AI 情报文章归档
浏览 AI圈报 已保存的 37974 条 AI 情报文章,覆盖模型、产品、行业、论文、教程与观点方法六大分类。
已收录文章37974
正式分类6
精选内容4748
全部文章
第 127 / 950 页 · 每页 40 条
教程 / 实战普通
Cloudflare's KV Cache Exposure Leaked Other Customers' Prompts for Two Weeks
Cloudflare's KV cache exposure on Workers AI let 155 responses leak other customers' context for two weeks, caused by a pool size mismatch and an SGLang bug.
信息来源:DEV Community
行业动态普通
MiniCPM-V 4.7 1B iPhone 本地 OCR 实测
真实场景 OCR,装进口袋!📱🔥 很高兴看到 MiniCPM-V 4.7 1B 在 iPhone 上本地运行,准确读出营养成分标签。 非常感谢测试并分享这段并排对比演示!💙🚀
信息来源:X:面壁智能 OpenBMB (@OpenBMB)
观点 / 方法普通
MiniCPM-V 4.7 iPhone 推理提速至 20.51 tokens/s
很喜欢这份拆解!🔥 在 iPhone 上从 3.38 提升到 20.51 tokens/s——这是推理优化如何释放强大端侧 AI 的一个绝佳例子。 感谢重点介绍 MiniCPM-V 4.7 背后的工程工作!🚀💙
信息来源:X:面壁智能 OpenBMB (@OpenBMB)
教程 / 实战普通
I Don't Write Code Anymore. I Still Read It.
Agents write all my code now. I still keep asking them to simplify it, because unnecessary complexity creates more to reason about and more places for bugs.
信息来源:DEV Community
教程 / 实战普通
Requiem: a crash post-mortem that speaks as the program that died
This is a submission for the Hacktoberfest Open Source AI Challenge: Week 4 - Ghost in the...
信息来源:DEV Community
论文研究精选
Ai2 发布 Old Ideas, Novel Problems 研究与 benchmark,揭示 LLM 新颖性裁判不稳定
Ai2 发布研究 Old Ideas, Novel Problems、配套 benchmark、数据、代码和自动化流水线,测试 LLM 作为研究想法新颖性裁判的可靠性。
信息来源:Ai2 / Allen Institute for AI(RSS)
教程 / 实战普通
My calibration study found the GEO score doesn't predict AI citations. Here's the data
I built a tool that scores websites 0 to 100 on how visible they are to AI answer engines. Then I ran...
信息来源:DEV Community
教程 / 实战普通
We sent 23 AI agents to our website. Most looked human
AI assistants now open websites for people. You ask ChatGPT to compare two pricing pages, and a...
信息来源:DEV Community · ChatGPT
教程 / 实战普通
How I Stopped an LLM From Skipping the Investigation, Looping Forever, and Inventing Pods — Without LangChain
An agent loop is just a for loop. The hard part is the invariants inside it — and the deterministic guardrails that catch every predictable failure mode.
信息来源:DEV Community
观点 / 方法普通
高盛:半导体板块对资产价格的影响力已超过油价
据《商业内幕》报道,高盛研究显示,过去三周内半导体板块行情差异可解释各类资产约 54% 的表现分化,油价仅能解释 2%。Anthropic 提交 IPO 申请后 AI 股热情提升,纳指和标普 500 涨约 1%,英伟达涨 2%,费城半导体指数涨约 3%;KOSPI 盘中一度涨 5%,较 6 月高点已下跌约 40%。
信息来源:IT之家·人工智能 · Claude / NVIDIA
教程 / 实战普通
How to Audit Whether Your Website Is Ready for AI Search Citations
A B2B page can rank in Google and still fail to appear when a buyer asks an AI assistant for an...
信息来源:DEV Community
行业动态精选
Latent Space 解读 Anthropic 提交 S-1:拟以代码 AN 上市纳斯达克,H1 2026 营收 16.3B 美元
Latent Space 汇总解读 Anthropic 公开提交的 S-1 文件,公司计划以代码 AN 上市纳斯达克,营收从 2023 年的 41M 美元增至 H1 2026 的 16.3B 美元,Q2 营收 11.6B 美元同比增长 1,368%,营业利润 974M 美元,H1 2026 毛利率达 85%。
信息来源:Latent Space(RSS) · Claude
教程 / 实战普通
Your AI had a $500 refund cap. It issued five of them in one conversation.
An HVAC company in the Midwest gave their AI agent a refund tool in February with one guardrail:...
信息来源:DEV Community
行业动态普通
英伟达向 Meta 投资 30 亿美元并扩大 Muse 智能助手合作
英伟达 10 月 27 日宣布与 Meta 扩大 Muse 智能助手方向的合作,并向 Meta 股权投资发行投资 30 亿美元(约合 200.51 亿元人民币)。
信息来源:IT之家·人工智能 · NVIDIA
教程 / 实战普通
Prompt Injection Defense for Production Agents: What Actually Works
Prompt injection cannot be reliably filtered, and any defense built on detecting malicious text will eventually lose. What works in production is...
信息来源:DEV Community
教程 / 实战普通
Chitthi: A Séance Where the Ghost Can Only Quote Itself
This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine ...
信息来源:DEV Community
教程 / 实战普通
back from the dead
Possessing fortune-mod: Bringing an abandoned UNIX oracle back from the dead For Hacktober...
信息来源:DEV Community
教程 / 实战普通
Last week in Agent Security Update 4: Cairn’t Touch This
What a fucking week.... If you've been living under a rock this week, two critical RCE...
信息来源:DEV Community
行业动态精选
OpenAI 扩大与 Oracle 合作,将 Oracle 应用与数据接入 ChatGPT
OpenAI 宣布扩大与 Oracle 的合作,将 Oracle 应用和数据接入 ChatGPT,并用 OpenAI 模型驱动 Oracle 新产品。
信息来源:OpenAI:官网动态(RSS · 排除企业/客户案例) · OpenAI / ChatGPT
教程 / 实战普通
Architectural Deep-Dive: Retriever-Free Retrieval: Navigating Documents with Decision Models
Replacing Vector Databases with Decision-Based Tree Navigation in Production RAG For the...
信息来源:DEV Community
行业动态普通
Elon 提议前沿 AI 公司互测新模型保安全
Elon 解释 SpaceXAI 向 Anthropic 提供算力前,曾亲自接触 Anthropic 团队考察其"邪恶探测器",因对方友善负责而决定洽谈合作。他还提议前沿模型公司每周通话、新模型发布前一周先给竞争对手试用,认为监管机构建立和立规太慢,竞争对手互盯更能保持诚实。
信息来源:X:indigo (@indigox) · Claude
教程 / 实战普通
To hint or not to hint
We run a practice platform for coding interviews with an AI assistant in the editor. That's the...
信息来源:DEV Community
教程 / 实战普通
ContextForge: Resurrecting Codebase Blueprints to Slash LLM Token Waste by 90%
This is a submission for the Hacktoberfest Open-Source AI Challenge: Week 4 — Ghost in the...
信息来源:DEV Community
教程 / 实战普通
Older Card: the later story is louder, and the line is thrown away
Gemma 3 1B may write a remembrance line. If it borrows the later story, or adds a word the older card never used, the line is thrown away.
信息来源:DEV Community
教程 / 实战普通
Wayback Séance, I asked dead websites to speak and made them cite their sources
What I Built Wayback Séance lets you talk to a website that died. You type an address such as...
信息来源:DEV Community
教程 / 实战普通
How to Turn Ollama Into an AI Agent That Can Use Tools (Free and Local)
Ollama runs AI models on your computer, but on its own it can only chat. Here's how to turn it into a local AI agent that can search the web, work with your files, and run tasks, with no code and no API bills.
信息来源:DEV Community · Ollama
观点 / 方法普通
Addy Osmani:何时该把子智能体收回主循环
Addy Osmani 提出判断子智能体的经验法则:只有当任务可并行、需要全新上下文或单个智能体确实无法选对工具时才保留子智能体,否则应将其收回主循环。他指出系统提示词臃肿往往是问题根源,可用 Skills 和工具搜索解决,而不必把上下文拆分到多个智能体。做法是把子智能体的提示词改写成 skill,跑完评测后保留效果更好的方案。
信息来源:X:Addy Osmani (@addyosmani)
教程 / 实战普通
The Last Directory: a local ghost story with a paper trail
This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine. ...
信息来源:DEV Community
教程 / 实战普通
Running a Green CI Pipeline Isn't the Hard Part Anymore
Six months ago, your team adopted an AI coding assistant. Pull requests doubled. Generated code...
信息来源:DEV Community
教程 / 实战普通
把15个国民级App接入蚂蚁灵光2.0:一份Personal Agent实操教程
作者分享了将米家、滴滴、美团、瑞幸、麦当劳、淘宝等15个国民级App接入蚂蚁灵光2.0 Personal Agent的实操过程,涵盖扫码授权、官方MCP、Skill、CLI等接入方式,并附上一段可在对话里逐步引导接入全部服务的配置提示语。实测包括迪士尼天气主动提醒、美团跑腿取号、支付宝账单统计等场景,作者认为手机端Personal Agent配置比电脑端Agent更简便,且会主动定时推送信息。
信息来源:公众号:卡尔的AI沃茨
教程 / 实战普通
Zero-Touch Open Source: How I Automated My GitLab MRs from Email to Merge Train with Antigravity IDE & CodeRabbit
A step-by-step guide to building a 100% autonomous workflow that reads GitLab maintainer emails, resolves reviews with CodeRabbit, heals CI/CD pipelines, and dispatches directly to the Merge Train.
信息来源:DEV Community
教程 / 实战普通
Your AI said "fixed" but never ran the tests? I made a plugin that won't let it say "done" until they pass
You ask Claude Code, Codex, Cursor, etc. for a quick fix. It changes the code and says "Fixed ✅"....
信息来源:DEV Community · Claude / Cursor
观点 / 方法普通
千问上线找题、拍题答疑与学习档案功能,帮学生精准练习
千问推出学习场景三项能力:输入地区、年级、学科和试卷关键词即可精准找题并免费下载打印;「拍题答疑」按阅卷标准给出解题步骤,卡住的步骤可进「小讲堂」由 AI 老师拆解讲解。日常错题会自动归入新升级的「学习档案」,通过“薄弱点分析”定位反复出错的知识点并针对性练习。
信息来源:公众号:千问APP(阿里) · Qwen
教程 / 实战普通
The Most Embarrassing Private Key in Web PKI History
The strange thing about this incident is that the certificates looked normal. They had familiar CA...
信息来源:DEV Community
行业动态普通
NVIDIA 与 Meta 扩大 Muse 模型合作并投资 30 亿美元
NVIDIA 与 Meta 宣布扩大围绕 Muse 及 Muse 系列模型的合作,优化推理与训练性能,并考虑在 NVIDIA OpenShell 参考架构和 NVIDIA Sentry 监控等方向继续合作。
信息来源:NVIDIA:Newsroom(RSS) · NVIDIA
模型发布 / 更新普通
vLLM Semantic Router 发布 27B 开源多模态决策模型 d3-ultra
vLLM Semantic Router 团队发布 Apache-2.0 开源的 27B 多模态决策模型 d3-ultra,属于 Decision 3.0 家族,不生成文本,而是单次读取输入后对每个问题返回概率,支持 Choice、Yes/No、Score、Box、Track 五类问题。
信息来源:MarkTechPost(RSS)
模型发布 / 更新普通
蚂蚁 inclusionAI 发布 Ling-3.1-flash-1019:560B 总参数 MoE,1M 上下文
蚂蚁 inclusionAI 发布 Ling-3.1-flash-1019,为 Ling-3.1-flash 的升级版,总参数 560B、每 token 激活 25B、上下文窗口 1M token。
信息来源:蚂蚁 inclusionAI:HuggingFace 新模型 · Hugging Face
教程 / 实战普通
Building an AI chatbot for UK users? A UK GDPR checklist for the code behind the chat window
An AI chatbot collects personal data from the very first message. People type their names, email...
信息来源:DEV Community
教程 / 实战普通
How I keep 13,000 kids' events fresh on Cloudflare Workers (and the 128 MB wall I hit)
Grassy Knees is a directory of real things to do with kids across 42 US metro areas. Its rule is...
信息来源:DEV Community
模型发布 / 更新普通
百度开源 Unlimited OCR 2,面向数百页长文档解析
百度开源 3.34B 参数的 OCR 模型 Unlimited OCR 2,MIT 许可,可单次请求解析数百页文档。核心是 SR-SWA 稀疏参考滑动窗口注意力,用 512 token 固定参考集加 128 token 局部窗口使每步解码开销基本不随页数增长。
信息来源:MarkTechPost(RSS)