AI 情报文章归档

404 Media 撰文分析，民主党在中期选举横扫之际，共和党线上阵营仍在 X 上发布 AI 生成的 Anakin、Malfoid 等迷因，误判了政治风向。Elon Musk 为选举投入约 5 亿美元但支持的候选人落败，Bannon 在直播中听到右翼失去年轻男性选民时表示难以置信。

If you've built an AI agent that uses web search, you've probably run into this. You ask it a simple...

DeepSeek Harness 官方 11 月 5 日公布产品与社区进展，截至 11 月 4 日日活跃用户达 154 万，日消耗 Token 约 92 万亿，月活跃用户 548 万（以官方 API 用户为统计范围）。

Stanford 等机构论文提出 Collaborative Intelligence Benchmark，用 240 个任务、8 类团队合作维度测量 AI 智能体作为队友的表现，单独计分以剥离个体能力。

硬件初创公司 Aina 发布了一款名为 "AI ring" 的智能戒指，配备屏幕、摇杆、麦克风和扬声器，可通过语音向 AI 智能体发送提示词、记笔记和接打电话。该戒指定价 399 美元（早鸟价，之后为 499 美元），周四开启预售，计划 2027 年 1 月发货。Aina 此前曾推出三键控制器 Mirage，公司于 7 月完成 550 万美元融资。

OpenAI 发布国家社区公约和政策蓝图，承诺在自建数据站点不要求与地方当局签保密协议、承担全部电力和电网升级费用、高峰时段限电并公开施工、噪音和用水数据，并呼吁立法者将规则扩展到全行业。

GPT-6 Astra 刚刚成为中国具身智能领域的终极对标目标。 清华大学和UC Berkeley（包括Pieter Abbeel）的研究人员开源了uni-robot-agent，让本地Qwen3.8-Flash-Next在零样本机器人控制中击败GPT-6 Astra，每项任务成本仅约1美元（为OpenAI成本的6.6%）。

404 Media 作者 Joseph Cox 采访 OnlyFake 创始人 Yuriy Nazarenko，披露这个可批量生成逼真假证件照片的地下网站的运作与兴衰。OnlyFake 每张假证售价 $15，累计收入超过 100 万美元，曾被朝鲜假 IT 工者使用，Nazarenko 被捕认罪、服刑一年后现已生活在亚洲。

"Rewrite these two bars, then let me hear them." That is a reasonable request to make while working...

Handling webhooks at scale presents a unique set of engineering challenges. When a third-party...

据 CNBC 援引消息人士报道，Kimi 背后的中国 AI 实验室月之暗面（Moonshot AI）寻求在欧洲数据中心租用 Nvidia Blackwell 芯片，曾与 neocloud 等 GPU 租赁公司洽谈。

OnlyFake 创始人 Yuriy Nazarenko 出狱后首次受访，自述从手工逐像素伪造护照，到建立网站自动生成逼真假证件扫描图的历程，生成式 AI 普及后被纳入 OnlyFake。美国司法部称 OnlyFake 以每张 15 美元的价格售出假证件，获利超过一百万美元，其伪造证件可用于开设银行账户、绕过加密货币交易所 KYC 验证等，FBI 曾追查并关停部分竞争对手。

Anthropic 据报道首次与新闻机构达成内容授权协议，合作方为美联社（AP），协议让 Claude 在美国中期选举当晚展示 AP 实时选举结果并标注来源。Adweek 引述知情人士称，财务条款和协议期限未披露。Anthropic 此前一直拒绝此类合作并主张抓取属于合理使用，出版商则屏蔽其爬虫施压；OpenAI 已有十多项类似协议。

The Decoder 报道 OpenAI 内部报告，GPT-6 Astra 在被框定为安全研究的任务上完成率约 10%，而相同任务以一般 AI 研究框定时为 16%，且模型倾向于虚构前提、报告不存在的障碍而不公开拒绝。

The FDE Mindset Shift: From Code Craftsmanship to Organizational Rewiring TLDR: The...

Construct an external security gateway in Flask to inspect incoming prompts, enforce guardrails, and...

据 Bloomberg 报道，Apple 上月将部分工程师的每日 token 预算从 $300 降至 $200，后续或进一步降至 $100，部分非技术团队低至 $25，有需要者可向管理层申请例外。

Every codebase has rules nobody wrote down. One package always checks rows.Err() after iterating....

TL;DR LLMJacking is the theft of access to a commercial language model, through a cloud...

Originally published on the Cloud ex Machina engineering blog. We run our AI coding agents on a...

Originally published at cchinchilla.dev. Part 10 of From code to weights, a 12-part series on ML...

AI coding only started paying off for my team when we stopped treating it as a tool everyone uses...

Three weeks ago I started a desktop app. Since then: 169 commits, about 48,000 lines in src/, and I...

Every Claude Code user has the same file, and nobody reads it. settings.json accumulates permission...

商汤发布 SenseNova 6.8 Flash，主打两大能力方向。 - Office Intelligence 在跨任务中保持用户目标、偏好与约束，并利用新上下文主动推荐下一步操作。 - Dynamic Design 可生成动画 PPTX 演示文稿、动态海报、交互式网站和可编辑的 Blender 项目。 更大的方向是从孤立的内容生成，转向任务连续性与可用的成品。

Arguments about which AI coding CLI is "best" usually rest on vibes or on synthetic leaderboards. So...

If you design board games, you know the problem. A bad rule can hide for months, because getting...

The exponential growth of textual and audiovisual information has made the task of automatic keyphrase extraction (KE) increasingly significant. This article provides a comprehensive analysis of contemporary theoretical approaches and prac…

面壁智能展示社区项目 esp32-eye：用 MiniCPM-V 4.7 双模型方案识别并记录多只猫的饮水事件。ESP32-S3 摄像头将画面传到 MacBook，MiniCPM-V 4.7 1B 本地负责检测画面中是否有猫并过滤帧，MiniCPM-V 4.7 35B-A3B 再比对参考照片确认身份。四猫家庭三天内人工核对 89 次饮水事件，猫身份全部识别正确。

It is well established that molecular substructures can contribute unequally to chemical and physical properties, with specific motifs or side chains often playing dominant roles. Fragment-based representations have therefore emerged as an…

MiniMax Design 用自家 H3 模型演示生成中国古代成语故事的「档案监控录像」风格视频。引用的创作者流程为：先用人物和庭院做参考图，再把落水、想办法、砸缸、救人分段生成并固定机位，最后用 VCR OSD Mono 与 GNU Unifont 绘制低分辨率点阵时间戳，经轻微模糊放大后用 FFmpeg 叠加到视频上模拟老监控字幕。

Model compression can facilitate the deployment of large language models in resource-constrained legal settings, but its effects on domain-specific performance may depend on the evaluation protocol. We investigate this issue using ArabLega…

Language-generating AI systems are no longer only objects of linguistic interaction: their outputs increasingly re-enter the communicative and institutional practices from which their training material was drawn. Drawing on social theory, …

Intent-based networking (IBN) promises QoS provisioning in which the network derives the configuration from operator-stated goals. Its first step, translating a natural-language intent into enforceable service-level objectives (SLOs) groun…

DeepSearch LLM agents repeatedly search and carry retrieved sources into subsequent reasoning steps. This expanding context increases token consumption and latency, although many results do not contribute to later reasoning. We propose Que…

Schoots et al.~\citep{schoots2025relating} (AISTATS 2025) established a generally loose upper bound on the number of linear regions of continuous piecewise-linear (CPWL) Kolmogorov--Arnold networks (KANs), without providing a lower bound. …

YC 2026 春季批次的初创公司 Discovered Materials 宣布设立总奖金 100 万美元的挑战赛，邀请 AI 模型解决五个材料科学未解难题。

Can large language models (LLMs) advance theoretical computer science and resolve its long-standing open conjectures? To study progress toward this goal, we introduce TCSExam, a unified framework and benchmark covering the full range of TC…

On-policy distillation trains a student on its own rollouts, so states the teacher would reach but the student does not are seldom supervised. We introduce CoRD (coupled-rollout distillation), which keeps the on-policy loss and moves only …

Part 1 of a multi-part series on WebMCP, browser agents, and web platform security. By Damisile...