankar; 资深软件工程师 | Raymond Lee; 资深软件工程师 | Chia-Wei Chen; 资深软件工程师 | Shunyao Li; 高级软件工程师 | Yi Li; 资深软件工程师 | Ambud Sharma; 首席工程师 | Saurabh Vishwas Joshi; 首席工程师 | Charles-A…

Vaibhav Shankar; Staff Software Engineer | Raymond Lee; Staff Software Engineer | Chia-Wei Chen; Staff Software Engineer | Shunyao Li; Sr. Software Engineer | Yi Li; Staff Software Engineer | Ambud Sharma; Principal Engineer | Saurabh Vishwas Joshi; Principal Engineer | Charles-A…

使用重要抽样结合微调的捐赠者预填充，以预测训练期间奖励作弊的出现

Using importance sampling with fine-tuned donor prefills to predict reward hacking emergence during training

今天，我们宣布 AWS Interconnect – multicloud 正式上线，这是一项托管的私有连接服务，可将您的 Amazon Virtual Private Cloud（Amazon VPC）直接连接到其他云提供商的 VPC。我们还推出了 AWS Interconnect – last mile，旨在…

Today, we’re announcing the general availability of AWS Interconnect – multicloud, a managed private connectivity service that connects your Amazon Virtual Private Cloud (Amazon VPC) directly to VPCs on other cloud providers. We’re also introducing AWS Interconnect – last mile, a…

了解 Airbnb 如何在构建更紧密社区的同时优先保护用户隐私，赋能客人进行社交互动、自信连接，并掌控个人数据。作者：Joy Jing ✨ 构建更紧密的社区 在 Airbnb，我们的房东和客人形成…

Discover how Airbnb prioritizes user privacy while building a more connected community, empowering guests to engage socially, connect confidently, and maintain control of their personal data. By: Joy Jing ✨ Building a more connected community At Airbnb, our hosts and guests form…

该抗体可开发性基准由约翰斯·霍普金斯惠廷工程学院的 Gray 实验室合作构建，基于公开文献中最丰富的抗体数据集之一，为 AI 引导的抗体设计提供透明的性能评估。

Built in collaboration with the Gray Lab at Johns Hopkins Whiting School of Engineering, the Antibody Developability Benchmark is powered by one of the most diverse antibody datasets in public literature, enabling transparent performance evaluation for AI-guided antibody design.

作者：Matt Lawhon | 高级机器学习工程师；Filip Ryzner | 机器学习工程师 II；Kousik Rajesh | 机器学习工程师 II；Chen Yang | 高级员工机器学习工程师；Saurabh Vishwas Joshi | 首席工程师 在 Pinterest，我们正在扩展推荐模型…

Authors: Matt Lawhon | Sr. Machine Learning Engineer; Filip Ryzner | Machine Learning Engineer II; Kousik Rajesh | Machine Learning Engineer II; Chen Yang | Sr. Staff Machine Learning Engineer; Saurabh Vishwas Joshi | Principal Engineer At Pinterest, scaling our recommendation mo…

教育创新

Education Innovation

摘录 在复杂的、长期运行的智能体系统中，保持智能体之间的一致性和连贯推理需要精心设计。在本系列的第二篇文章中，我们探讨这些挑战以及我们构建的机制，以确保智能体团队能够高效协作……

Excerpt In complex, long-running agentic systems, maintaining alignment and coherent reasoning between agents requires careful design. In this second article of our series, we explore these challenges and the mechanisms we built to keep teams of agents working productively over l…

在我上一次的周度回顾文章中，我提到今年我在与客户进行 AI 驱动开发生命周期（AI-DLC）研讨会时花了大量时间。这些会议的一个共同主题是对成本可视化的需求。团队在快速采用 AI，但在从……

In my last Week in Review post, I mentioned how much time I’ve been spending on AI-Driven Development Lifecycle (AI-DLC) workshops with customers this year. A common theme in those sessions is the need for better cost visibility. Teams are moving fast with AI, but as they go from…

Gemini Robotics ER 1.6：提升空间推理和多视角理解，以支持自主机器人。

Gemini Robotics ER 1.6: Enhancing spatial reasoning and multi-view understanding for autonomous robotics.

作者：Gabriela Alessio、Cameron Taylor 和 Cameron R. Wolfe 引言 当用户登录 Netflix 时，最难的选择之一是观看什么。挑战不在于选项不足——有成千上万的标题——而是找到最吸引人的作品既复杂又深层 …

by Gabriela Alessio , Cameron Taylor , and Cameron R. Wolfe Introduction When members log into Netflix, one of the hardest choices is what to watch. The challenge isn’t a lack of options — there are thousands of titles — but finding the most intriguing one is complex and deeply p…

在 Meta，WebRTC 为各种平台提供实时音视频。但在我们的 monorepo 中对像 WebRTC 这样的大型开源项目进行分叉会带来独特挑战——随着时间推移，内部分叉可能会落后于上游，导致与社区升级脱节。我们…

At Meta, WebRTC powers real-time audio and video across various platforms. But forking a large open-source project like WebRTC within our monorepo presents unique challenges – over time, an internal fork can drift behind upstream, cutting itself off from community upgrades. We’re…

生成式 AI

Generative AI

生成式 AI

Generative AI

随着 AI 提升开发者的速度和生产力，也增加了对安全防护的需求。在本期 Meta Tech Podcast 中，Pascal Hartig 与 Meta 配置团队的 Ishwari 和 Joe 对话，讨论 Meta 如何在大规模下确保配置发布的安全。敬请收听…

As AI increases developer speed and productivity it also increases the need for safeguards. On this episode of the Meta Tech Podcast, Pascal Hartig sits down with Ishwari and Joe from Meta’s Configurations team to discuss how Meta makes config rollouts safe at scale. Listen in to…

亚马逊的 RuleForge 系统利用代理式 AI 生成可投入生产的检测规则，速度比传统方法快 336%。

Amazon’s RuleForge system uses agentic AI to generate production-ready detection rules 336% faster than traditional methods.

作者：Lin Wang（Android 性能工程师） 默认特性 对于移动应用，性能被视为“默认特性”，这意味着应用应运行快速且响应灵敏。就像我们理所当然地期待手表显示时间一样。在 Pinterest 没有例外，w…

Author: Lin Wang (Android Performance Engineer) Default Feature For mobile apps, performance is considered as the “default feature”, which means apps are expected to run fast and be responsive. It’s just as if we expect a watch to show the time. With no exceptions at Pinterest, w…

Amazon S3 Files 将 S3 存储桶转变为在 AWS 计算资源上的高性能文件系统，消除对象存储优势与交互式文件功能之间的权衡，并实现约 1 毫秒延迟的无缝数据共享

Amazon S3 Files makes S3 buckets accessible as high-performance file systems on AWS compute resources, eliminating the tradeoff between object storage benefits and interactive file capabilities while enabling seamless data sharing with ~1ms latencies.

一种经生产验证的方法，将大规模指标管道从 StatsD 迁移到 OpenTelemetry 和 Prometheus。作者：Eugene Ma、Natasha Aleksandrova 在迁移到新监控系统时，您需要提前收集所有指标。这会暴露瓶颈…

A production-tested approach for moving a large-scale metrics pipeline from StatsD to OpenTelemetry and Prometheus. By: Eugene Ma , Natasha Aleksandrova When migrating to a new monitoring system, you’ll want to frontload the work to collect all your metrics. This exposes bottlene…

首页信息流：何佳聪、何大方、程杰（前）、Andreanne Lemay、Mostafa Keikha、Rahul Goutam、Dhruvil Deven Badani、Dylan Wang 内容质量：孙健宁、曾青龙 机器学习服务：李唐 引言 在信息流推荐中，我们为用户推荐一系列项目以 c…

Homefeed: Jiacong He, Dafang He, Jie Cheng (former), Andreanne Lemay, Mostafa Keikha, Rahul Goutam, Dhruvil Deven Badani, Dylan Wang Content Quality: Jianing Sun, Qinglong Zeng ML Serving: Li Tang Introduction In feed recommendation, we recommend a list of items for the user to c…

自动推理如何兼顾安全性、性能和可维护性的需求.

How automated reasoning reconciles the demands of security, performance, and maintainability.

我们分析了超过两万家企业的结账活动，调查了购物者和电商领袖，并收集了 Stripe 网络上企业的洞见，以了解在线转化率的变化.

We analyzed checkout activity across more than 20K businesses, surveyed shoppers and ecommerce leaders, and gathered insights from businesses on the Stripe network to understand what’s changing in online conversion.

在常规的数学解题中，模型只需要“答对最终数值”即可，但数学定理证明不同，它要求极度严苛的逻辑链条，任何一句自然语言的模棱两可，都可能导致整个证明的崩塌。那么，如何让 AI 从“猜答案”走向“严谨证明”，成为复杂推理具有挑战的课题。为了解答这个问题，我们开源了专门用于数学形式化与定理证明的模型 —— LongCat-Flash-Prover。

In typical math problem solving, a model only needs to “get the final answer” right, but mathematical theorem proving is different; it demands an extremely strict logical chain, and any ambiguity in natural language can cause the entire proof to collapse. So, how to move AI from “guessing answers” to “rigorous proofs” becomes a challenging problem of complex reasoning. To address this, we have open-sourced a model specifically for mathematical formalization and theorem proving — LongCat-Flash-Prover.

作者 Ben Sykes 在之前的文章中，我们描述了 Netflix 如何使用 Apache Druid 每秒摄取数百万事件并查询数万亿行数据，提供实时洞察以确保会员获得高质量体验。自那篇文章以来，我们的规模已持续增长…

By Ben Sykes In a previous post , we described how Netflix uses Apache Druid to ingest millions of events per second and query trillions of rows, providing the real-time insights needed to ensure a high-quality experience for our members. Since that post, our scale has grown cons…

AI 代理正在改变我们的构建方式——甚至改变我们对自己作为软件开发者的认知。两者……文章《Let’s Talk Agentic Development: Spotify x Anthropic Live》首次发表于 Spotify Engineering。

AI agents are transforming the way we build — and even how we think of ourselves as software developers. Both... The post Let’s Talk Agentic Development: Spotify x Anthropic Live appeared first on Spotify Engineering.

上周，我和团队拜访了 AWS 香港用户组。香港的社区规模虽小却很强大，活力与热情高涨。他们最近成立了新的 AI 用户组，我们希望有更多人加入。我通过……加强了与社区的联系。

Last week, I visited AWS Hong Kong User Group with my team. Hong Kong has a small but strong community, and their energy and passion are high. They recently started a new AI user group, and we hope more people will join. I was able to strengthen my bond with the community through…

组织级安全防护现已在 Amazon Bedrock Guardrails 中正式上线，可在 AWS 组织内对多个 AWS 账户实现安全控制的集中执行和管理

Organizational safeguards are now generally available in Amazon Bedrock Guardrails, enabling centralized enforcement and management of safety controls across multiple AWS accounts within an AWS Organization.

生成式 AI

Generative AI

Jonathan Woodard 对防守了如指掌。在职业足球担任防守端六年后，Jonathan 明白自己想要一份能够不断面对新挑战、学习新技能并在快节奏环境中持续成长的职业。在发现对…的热情后

Jonathan Woodard knows defense. After playing professional football as a defensive end for six years, Jonathan knew he wanted a career where he could always face new challenges, learn new skills, and continue to thrive in high-paced environments. After discovering a passion for p…

通过将压缩转变为分层自适应流水线并加强监控和控制，我们使 Magic Pocket 对工作负载变化更具弹性

By turning compaction into a layered, adaptive pipeline and strengthening our monitoring and controls, we made Magic Pocket more resilient to workload changes.

Gemma 4：迄今为止我们最智能的开源模型，专为高级推理和代理工作流打造

Gemma 4: Our most intelligent open models to date, purpose-built for advanced reasoning and agentic workflows.

零售商知道搜索和发现已经发生了变化。接下来会怎样尚不明朗。从嵌入式结账到新兴的第三方渠道，以下是电商和 AI 领袖如何整合代理商务的方式

Retailers know search and discovery have already shifted. What comes next is less settled. From embedded checkout to emerging third-party surfaces, here’s how ecommerce and AI leaders are integrating agentic commerce.

LongCat-Next 是我们在通往物理世界 AI 道路上的一次探索。今天，我们把研究思路的核心——LongCat-Next 模型和它的离散分词器全部开源，希望更多开发者能基于它，构建真正能感知、理解并作用于真实世界的 AI。

LongCat-Next is our exploration on the path to AI in the physical world. Today we open source the core of our research approach—the LongCat-Next model and its discrete tokenizer—hoping more developers can build AI that truly perceives, understands, and acts in the real world based on it.

Amazon ECS Managed Daemons 为平台工程师提供对监控、日志和追踪代理的独立控制，无需应用团队协作，确保守护进程的一致部署以及大规模的全面主机级可观测性。

Amazon ECS Managed Daemons gives platform engineers independent control over monitoring, logging, and tracing agents without application team coordination, ensuring consistent daemon deployment and comprehensive host-level observability at scale.

低秩适配、数据增强和链式思考推理是实现无口音多语言输出、提升表达力和可靠合成的技术之一。

Low-rank adaptation, data augmentation, and chain-of-thought reasoning are among the techniques enabling accent-free polyglot outputs, improved expressiveness, and reliable synthesis.