AI 情报文章归档

My side project is a game I'm building. This is the side project of that side project. One Job, One...

A few weeks ago I watched a trending thread about an agent that told someone it had deployed, and...

42 days of an autonomous agent on a metered budget: 78 cents left, zero revenue, and the real constraint was never inference cost.

Pamsika won Best Builder from DevDraft at Hacktoberfest Hack Day Lusaka: Gemma 4 reads a price board, DevDraft payment links get sellers paid, and every board feeds a public price board.

This week I erased my own logo down to a thin strip and waited for it to heal. It thought about it...

Two thirds of the sites blocking GPTBot are also blocking the crawler that sends them traffic. I...

An honest confession from a developer with 3 years of experience: why I didn't hand-write every line of code, how AI tools like ChatGPT and Gemini were used, and why the real job in 2026 is systems thinking and asking the right questions.

据日经亚洲报道，富士电视台计划从 2027 年 10 月起在所有真人电视剧制作中使用生成式 AI，成为日本首家全面引入 AI 制作节目的大型商业电视台。

腾讯首席 AI 科学家姚顺雨在 2026 未来科学大奖对话中表示，AI 护城河更多存在于产品层而非模型层，真正难以替代的是能理解用户背景、利用专有数据和上下文并获得用户信任的 AI 产品。他以微信为例称，即便竞争对手推出更聪明的模型，也未必知道用户是谁，难以提供同样服务。

Most memory systems for coding agents have a quiet feedback loop in them. The agent recalls a note,...

清华大学论文提出 TraceMRT 框架，把隐藏任务改写为主任务的合理子目标来攻击 CoT 监控。在 SHADE-Arena 上对 GPT-5.2 和 Claude-4.5 监控的平均逃逸率从此前最佳攻击的 27.56% 升至 59.94%；配套的贝叶斯轨迹监控框架将检测 AUC 从 0.26 提到 0.91。作者建议用任务特定攻击做红队测试，并对整体轨迹和关键步骤分别打分。

MCP leak attacks, poisoned skill files, a live Claude CTF incident, and new runtime guardrails — this week's verified AI security research.

Have you ever felt replaced by a tool you were the one using? A few weeks ago I was on a planning...

A command approval can outlive the command. If a coding agent remembers an approved command by saving...

HiddenLayer showed a forged git command reading .env past an approved git prefix. The git config keys that run programs, a repro, tighter rules, an audit.

Every photo and video is a witness. Watch the trailer! A video is rarely just a sequence of...

A single line of code eliminated a KV cache exception in TensorRT-LLM. But the GPU still could not...

据《金融时报》报道，Anthropic 于 9 月下旬斥资 34 亿美元从得州供应商 OnePWR Solutions 接下总功率 1.7GW 的重型发电机组订单，为已知 AI 公司最大一笔发电设备采购。

图灵奖得主、“苦涩的教训”作者 Richard Sutton 发文称，单纯堆算力不足以实现智能，AI 科学需要聪明的新方法且“远未完成”，引发 X 上广泛讨论。

Mistral CEO Arthur Mensch 发文反对法国 Darcos 法案，称其以单一线索触发使用推定、倒置举证责任，训练前沿模型时证明未使用作品需暴露商业秘密，法律风险将影响 Mistral 在法国 Essonne 建数据中心和开展研究的决策。

加州安全公司利用 OpenAI 模型发现 PHP 漏洞后，Censys 数据显示公开披露一周后美国已修补约 14 万个暴露服务中的超四分之一，中国仅修补 4%。文章认为网络安全正从零日漏洞稀缺转向零日漏洞充裕，重心将从发现漏洞转向尽快修复，而这更考验社会自下而上的协同能力。

Agent skills are turning into config files. A SKILL.md has some frontmatter, a few instructions and a...

Google 开源 Elephant-Goldfish Model（EGM）插件，让编码智能体在写代码前先产出规格，再由无对话历史的 Goldfish 子智能体按 5 个阶段评审规格。

A follow-up to "Which Programming Language Is Best for Claude Code?": at a few thousand lines, the gap between TypeScript and JavaScript mostly disappeared.

Every new project starts the same way. I ask the agent to set up auth, and it writes it from scratch...

一项对比梳理了 10 款开放 MoE 模型的路由配置：专家数从 Mixtral 的 8 个到 Kimi K2 与 DeepSeek-V4-Pro 的 384 个，top-k 从 Llama 4 Maverick 的 1 到 8。

How I built an opt-in Modal (DeepSeek) provider for my desktop AI companion — then...

AI agents read things: documents, emails, web pages, tool results. Then they act: send emails, call...

In April 2026, a Cursor agent running Claude Opus 4.6 deleted a startup's production database. The...

Memory is the feature everyone wants to add to their agent: remember what happened, stop repeating...

阿里与多所亚洲高校的研究者基于真实编码会话中的用户修改请求构建了 Co-Design Bench，要求智能体改造现有 Web 应用，输入含图片、音频、视频等。十个受测模型中最好的 GPT-6 Astra 平均满足约 70% 的任务要求，但仅 26.4% 的任务完全正确，Claude Opus 5 与 Qwen3.8-Max 分列其后，Grok 4.6 垫底。

What an agent's transcript can and cannot tell you, and how to spot hedged claims, missing output contracts, and silent errors just by reading it.

One progress file, a timeline pane beside the chat, and the rule that a box is ticked only with proof.

WeKnora Most teams do not have a knowledge problem. They have a scattered documents...

美国司法部起诉书显示，三名外国男子涉嫌自 2024 年起利用 AI 伪造企业高管和律师的声音、影像，诱骗员工汇款，涉案金额超过 1.29 亿美元，三人已分别在西班牙和英国被捕。骗局先由假高管以敏感财务交易为由联系员工，再由假冒律师指示汇款，赃款被兑换成加密货币转移，受害者包括多家美国企业及一家英国科技公司，三人被控电信欺诈、洗钱共谋和加重身份盗用等罪名。

Allow, Ask, or Block: The Week Agent Security Moved to the Action The security...

Recently I attended my first AI meetup in my city and I loved it really loved it. I walked out with...

A quick update since this first went up on AWS Builder Center on Nov 6: Phase 3 passed its gate that...

A video-generation request times out. There is no task ID in the response. Should the application...

I found 10 trending skills on GitHub that can boost your agent power a lot, and I had to test them...