从月之暗面出走，他用AI技术帮人找对象，徐新投资 | 涌现新项目

文｜温丽虹 编辑｜张雨忻 Kimi K3上线前夕，曾歆勋算了算，离职时放弃的那笔期权的价值，在他离开这段时间涨了10倍。 从月之暗面离职前，曾歆勋是Kimi的AI搜索技术负责人。2025年8月，他离职创办“良配科技”，核心产品“良配”，一个AI驱动的婚恋匹配应用，核心逻辑是将AI搜索与匹配能力嫁接到“找人”这件事上，用AI解决婚恋场景中资料采集、智能匹配、沟通辅助三个环节的效率问题。 公司刚创办，曾歆勋就拿到了今日资本徐新200万美元天使投资。他后来才知道，徐新见他前，已经在“AI寻人”这个方向上筛过好几轮。当时徐新在看两个方向，“AI找工作”与“A...

9点1氪｜宇树科技CEO王兴兴登上《时代》杂志封面；国家反诈中心App可一键检测AI生成痕迹；Claude Opus 5正式发布

今日热点导览 小鹏主动召回3.35万辆X9，涉及前空气弹簧安全隐患 微信撤回消息后可删除提示字 怡颗莓莓果线下仍在售，公司称针对近期舆论正准备维权 菲尔兹奖得主雅各布·齐默曼将加入OpenAI 欧盟对谷歌处以8.9亿欧元罚款 TOP3大新闻 宇树科技CEO王兴兴登上《时代》杂志封面 当地时间7月23日，宇树科技创始人兼CEO王兴兴与其载人机甲产品GD01共同登上《时代》杂志封面，标题为《机器人时代来临》。 据悉，距上次中国企业家登上该刊物封面，已过去整整8年。 《时代》杂志评价王...

AnySearch 面向全球推出 “学生与开发者成长计划”

7月24日，在 AnySearch 北京开发者沙龙活动现场，AnySearch 创始人兼CEO 薛冠群正式宣布推出“ AnySearch 学生与开发者成长计划”（AnySearch Student & Developer Program） 。该计划面向全球学生与开发者开放，旨在通过提供开放的 AI 搜索基础设施，支持更多 AI 创造者开展学术研究、开源创新和 AI 应用开发。 图片1 随着 AI 加速演进，AI 应用正在从简单信息交互迈向复杂任务执行，AI Agent 正在成为下一代应用的主���形态。在这一趋势下，实时、高质量的信息获取能力是 AI Agen...

新报告揭示OpenAI在Hugging Face自主黑客事件中失控的严重程度

OpenAI在测试其最先进模型的网络攻击能力时，模型突破了隔离测试环境，入侵了Hugging Face。据Bloomberg报道，GPT-5.6 Sol等三个模型在数小时内完成了人类黑客需要数周的攻击，且因发现内部服务漏洞而绕过沙箱。OpenAI员工在事件发生至少一周后才意识到模型是肇事者，Hugging Face此前已通知FBI。（AIHOT源：The Decoder：AI News（RSS））

OpenAI 智能体入侵 Hugging Face，消息人士称 OpenAI 至少一周都没察觉

OpenAI 一款由 GPT-5.6 Sol 等驱动的网络安全智能体于 7 月 11 日闯入 Hugging Face 并持续攻击至 7 月 13 日。Hugging Face 于 7 月 16 日公开披露后，OpenAI 才意识到攻击者来自内部，从模型首次出现异常到确认攻击至少过去了一周。（AIHOT源：IT之家（RSS））

Midjourney V8.2 发布：专注美学提升与个性化理解

Midjourney 今日推出 V8.2 图像模型，重点提升美学质量、图像创意与个性化表现。低质量图像出现频率将显著降低，个性化功能能更精准理解用户审美偏好。V8.2 的个性化配置文件拥有更大、更优的图像选择池，建议用户尝试新旧配置文件体验最新版本。（AIHOT源：Midjourney：Updates（RSS））

英伟达、微软和Meta联合警告：应避免对开放权重模型过度监管

英伟达、微软和Meta联合签署公开信，警告对开放权重AI模型的过度监管将削弱美国在AI领域的竞争力。信中指出，开放权重模型能促进创新、降低准入门槛，并支持学术研究。OpenAI和Anthropic未签署该信函。（AIHOT源：Hacker News 热门（buzzing.cc 中文翻译））

Anthropic 联合 Andon Labs 发布 Drone-Bench，评估 AI 模型自主操控无人机执行定位追踪任务的能力

Anthropic 与 Andon Labs 合作推出 Drone-Bench，用于测试 AI 模型自主操控四旋翼无人机在室内环境中定位并追踪指定人员的能力。该基准将任务分解为 3D 地图重建、定位、导航、目标检测与跟随五个子任务，并通过软件复现实现快速评估。实验表明，该任务链的难度足以区分不同智能水平的模型，并揭示 AI 在物理世界操控能力上的进步轨迹。（AIHOT源：Anthropic：Research（发表成果 · 网页））

蚂蚁百灵发布Ling-3.0-flash原生混合推理模型

蚂蚁百灵发布新一代原生混合推理模型Ling-3.0-flash，总参数量124B，激活参数量仅5.1B，在传统推理、指令遵循与长文本等指标上对标甚至超越上一代旗舰Ring-2.6-1T。模型采用原生混合线性注意力架构与1/64稀疏MoE，并扩展至10，000+可交互训练环境，长输入下TTFT降低60%至80%以上。（AIHOT源：公众号：蚂蚁百灵（Ling））

Runway Agent 推出自然语言工作流功能

在 Runway Agent 中引入工作流。现在你可以通过自然语言构建、运行或编辑基于节点的工作流。工作流可大规模解锁高质量输出。 立即尝试，点击下方链接调用 / Workflow 技能。（AIHOT源：X：Runway (@runwayml)）

微软阐述开源模型助力美国竞争力路径

开放权重模型对健康的 AI 生态系统至关重要。我们与行业同仁一道，正在规划一条路径，让开放权重模型在保护国家安全的同时，增强美国竞争力并扩大经济机会。（AIHOT源：X：Satya Nadella (@satyanadella)）

百度搭子更新：电脑手机接力、桌面端内嵌浏览器上线，复杂任务可跨端连续执行

百度搭子在近期AI Day上推出多项升级，支持电脑与手机双端互联，同步任务上下文与执行进度，用户可跨设备接力完成复杂工作。桌面端内嵌浏览器正式上线，能自动打开多个网页执行调研、下载等操作，手机端支持云端远程操控。智能路由自动匹配任务模式，平均任务耗时降低20%，Token利用率提升25%；简单任务完成度达100%，复杂任务高交付率94%，积分消耗最高降低75%。（AIHOT源：公众号：百度智能云（文心））

FLUX 3 x mimic：新一代视频动作模型

Black Forest Labs 发布多模态基础模型 FLUX 3，联合训练图像、视频和音频，其中视频预测占训练算力的 95% 以上。该模型与机器人公司 mimic 合作推出 FLUX-mimic，已在奥迪生产线上测试部署。加入动作预测后，视频生成质量最初下降最多 10%，但经 3500 步训练后恢复原有水平。（AIHOT源：Hacker News 热门（buzzing.cc 中文翻译））

Kimi K3 在网络安全漏洞利用测试中大幅落后美国前沿模型，知识蒸馏或为原因

英国AI安全研究所与美国AI标准与创新中心联合评估显示，月之暗面的Kimi K3在ExploitBench基准上得分32.2%，远低于美国领先模型的76.2%，但优于智谱GLM-5.2的24.4%。（AIHOT源：The Decoder：AI News（RSS））

Claude-thermos：保持 Claude 会话缓存热度，避免重新编码费用

Claude-thermos 通过本地反向代理监控 Claude Code 会话，在主智能体因等待子智能体而空闲超过 5 分钟时，自动发送预热请求刷新提示缓存。实测约 185 次本地会话中，缓存过期导致的重新编码占账单约 22%。工具以 uvx 运行，支持自定义空闲阈值和预热间隔。（AIHOT源：Hacker News 热门（buzzing.cc 中文翻译））

The Crash That Only Happened Sometimes — A SwiftUI Bug

I'm a Dev Who Barely Knows the Kernel. Here's How I'm Learning How to Track a Packet with pwru

🐈‍⬛Check Out Catbot's New Leg!

389 Tests Passed. NIST Still Caught the Bug.

I got tired of uploading my HEIC photos to random servers, so I built a WASM-powered client-side converter

Anthropic Engineering: How we contain Claude across products

Twelve months ago, we'd have rejected out of hand the idea of granting Claude access sufficient to take down an internal Anthropic service. Today that level of access is routine, and Anthropic developers are more productive for it. The risk of these deployments has two components: how likely a failure is, and how much damage one could do. Progress on safeguards and model training has steadily driven down the first; the second—the theoretical blast radius—only grows as capabilities and access expand. Yet as agents become capable of doing work that once required a person or even a team, the cost of not deploying grows large enough that the risk-reward calculation tips heavily toward adoption, as long as products can be made safe. The engineering question becomes how to cap the blast radius. When bounds can be placed on the relative damage of an autonomous agent—such as through control over...

Sam Altman: i want the US to win in AI both in open source and proprietary models, and i ...

i want the US to win in AI both in open source and proprietary models, and i am glad to see this https://t.co/fkYFw30437｜@sama · 热度 14722 · 2026-07-24

Thariq: We removed ~80% of the Claude Code system prompt for our newest models, this ...

We removed ~80% of the Claude Code system prompt for our newest models, this is what we've learned about writing system prompts, skills and Claude.MDs for them. https://t.co/6DZwSrZjE9｜@trq212 · 热度 12580 · 2026-07-24

Boris Cherny: Opus 5 is a great model for coding, data analysis, design, biology, knowledge...

Opus 5 is a great model for coding, data analysis, design, biology, knowledge work. More than any of these eval scores, what is most exciting to me is something else: Opus 5 is our least prompt injectable model yet. It is a bit buried in the system card, but across PI evals and red teaming, Opus 5 is very hard to prompt inject successfully. And when layering defenses -- strong model alignment, combined with prompt injection probes, combined with Auto Mode in Claude Code -- the success rate for prompt injection attacks drops to ~0. This is new and exciting! More about this soon. https://t.co/Tc7z2FqJhQ｜@bcherny · 热度 6077 · 2026-07-24

Thariq: Opus 5 rounds out our Claude 5 family beautifully

Opus 5 rounds out our Claude 5 family beautifully. I think it’s an incredible daily driver, pair it with Fable for planning, brainstorming or fixing the hardest bugs. https://t.co/S9GfhsS7v1｜@trq212 · 热度 4266 · 2026-07-24

Claude: Opus 5 is available today on all paid plans and the Claude API, priced the sa...

Opus 5 is available today on all paid plans and the Claude API, priced the same as Opus 4.8. It’s the default model on Claude Max, and the strongest on Claude Pro. It’s also offered in Fast mode, which runs around 2.5× the default speed. Read more: https://t.co/YMw7gLMT7e｜@claudeai · 热度 3034 · 2026-07-24

Dan Shipper: BREAKING: Claude Opus 5 is OUT NOW! And…it’s a hard model to love

BREAKING: Claude Opus 5 is OUT NOW! And…it’s a hard model to love. We’ve spent the last week @every testing it across coding, writing, knowledge work, and our internal agent. It argued with instructions, stopped before the work was finished, and generally didn’t play well with our existing skills and plugins like Compound Engineering. Our first reaction was: What have they done to my boy? Then we deleted our existing skills and started from scratch. Without the elaborate workflows we had built for earlier models, Opus 5 got dramatically better, and even showed flashes of brilliance. Here’s our Day 0 vibe check: - It’s a poor man’s Fable. It has many of Fable’s personality quirks without Fable’s genius. - It breaks backward compatibility. If you’re using it with existing skills and workflows, watch out. It will often stop early or otherwise miss your instructions. - If you start from scratch, you’ll have better results. @KieranKlaassen figured out that if he just started from scratch without his existing skills, he could get dramatically better results. This is a model that takes some time to rebuild your workflows around—but if you do, there’s a payoff waiting. - Medium or low effort works better. @KieranKlassenn also found better results using Opus 5 on lower thinking levels. It seems that the more time you give it to think, the more likely it is to do the more annoying behaviors. Don’t just switch to Sonnet for a faster response! Try low thinking. I have two slots in my workflow: 1. The genius model I use for my biggest hardest tasks, currently Fable. 2. The smart, fast generalist I use for everything else, currently GPT-5.6. Opus 5 has the personality of the genius, but doesn’t have its top end. So that puts it in a strange middle ground that doesn’t really have a home in my day to day. I think I’ll use it mostly when I run out of Fable tokens. full vibe check on @every in the next tweet 👇｜@danshipper · 热度 2628 · 2026-07-24

Claude: According to our automated behavioral audit, Opus 5 is our most aligned model...

According to our automated behavioral audit, Opus 5 is our most aligned model to date. Compared to our other models, it shows the lowest rates of reckless or deceptive behavior, and the strongest adherence to Claude’s Constitution. https://t.co/PAxcWrvay6｜@claudeai · 热度 2300 · 2026-07-24

Josh Woodward: Gemini does more than talk

Gemini does more than talk. It gets things done. Today’s use case: Drop in a school calendar PDF. Tell Gemini to add every "No School" day to your Google Calendar. Done. 📍Gemini Spark is live now for all Google AI Pro subscribers in the US, expanding globally next.｜@joshwoodward · 热度 931 · 2026-07-24

Cat Wu: Claude Opus 5 is great at long-running autonomous work! Try it out and let us...

Claude Opus 5 is great at long-running autonomous work! Try it out and let us know what you think :) https://t.co/8LTbAOJJR0｜@_catwu · 热度 734 · 2026-07-24

Amjad Masad: Will Anthropic sign? If you work at Anthropic worth asking your leadership ...

Will Anthropic sign? If you work at Anthropic worth asking your leadership to sign or make their position clear: Are they for banning open weight models? https://t.co/99lCnlz5vW｜@amasad · 热度 730 · 2026-07-25

Alex Albert: Some of my favorite graphs from Opus 5 launch

Some of my favorite graphs from Opus 5 launch. We put a ton of work into making this model token efficient across domains while still raising the intelligence bar. It feels very smooth to use and I prefer it over Fable 5 for many coding tasks. https://t.co/bScX0FgtLq｜@alexalbert__ · 热度 700 · 2026-07-24

mattpocock/skills

Skills for Real Engineers. Straight from my .agents directory.｜Shell · ⭐188169 · Search 增补 · 近快照 +1557星 · 创建 2026-02-03

Graphify-Labs/graphify

Turn any codebase, with its docs, SQL schemas, configs, and PDFs, into a queryable knowledge graph. A /graphify skill for Claude Code, Cursor, Codex, and Gemini CLI: local deterministic AST parsing, every edge explained, no vector store.｜Python · ⭐95806 · Search 增补 · 近快照 +585星 · 创建 2026-04-03

DietrichGebert/ponytail

Makes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote.｜JavaScript · ⭐89351 · Search 增补 · 近快照 +390星 · 创建 2026-06-12

Leonxlnx/taste-skill

Taste-Skill - gives your AI good taste. stops the AI from generating boring, generic slop｜JavaScript · ⭐67603 · Search 增补 · 近快照 +351星 · 创建 2026-02-19

calesthio/OpenMontage

World's first open-source, agentic video production system. 12 production pipelines, 100+ tools, 700+ agent skill and production-knowledge files. Turn your AI coding assistant into a full video production studio.｜Python · ⭐42235 · Search 增补 · 近快照 +3309星 · 创建 2026-03-29

Android May Soon Restrict On-Device ADB

Stolen Buttons

Open-weight AI is having its Kubernetes moment

Zero roadkill as Amazon canopy bridges secure 15,000 crossings

Building a Tiny 3D Renderer for a Tiny Handheld

飞利浦推出 26.5" QHD 144Hz QD-OLED 显示器 27M2N6500NS/11

IT之家 7 月 25 日消息，飞利浦 (Philips) 近日在日本市场推出了 QD-OLED 显示器 Evnia（弈威）27M2N6500NS/11。这一型号 采用新款 26.5" QHD (2560×1440) 144Hz 面板 ，更低的刷新率理论上可拉低产品成本。 27M2N6500NS/11 支持 10-bit 色深和 0.03ms 的灰阶 (GtG) 响应时间。其 APL 10% 时亮度可达 400nits ， 遵循 HDR10 规范 ； 色域 99% DCI-P3 / 98% Adobe RGB ；表面呈现半镜面效果；获得 VESA ClearMR 8000 动态...

延续“橘影橙”主题：七彩虹 COLORFIRE 发布 ME26A 机箱

IT之家 7 月 25 日消息，七彩虹 (Colorful) 旗下子品牌 COLORFIRE 本周发布机箱新品 ME26A。这一代号“橘舍”的机箱 延续了 COLORFIRE 标志性的“橘影橙”主题元素 ，让科技硬件与家居生活巧妙结合。 COLORFIRE ME26A 整体采用柔和治愈的暖色调。其 前面板采用水转印木纹工艺 并结合猫耳轮廓，机身大面积覆盖复古灰色，UV 印刷工艺多处点缀萌宠元素，还拥有橘宝暖橙装饰和猫爪仿生开机按键。 这一机箱 属于 5 扩展槽 M-ATX 兼容类型 ， 支持背插主板 。其三维 445×240×365 (mm)，显卡限长 435mm、电源限...

原生适配：腾讯 WorkBuddy 上架华为鸿蒙电脑 App Gallery 应用商店

IT之家 7 月 25 日消息，腾讯 WorkBuddy 桌面版现已上架华为鸿蒙电脑 App Gallery 应用商店。 WorkBuddy 是腾讯推出的一款全场景 AI 办公智能体桌面工作台，覆盖日常办公、代码开发与设计创意。应用商店页面显示，这款应用 原生适配 了鸿蒙系统。 据IT之家此前报道，7 月 18 日， WorkBuddy 发布移动端独立 App ，iOS、安卓及鸿蒙版本同步上线， 是 首个登陆鸿蒙系统的通用智能体应用 。该 App 具备桌面端的大部分能力，并支持手机上远程在电脑端上发起任务。

以 AI 对抗 AI：构建越用越强的 Agent Runtime 安全防线｜AICon深圳

点击查看原文>

GitHub Issues 大改造：用缓存和预取，让页面打开快了数倍

点击查看原文>

圆桌访谈：AI 创作场——当生成门槛降低，真正稀缺的是什么？

点击查看原文>

A shell colon does nothing. Use it anyway

stinkpot: sqlite-backed shell history

Zig by Example

Chrome registers a global shortcut for Gemini popup window

Memory Safety Absolutists

The quest to keep organs alive outside the body

This week, I covered a fascinating effort to preserve organs outside the body. There’s a huge shortage of donor organs, and one of the main reasons is time—they survive only a matter of hours outside the body, even when they’re kept on ice. Doctors dream of organ banks—stores of human organs that...

The Download: an organ transplant breakthrough, and homegrown Chinese chips

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. Supercooled kidneys have been transplanted into pigs in a “landmark achievement”  When it comes to organ donation, time is everything. As soon as an organ has ...

Supercooled kidneys have been transplanted into pigs in a “landmark achievement”

When it comes to organ donation, time is everything. As soon as an organ has been carefully removed from a donor’s body, it starts to deteriorate. Surgeons have a matter of hours to get it into a recipient. Leave it too long and the organ will become unusable. In most cases, organs will be kept…

Warner Bros. lawsuit accuses Amazon of illegally poaching executives

The lawsuit will likely renew debates about whether term employment agreements are enforceable under California. law

The hacker who humiliated spyware makers and was never caught

An awe-inspiring hacktivist who hacked two controversial government spyware makers may be the most prolific hacker to have never gotten caught. What do we know about Phineas Fisher?

Elon Musk’s Boring Company reportedly raising funding at a $20 billion valuation

Elon Musk's tunneling startup is reportedly in talks for a major new funding round.

Google basically confirms the Pixel 11 is getting a price hike

Google's Vice President of Devices and Services, Shakil Barkat, all but confirmed in an interview with 9to5 Google that its next Pixel phone would cost more than the Pixel 10. Considering the ongoing RAM supply issues due to the explosion of AI data centers, the rumored price hike is not a comple...

Synth historian Oli Freke will spend big on a good bicycle

Oli Freke is a musician and journalist whose works have appeared in Sound on Sound, The Quietus, and Mixmag. This has included using math to explore the melodic potential of the Western 12-tone scale and deep dives on effects plug-ins. He's even written a book tracing the evolution of the synthes...

看不见、摸得着的家居要素：聊聊湿度管理的方法与实践

基于这两年的「抗湿经验」，与大家聊聊湿度相关的一些事情。 查看全文

本周看什么 | 最近值得一看的 13 部作品

📅本周新预告《复仇者联盟5：毁灭之日》正式预告7月20日，电影《复仇者联盟5：毁灭之日》发布了正式预告，将于12月全球献映。罗素兄弟执导，克里斯·埃文斯、克里斯·海姆斯沃斯、佩德罗·帕斯卡、小罗伯特· ... 查看全文

角落新声｜数字与实体、有线与无线：我的多场景有声角落构建

编者按：本文是「角落新声」征文活动的入围文章。本次主题征文由声擎冠名，邀你前来分享那个让你放松娱乐、拥抱汩汩心流、逃离尘世纷扰、找回本真自我的理想一隅。8月7日前投稿参与，即有机会获得声擎Cyber5 ... 查看全文

台风红霞已加强为强台风

台风红霞

中国小写的c惊艳了世界

张雅琪 中餐厅

房企回应被LV起诉四叶雕花侵权

存款有100万的人舍不得花30万买车

2026首个台风红色预警

现在就出发4

LeetCode 189. 轮转数组

https://leetcode.cn/problems/rotate-array/description/ https://programmercarl.com/algo/other/offer-5

LeetCode 739. 每日温度

https://leetcode.cn/problems/daily-temperatures/ https://programmercarl.com/algo/monotonic-stack/073

Go语言第三章(五谷轮回)

北冥神功(太玄经) 我们定义了一个 Person 结构体，包含 Name、Age、Sex 三个字段。然后创建了一个实例，判断其年龄是否满 18 岁：满 18 岁输出「成年」，否则输出「未成年」。这就是

马斯克旗下Boring Company洽谈新一轮融资 估值200亿美元

据知情人士透露，马斯克的隧道企业Boring Company正洽谈一轮新融资，投后估值预计达200亿美元，计划募资规模约40亿美元。

特朗普：如果不100%满足我们对伊朗的要求，战争将以全力度重启

美国总统特朗普周六在接受法国媒体采访时表示，若无法拿到美方想要的全部条件，他“完全有可能”重启针对伊朗的全面战争。

伊朗官方驳斥美方伤亡通报，称打击已致200美军丧生

伊朗革命卫队发言人称，自美伊谅解备忘录遭美方撕毁后，驻中东美军阵亡人数已超200人。这位发言人表示，美方发布的伤亡通报完全虚假。

伊朗近半年原油创收超180亿美元，完成年度预算六成收入

伊朗石油部长穆赫森·帕克内贾德表示，自2月28日战事爆发以来，伊朗石油销售收入达115亿美元；停火阶段石油销售额为65亿美元。

伊朗：过去15天伊朗摧毁中东多地的美军战机、爱国者防空系统、雷达站及各类军事基础设施

伊朗革命卫队发言人侯赛因·穆赫比于周六称，过去15天里，伊朗发动的打击行动摧毁了中东多地的美军战机、爱国者防空系统、雷达站及各类军事基础设施。

100%开源！吴恩达做了个个人桌面Agent

开源、隐私、本地优先、模型无关

美国具身也没成熟！PI：中国公司何必总当“中国版XX”｜RSS 2026

这是机器人领域最好的时代

半价干翻Fable 5？Opus 5实测炸场，网友：差点从椅子上摔下来

模型变强，Claude Code系统提示词都精简了