AI 情报文章归档

Cyber-Physical Systems are developed through multiple model based activities, comprising several stages of development. As the development progresses through simulation, experimentation, and operation, observations of system execution may …

Large language models (LLMs) have taken the world by storm. Their ability to communicate with us in our own language, and to reason about our world, has made them a favorite tool for writing, answering questions, and much more. Their answe…

Meta 新论文发现，智能体记忆系统对用户价值观与信念类声明应单独提高置信阈值，而非全局统一。这类声明仅占候选记忆的 21.5%，却贡献了超过其他所有类别总和的缺乏支撑的声明，其来源支持率仅 77.9%，其他事实为 96.2%。仅对价值观类收紧标准后，无支撑记忆从 6.2% 降至 4.0%，同时多保留约 13 个百分点的来源关键事实。

On September 4 I started a ComfyUI custom node: a canvas with layers and selection tools, so I could...

Speech recognition on low-cost microcontrollers is, in practice, limited to keyword spotting or a fixed list of commands. We present Oído, an inference engine that runs open-vocabulary English speech recognition entirely on an ESP32-S3 (24…

There is a bug in these six lines. Find it before you scroll. `def get_discount(price, user): ...

The Decoder 报道，OpenAI 解雇了 Tomek Korbak、Jasmine Wang 和 Mikita Balesni 三名与 Hugging Face 事件调查相关的安全研究员，三人在公开信中表示此举令在职员工恐惧，并提出落实外部审计员驻场、保持前沿模型可监控性、明确与外部安全组织合作规则三项要求。

Compression schemes provide a general framework for deriving data-dependent risk bounds for learning algorithms. For learning rules based on preferent compression, existing results upper-bound the risk in terms of a sample-dependent comple…

Have you ever wanted to upload your college notes and simply ask an AI, “Explain Unit 3” or “What is...

Recent accent classification systems can achieve high accuracy in standard in-corpus evaluations. However, when speakers overlap between the training and test sets, such results may overestimate model robustness compared with real-world de…

Collective Autonomous Vehicle (AV) route choices in real-world traffic networks can create externalities for Human-Driven Vehicles (HDVs), as routes that benefit AVs may increase travel times for other road users. To address this, we intro…

Reinforcement learning (RL) has emerged as a key paradigm for improving large language model (LLM) agents, enabling stronger reasoning, planning, and tool-use in interactive environments. However, existing agent RL methods remain fundament…

I built an MCP server so AI agents can find and query our company data. Most of my time did not go...

If you run an LLM-backed endpoint (a support bot, a docs assistant, a "chat with your data" feature),...

作者发布开源工具 bigarrow（big-arrow-on-the-screen），一个 macOS CLI 和智能体技能，让 AI 智能体在屏幕上画出带文字的大箭头、方框，指向需要人工点击的按钮或窗口。

互联网上非人类流量与人类的比例已达 144:1，自主 AI 智能体兼具类人的操作范围和概率性推理，使传统身份验证体系失效。Web Bot Auth 等签名机制只能证明请求者身份，无法验证意图，提示词注入可劫持已验证的智能体。企业需转向执行时持续验证，包括短时凭证、加密入口验证、浏览器层意图分类和提示词注入防护。

At a glance Problem: Let a coding agent run, tap through and profile a React Native app by...

据 Bloomberg 报道，腾讯考虑通过离岸债券融资至多 50 亿美元，债券可能以美元和离岸人民币发行，或于本月初进行，用于加码 AI 与计算基础设施投资。此举 follows 腾讯 6 月发行的 46.6 亿美元债券，是 2020 年以来最大债务交易，资金用于包括 AI 开发在内的一般公司用途。随着 AI 基础设施支出增长，大型科技公司越来越多地在现有现金流之外寻求融资。

I'm Alina, Head of Web Applications at EXANTE. I'm responsible for the technical strategy of our web...

This is a submission for the Kaggle Benchmarking Challenge What I Benchmarked Every...

Three AI API changes from the last two weeks that quietly break shipped code: Claude Haiku 5.5 migration 400s, Managed Agents web_fetch gating, and the GPT-6 plan split.

How I built memory for my desktop AI companion: every memory must quote the...

复刻 Adobe 全家桶的 ArtCraft 开源项目将目标转向 Office，对应项目为 deckcraft（PPT）、wordcraft（Word）、gridcraft（Excel）。其中 WordCraft 是用纯 Rust 重写的开源文字处理器，读写 .docx，可在 macOS、Windows、Linux、BSD 原生运行，也可通过 WebAssembly 在浏览器运行，支持 ribbon、样式、表格、修订、引用和邮件合并，许可证为 MIT / Apache-2.…

The last piece ended with a correction. On the HTTP path of aine-control-plane, the field I had...

Building an AI technical interviewer presents a unique physics problem. When a human engineer pauses...

火山引擎正式发布 AI 安全网关，可通过代理、安全插件或旁路方式部署在企业办公网出口，统一管理企业内 Agent 和 AI 应用的交互流量。该网关提供旁路发现能力，基于 7 层流量镜像识别未装端侧插件的设备，并支持 MCP、Function Calling 等协议特征识别工具调用行为。其 PII 识别率超 95%，高敏感数据可自动路由到 AICC 机密推理环境，产品已开放 POC 测试。

Kevin Roose 与研究者 Jasmine Sun 一起，梳理了 OpenAI、Anthropic、DeepMind 以及那群过度思考并担忧人工智能的人的最新历史。Roose 此前供职于 Hard Fork，并著有《The AGI Chronicles》，现已离开《纽约时报》。

Build an admin table of your blogs, then turn each post into Instagram carousels using Google Nano Banana images via OpenRouter, AI HTML slides, or an AI React code generator.

小红书开源自研 Java 堆外缓存引擎 RED OHC（Apache 2.0），在 AMD EPYC 7T83、JDK 11 环境下对照 Cassandra OHC 的 5 轮基准中，32 线程纯读吞吐达基线 3.38 倍，32 线程读写 9:1 达 2.60 倍，单线程读、写、读写 9:1 分别为 1.13、1.26、1.25 倍。

Build an open-source, OTel-compliant LLM reverse proxy to capture micro-cent token costs and trace inference to Tempo, ClickHouse, or Jaeger

OpenAI 在 X 发文坚持解雇 Jasmine Wang、Tomek Korbak 和 Mikita Balesni 三名安全研究人员，称内部调查发现他们违反了处理敏感信息的明确政策，构成重大信任违规。此声明是对三人周四发布公开信要求提高透明度的直接回应，三人在信中称自己因提出安全担忧而被解雇；OpenAI 表示调查发现的问题超出公开信所述，但未提供细节。

This is a submission for the Kaggle Benchmarking Challenge. What I Benchmarked Hindsight...

Anthropic 的 Claude Science 协调多个 AI 智能体，下载、校准并合并多个太空任务的数据，首次绘制完整紫外线全天图。由于臭氧层阻挡紫外线只能从太空测量，此前 NASA 的 GALEX 任务仅覆盖约三分之二天区并跳过明亮恒星形成区；缺失区域用 inpainting 技术重建，测试中预测与实测平均偏差约百分之十。

⚠️ 观众酌情观看：Microduck 遭受了一次微冒犯

豆包工作宣布任务模式新增画布功能，支持无限画布查看整理素材与创作成果，并可逐步编辑文字、配色与布局。同时上线豆包 2.1 Lite 模型，主打更省额度、更快交付，覆盖文档撰写、表格处理、PPT 制作等场景。此外新增图片模型 Seedream 5.0 Flash，为生图创作提供更多风格选择。

Send the enemy beautiful women to cause discord within his camp. — The 36 Stratagems, The Beauty...

docker-agent comes preinstalled with Docker Desktop 4.63 — declarative YAML agents, MCP toolsets, and a VM sandbox with default-deny egress you have to opt into. A source read.

Someone here (@soumyadeepdey) ran 15 models against a target that was secretly a real company. As the...

OpenAI 研究负责人发布声明，称与 Jasmine、Mikita 和 Tomek 解约是因调查发现他们违反敏感信息处理政策，并否认解雇与提出安全担忧有关，表示未且不会因员工提出担忧而终止雇用。声明还提到正在敲定第三方安全评估机构的合同，细节将在未来几周公布，并认同保持前沿模型可监控性需要全行业承诺。

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass ...