[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"consumer-news-home-v1":3,"consumer-news-latest-feed-v1":550},{"channel":4,"generatedAt":5,"operationBlocks":6},"news",1788862947303,[7,119,295,408],{"blockCode":8,"blockType":9,"layoutType":10,"title":11,"subtitle":12,"itemLimit":13,"moreUrl":14,"cardVariant":15,"themeVariant":16,"rankCode":17,"rankName":18,"cards":19,"rankEntries":89},"news_lead","FEED","LEAD_GRID","今日焦点","编辑判断与实时热点并重",5,"\u002Fnews\u002Fsearch","NEWS_LEAD_STORY","SIGNAL_RED","news_lead_visual","资讯有图头条榜",[20,40,52,64,77],{"id":21,"kind":4,"targetType":22,"targetId":23,"subtype":4,"typeLabel":24,"title":25,"subtitle":26,"summary":27,"coverUrl":28,"badgeText":29,"href":30,"sourceName":26,"meta":31,"metrics":34,"tags":35,"operationItemId":36,"sortOrder":37,"targetCode":38,"resolved":39},"NEWS_ARTICLE:863","NEWS_ARTICLE",863,"资讯","弱模型不能裸奔：Agent Harness 凭什么真实有效","博客园","Harness 不是给弱模型贴的创可贴。它是把工程纪律——验证、门禁、不变量、路由——变成架构里一等公民的方式。模型每半年换一代，今天省钱的 Flash 明天可能就过时了，但那套'默认怀疑、机器校验、按决策密度调度'的流程会留下来，并且越跑越值钱。\n弱模型不能裸奔。给它穿上 harness，便宜才真","https:\u002F\u002Fimg2024.cnblogs.com\u002Fblog\u002F510\u002F202608\u002F510-20260830181633785-1171146572.jpg","","\u002Fnews\u002F863",[32,33],"2026","综合技术",{},[33],1322452,1,null,true,{"id":41,"kind":4,"targetType":22,"targetId":42,"subtype":4,"typeLabel":24,"title":43,"subtitle":26,"summary":44,"coverUrl":45,"badgeText":29,"href":46,"sourceName":26,"meta":47,"metrics":48,"tags":49,"operationItemId":50,"sortOrder":51,"targetCode":38,"resolved":39},"NEWS_ARTICLE:858",858,"Agent上下文管理概述-1","Agent上下文管理核心是组织与压缩。组织采用静态+动态提示词：启动时载入系统提示词、工具描述、Skill名称路径；运行时拼接用户消息、工具调用及结果，并依KV缓存命中优化排列。压缩需兼顾速度与token占用。工业界常用确定性清理加摘要：Pi\n  Agent压缩历史后重建上下文，OpenCode对工","https:\u002F\u002Ffiles.seeusercontent.com\u002F2026\u002F08\u002F19\u002FxH5o\u002F20260819210036197.png","\u002Fnews\u002F858",[32],{},[],1322453,2,{"id":53,"kind":4,"targetType":22,"targetId":54,"subtype":4,"typeLabel":24,"title":55,"subtitle":26,"summary":56,"coverUrl":57,"badgeText":29,"href":58,"sourceName":26,"meta":59,"metrics":60,"tags":61,"operationItemId":62,"sortOrder":63,"targetCode":38,"resolved":39},"NEWS_ARTICLE:857",857,"Palantir 眼中的一次“订单可能延期”，如何成为实时决策的起点？","关键客户订单 SO-2026-0812 明早必须发货，但产线设备异常、核心零件库存不足。这个场景的重点不只是识别风险，而是让分散信息支撑后续判断与行动。下面以此示例来理解 Palantir 的 Foundry、Ontology、AIP、Apollo 与 Gotham 都将扮演怎样的角色；当然，这里是","https:\u002F\u002Fimg2024.cnblogs.com\u002Fblog\u002F635610\u002F202609\u002F635610-20260902065517024-771811555.png","\u002Fnews\u002F857",[32],{},[],1322454,3,{"id":65,"kind":4,"targetType":22,"targetId":66,"subtype":4,"typeLabel":24,"title":67,"subtitle":26,"summary":68,"coverUrl":69,"badgeText":29,"href":70,"sourceName":26,"meta":71,"metrics":73,"tags":74,"operationItemId":75,"sortOrder":76,"targetCode":38,"resolved":39},"NEWS_ARTICLE:855",855,"被罚了500后，整个人都变老实了","我们组有个同事，以前是那种闲不住的人。现在也闲不住 只不过是不敢动了。 去年部门空降了一位新总监，阿里P8，第一次全员会PPT上就八个大字：拥抱变化，永不言弃。 新领导上来搞流程改革，每个模块指定Owner，出事追责到人，A级事故罚款500起步，B级300 依次类推，发版上线改配置全走审批。说实话之","https:\u002F\u002Fimg2024.cnblogs.com\u002Fblog\u002F273387\u002F202608\u002F273387-20260818222617092-925481998.png","\u002Fnews\u002F855",[32,72],"软件开发",{},[72],1322455,4,{"id":78,"kind":4,"targetType":22,"targetId":79,"subtype":4,"typeLabel":24,"title":80,"subtitle":26,"summary":81,"coverUrl":82,"badgeText":29,"href":83,"sourceName":26,"meta":84,"metrics":86,"tags":87,"operationItemId":88,"sortOrder":13,"targetCode":38,"resolved":39},"NEWS_ARTICLE:853",853,"为什么团队用了 AI 之后，测试质量反而下降了！","一、AI 提效之后，往往先崩的是质量 前两天刷到一篇同行的复盘。 「AI 提效半年后，我们的测试团队规模反而更大了」 用AI 提效半年，人数不降，反而变更多了？ 这位老哥是认真做过团队实践的，观察写得很扎实。提测质量肉眼可见地下滑，需求做漏了，单位时间里测试捞出来的 bug 反而变多，最后测试团队加","https:\u002F\u002Fimage.kjdaohang.com\u002Fimg\u002F20260831223910845.png","\u002Fnews\u002F853",[32,85],"人工智能",{},[85],1322456,[90,99,104,109,114],{"rankNo":37,"score":91,"trendType":92,"trendValue":93,"reason":94,"card":95},0,"SAME","持平","阅读、互动与发布时间综合热度",{"id":21,"kind":4,"targetType":22,"targetId":23,"subtype":4,"typeLabel":24,"title":25,"subtitle":26,"summary":27,"coverUrl":28,"badgeText":29,"href":30,"sourceName":26,"meta":96,"metrics":97,"tags":98,"operationItemId":36,"sortOrder":37,"targetCode":38,"resolved":39},[32,33],{},[33],{"rankNo":51,"score":91,"trendType":92,"trendValue":93,"reason":94,"card":100},{"id":41,"kind":4,"targetType":22,"targetId":42,"subtype":4,"typeLabel":24,"title":43,"subtitle":26,"summary":44,"coverUrl":45,"badgeText":29,"href":46,"sourceName":26,"meta":101,"metrics":102,"tags":103,"operationItemId":50,"sortOrder":51,"targetCode":38,"resolved":39},[32],{},[],{"rankNo":63,"score":91,"trendType":92,"trendValue":93,"reason":94,"card":105},{"id":53,"kind":4,"targetType":22,"targetId":54,"subtype":4,"typeLabel":24,"title":55,"subtitle":26,"summary":56,"coverUrl":57,"badgeText":29,"href":58,"sourceName":26,"meta":106,"metrics":107,"tags":108,"operationItemId":62,"sortOrder":63,"targetCode":38,"resolved":39},[32],{},[],{"rankNo":76,"score":91,"trendType":92,"trendValue":93,"reason":94,"card":110},{"id":65,"kind":4,"targetType":22,"targetId":66,"subtype":4,"typeLabel":24,"title":67,"subtitle":26,"summary":68,"coverUrl":69,"badgeText":29,"href":70,"sourceName":26,"meta":111,"metrics":112,"tags":113,"operationItemId":75,"sortOrder":76,"targetCode":38,"resolved":39},[32,72],{},[72],{"rankNo":13,"score":91,"trendType":92,"trendValue":93,"reason":94,"card":115},{"id":78,"kind":4,"targetType":22,"targetId":79,"subtype":4,"typeLabel":24,"title":80,"subtitle":26,"summary":81,"coverUrl":82,"badgeText":29,"href":83,"sourceName":26,"meta":116,"metrics":117,"tags":118,"operationItemId":88,"sortOrder":13,"targetCode":38,"resolved":39},[32,85],{},[85],{"blockCode":120,"blockType":9,"layoutType":121,"title":122,"subtitle":123,"itemLimit":124,"moreUrl":14,"cardVariant":125,"themeVariant":126,"rankCode":127,"rankName":128,"cards":129,"rankEntries":243},"news_latest_feed","TIMELINE","刚刚发生","按发布时间连续阅读",10,"NEWS_TIMELINE_FEED","INK_BLUE","news_latest","资讯最新发布榜",[130,141,152,163,174,185,197,209,221,232],{"id":131,"kind":4,"targetType":22,"targetId":132,"subtype":4,"typeLabel":24,"title":133,"subtitle":26,"summary":134,"coverUrl":135,"badgeText":29,"href":136,"sourceName":26,"meta":137,"metrics":138,"tags":139,"operationItemId":140,"sortOrder":37,"targetCode":38,"resolved":39},"NEWS_ARTICLE:926",926,"数学建模第五课：从矩形面积最大问题学会建立优化模型","面向大学生与数学建模爱好者的优化模型入门课：讲解优化问题建模的通用格式与三要素，以矩形面积最大问题完整演示「问题提出→问题分析→建立模型→模型求解→结果分析」的建模主要流程，并介绍敏感性分析与稳健性分析的基本概念，以及优化模型的多维度五层分类体系。","https:\u002F\u002Fimg2024.cnblogs.com\u002Fblog\u002F3840373\u002F202608\u002F3840373-20260829142423712-1690016609.jpg","\u002Fnews\u002F926",[32,33],{},[33],1322471,{"id":142,"kind":4,"targetType":22,"targetId":143,"subtype":4,"typeLabel":24,"title":144,"subtitle":26,"summary":145,"coverUrl":29,"badgeText":29,"href":146,"sourceName":26,"meta":147,"metrics":149,"tags":150,"operationItemId":151,"sortOrder":51,"targetCode":38,"resolved":39},"NEWS_ARTICLE:925",925,"订单的含金量在分化","复杂的流程，在产品设计阶段就需要综合考虑多方面的建议，前期业务和技术的介入，避免产品层面出现合理但不合适的设计，过繁或者过简都不利于整体的迭代节奏。","\u002Fnews\u002F925",[32,148],"数据与架构",{},[148],1322472,{"id":153,"kind":4,"targetType":22,"targetId":154,"subtype":4,"typeLabel":24,"title":155,"subtitle":26,"summary":156,"coverUrl":157,"badgeText":29,"href":158,"sourceName":26,"meta":159,"metrics":160,"tags":161,"operationItemId":162,"sortOrder":63,"targetCode":38,"resolved":39},"NEWS_ARTICLE:924",924,"实战对比：用 mpremote\u002FPutty\u002FMobaXterm 连接树莓派 Pico REPL，谁更省心？","我们可以利用多个远程控制终端通过 USB 线和树莓派 Pico 进行连接通信，常见的远程控制终端包括 MobaXterm、Putty、mpremote、Xshell 等。","http:\u002F\u002Fopenwrite.cn\u002Fuploads\u002F21597\u002F62887\u002F0876b5b0-dd6c-4902-8264-ee8e40b45997.png","\u002Fnews\u002F924",[32],{},[],1322473,{"id":164,"kind":4,"targetType":22,"targetId":165,"subtype":4,"typeLabel":24,"title":166,"subtitle":26,"summary":167,"coverUrl":168,"badgeText":29,"href":169,"sourceName":26,"meta":170,"metrics":171,"tags":172,"operationItemId":173,"sortOrder":76,"targetCode":38,"resolved":39},"NEWS_ARTICLE:923",923,"为什么说WEB框架PasteApeart是AI Coding时代的黄金搭档，从此开启敏捷开发新思路","在AI代码生成工具大行其道的今天，一个大胆的论断正在开发者社区流传：有一个WEB框架，其开发效率可以超越AI。这听起来像是天方夜谭，但PasteApeart（即升级版PasteForm）正将这一设想变为现实。它并非又一个功能堆砌的框架，而是一套从源头解决“代码该写在哪”这一核心痛点的思想与约定体系。","https:\u002F\u002Fsoft.pastecode.cn\u002Fupload\u002Fbookmark\u002F202608\u002F2f7e8905b81611ff5302a1f01174b2d9.webp","\u002Fnews\u002F923",[32],{},[],1322474,{"id":175,"kind":4,"targetType":22,"targetId":176,"subtype":4,"typeLabel":24,"title":177,"subtitle":26,"summary":178,"coverUrl":179,"badgeText":29,"href":180,"sourceName":26,"meta":181,"metrics":182,"tags":183,"operationItemId":184,"sortOrder":13,"targetCode":38,"resolved":39},"NEWS_ARTICLE:922",922,"用 crontab 给 LLM 使用量装上“监控眼”","在把大模型接入日常工作流之后，笔者很快遇到了一个新问题：模型到底被用了多少次？每天的高峰时段是什么时候？周末是不是真的没人调用？如果对这些数据一无所知，就谈不上优化成本、排查异常，更谈不上为后续扩容做规划。 于是，笔者为 LLM 使用量加了一层可观测性，做法很简单——用 Linux 自带的 cron","https:\u002F\u002Fimg2024.cnblogs.com\u002Fblog\u002F635610\u002F202608\u002F635610-20260830013552425-1810240964.png","\u002Fnews\u002F922",[32],{},[],1322475,{"id":186,"kind":4,"targetType":22,"targetId":187,"subtype":4,"typeLabel":24,"title":188,"subtitle":26,"summary":189,"coverUrl":190,"badgeText":29,"href":191,"sourceName":26,"meta":192,"metrics":193,"tags":194,"operationItemId":195,"sortOrder":196,"targetCode":38,"resolved":39},"NEWS_ARTICLE:921",921,"编程语言的「第三条道路」上，走得最远的其实是 C#","不必要求每个开发者都成为证明专家，而是把验证、内存安全、类型纪律，一层层织进语言和平台的基础设施里，让正确的代码成为默认路径","https:\u002F\u002Fimg2024.cnblogs.com\u002Fblog\u002F510\u002F202608\u002F510-20260830081613420-742582120.jpg","\u002Fnews\u002F921",[32,72],{},[72],1322476,6,{"id":198,"kind":4,"targetType":22,"targetId":199,"subtype":4,"typeLabel":24,"title":200,"subtitle":26,"summary":201,"coverUrl":202,"badgeText":29,"href":203,"sourceName":26,"meta":204,"metrics":205,"tags":206,"operationItemId":207,"sortOrder":208,"targetCode":38,"resolved":39},"NEWS_ARTICLE:920",920,"AI编程实战：博客建站+RAG知识库","大家好，我是Java烘焙师。最近利用业余时间，完成了博客建站+RAG知识库的搭建，分享一下过程中遇到的选型问题、实现步骤。 搭建博客站点和RAG知识库的初衷，是因为日积月累写了几十篇技术文章，希望有一个独立的站点，并且能用自然语言问答、查找知识点。 下面是用到的技术栈： 静态页面构建：docmd 网","https:\u002F\u002Fimg2024.cnblogs.com\u002Fblog\u002F1247698\u002F202608\u002F1247698-20260829115308559-547026660.png","\u002Fnews\u002F920",[32,85],{},[85],1322477,7,{"id":210,"kind":4,"targetType":22,"targetId":211,"subtype":4,"typeLabel":24,"title":212,"subtitle":26,"summary":213,"coverUrl":214,"badgeText":29,"href":215,"sourceName":26,"meta":216,"metrics":217,"tags":218,"operationItemId":219,"sortOrder":220,"targetCode":38,"resolved":39},"NEWS_ARTICLE:919",919,"MicroPython 开发避坑：用 Signal 类解决不同电路的电平兼容问题","在以下例程中，我们通过 Signal 类控制两个不同电路连接的 LED 灯，其中 `GP17` 连接的 LED 灯负极连接到了地，对应 GPIO 输出高电平点亮；而 `GP16` 连接的 LED 灯正极连接到了 `+3V3`，对应 GPIO 输出低电平点亮。我们通过 `Signal` 类来包装现有的","http:\u002F\u002Fopenwrite.cn\u002Fuploads\u002F21597\u002F62896\u002Fad2af538-1ed7-46d9-a676-f8cdceefd7a1.png","\u002Fnews\u002F919",[32],{},[],1322478,8,{"id":222,"kind":4,"targetType":22,"targetId":223,"subtype":4,"typeLabel":24,"title":224,"subtitle":26,"summary":225,"coverUrl":29,"badgeText":29,"href":226,"sourceName":26,"meta":227,"metrics":228,"tags":229,"operationItemId":230,"sortOrder":231,"targetCode":38,"resolved":39},"NEWS_ARTICLE:918",918,"AI加持后2天时间将公司的运维自动化提高了一个层次","近期工作安排包括自动化测试、自动化运维、自动化运营和安全等一些工作。自己做产品、自己做设计、自己做开发、自己验收上线还是挺爽滴。 这其中工作简单的就是自动化运维。但我今天真正要讲的不是做了什么，或者用了什么技术，这些都不是核心问题。核心是面对一句话需求，要怎样去思考。 我拿到的需求就是 做全公司的自","\u002Fnews\u002F918",[32],{},[],1322479,9,{"id":233,"kind":4,"targetType":22,"targetId":234,"subtype":4,"typeLabel":24,"title":235,"subtitle":26,"summary":236,"coverUrl":237,"badgeText":29,"href":238,"sourceName":26,"meta":239,"metrics":240,"tags":241,"operationItemId":242,"sortOrder":124,"targetCode":38,"resolved":39},"NEWS_ARTICLE:917",917,"[python] pywinauto使用指北","当业务系统没有API、命令行接口或可直接集成的数据通道时，桌面自动化往往是打通业务流程的最后一公里。pywinauto库通过Win32 API与Microsoft UI Automation（UIA）访问Windows窗口及控件，使Python脚本能够驱动桌面应用。本文聚焦于pywinauto的基础","https:\u002F\u002Fgitlab.com\u002Fluohenyueji\u002Farticle_picture_warehouse\u002F-\u002Fraw\u002Fmain\u002FCSDN\u002F%5Bpython%5D%20pywinauto%E4%BD%BF%E7%94%A8%E6%8C%87%E5%8C%97\u002Fimgs\u002F1.jpg","\u002Fnews\u002F917",[32,72],{},[72],1322480,[244,250,255,260,265,270,275,280,285,290],{"rankNo":37,"score":91,"trendType":92,"trendValue":93,"reason":245,"card":246},"按发布时间更新的最新资讯",{"id":131,"kind":4,"targetType":22,"targetId":132,"subtype":4,"typeLabel":24,"title":133,"subtitle":26,"summary":134,"coverUrl":135,"badgeText":29,"href":136,"sourceName":26,"meta":247,"metrics":248,"tags":249,"operationItemId":140,"sortOrder":37,"targetCode":38,"resolved":39},[32,33],{},[33],{"rankNo":51,"score":91,"trendType":92,"trendValue":93,"reason":245,"card":251},{"id":142,"kind":4,"targetType":22,"targetId":143,"subtype":4,"typeLabel":24,"title":144,"subtitle":26,"summary":145,"coverUrl":29,"badgeText":29,"href":146,"sourceName":26,"meta":252,"metrics":253,"tags":254,"operationItemId":151,"sortOrder":51,"targetCode":38,"resolved":39},[32,148],{},[148],{"rankNo":63,"score":91,"trendType":92,"trendValue":93,"reason":245,"card":256},{"id":153,"kind":4,"targetType":22,"targetId":154,"subtype":4,"typeLabel":24,"title":155,"subtitle":26,"summary":156,"coverUrl":157,"badgeText":29,"href":158,"sourceName":26,"meta":257,"metrics":258,"tags":259,"operationItemId":162,"sortOrder":63,"targetCode":38,"resolved":39},[32],{},[],{"rankNo":76,"score":91,"trendType":92,"trendValue":93,"reason":245,"card":261},{"id":164,"kind":4,"targetType":22,"targetId":165,"subtype":4,"typeLabel":24,"title":166,"subtitle":26,"summary":167,"coverUrl":168,"badgeText":29,"href":169,"sourceName":26,"meta":262,"metrics":263,"tags":264,"operationItemId":173,"sortOrder":76,"targetCode":38,"resolved":39},[32],{},[],{"rankNo":13,"score":91,"trendType":92,"trendValue":93,"reason":245,"card":266},{"id":175,"kind":4,"targetType":22,"targetId":176,"subtype":4,"typeLabel":24,"title":177,"subtitle":26,"summary":178,"coverUrl":179,"badgeText":29,"href":180,"sourceName":26,"meta":267,"metrics":268,"tags":269,"operationItemId":184,"sortOrder":13,"targetCode":38,"resolved":39},[32],{},[],{"rankNo":196,"score":91,"trendType":92,"trendValue":93,"reason":245,"card":271},{"id":186,"kind":4,"targetType":22,"targetId":187,"subtype":4,"typeLabel":24,"title":188,"subtitle":26,"summary":189,"coverUrl":190,"badgeText":29,"href":191,"sourceName":26,"meta":272,"metrics":273,"tags":274,"operationItemId":195,"sortOrder":196,"targetCode":38,"resolved":39},[32,72],{},[72],{"rankNo":208,"score":91,"trendType":92,"trendValue":93,"reason":245,"card":276},{"id":198,"kind":4,"targetType":22,"targetId":199,"subtype":4,"typeLabel":24,"title":200,"subtitle":26,"summary":201,"coverUrl":202,"badgeText":29,"href":203,"sourceName":26,"meta":277,"metrics":278,"tags":279,"operationItemId":207,"sortOrder":208,"targetCode":38,"resolved":39},[32,85],{},[85],{"rankNo":220,"score":91,"trendType":92,"trendValue":93,"reason":245,"card":281},{"id":210,"kind":4,"targetType":22,"targetId":211,"subtype":4,"typeLabel":24,"title":212,"subtitle":26,"summary":213,"coverUrl":214,"badgeText":29,"href":215,"sourceName":26,"meta":282,"metrics":283,"tags":284,"operationItemId":219,"sortOrder":220,"targetCode":38,"resolved":39},[32],{},[],{"rankNo":231,"score":91,"trendType":92,"trendValue":93,"reason":245,"card":286},{"id":222,"kind":4,"targetType":22,"targetId":223,"subtype":4,"typeLabel":24,"title":224,"subtitle":26,"summary":225,"coverUrl":29,"badgeText":29,"href":226,"sourceName":26,"meta":287,"metrics":288,"tags":289,"operationItemId":230,"sortOrder":231,"targetCode":38,"resolved":39},[32],{},[],{"rankNo":124,"score":91,"trendType":92,"trendValue":93,"reason":245,"card":291},{"id":233,"kind":4,"targetType":22,"targetId":234,"subtype":4,"typeLabel":24,"title":235,"subtitle":26,"summary":236,"coverUrl":237,"badgeText":29,"href":238,"sourceName":26,"meta":292,"metrics":293,"tags":294,"operationItemId":242,"sortOrder":124,"targetCode":38,"resolved":39},[32,72],{},[72],{"blockCode":296,"blockType":297,"layoutType":298,"title":299,"subtitle":300,"itemLimit":196,"moreUrl":14,"cardVariant":301,"themeVariant":302,"rankCode":127,"rankName":128,"cards":303,"rankEntries":377},"news_editorial","GRID","EDITORIAL_MOSAIC","值得细读","编辑精选的深度内容","NEWS_EDITORIAL_MOSAIC","CYAN_FOCUS",[304,316,328,340,352,365],{"id":305,"kind":4,"targetType":22,"targetId":306,"subtype":4,"typeLabel":24,"title":307,"subtitle":26,"summary":308,"coverUrl":309,"badgeText":29,"href":310,"sourceName":26,"meta":311,"metrics":312,"tags":313,"operationItemId":314,"sortOrder":315,"targetCode":38,"resolved":39},"NEWS_ARTICLE:916",916,"经常用 Codex 后，我发现 AGENTS.md 只该管一件事","让 Codex 少走弯路：一份全局 AGENTS.md 的取舍 Codex GPT-5.6、Claude Fable 这一代模型变强后，我反而开始删 Prompt。 以前总怕 AI 理解错。 恨不得把“先看什么、再搜什么、怎么改、最后怎么回复”，一条条写进全局 AGENTS.md。 结果呢？ 规则越","https:\u002F\u002Fimg2024.cnblogs.com\u002Fblog\u002F1001668\u002F202608\u002F1001668-20260830114848294-1139249921.png","\u002Fnews\u002F916",[32],{},[],1322481,11,{"id":317,"kind":4,"targetType":22,"targetId":318,"subtype":4,"typeLabel":24,"title":319,"subtitle":26,"summary":320,"coverUrl":321,"badgeText":29,"href":322,"sourceName":26,"meta":323,"metrics":324,"tags":325,"operationItemId":326,"sortOrder":327,"targetCode":38,"resolved":39},"NEWS_ARTICLE:915",915,"codex对接智谱coding plan，我发现连cc switch也不用了","背景 大家好，我是逐日。周末在家里电脑折腾，本来还是按惯例在弄codex--》cc switch--》智谱，翻官方文档的过程中，发现智谱好像已经原生支持了openai的response接口。 试了试，连cc swtich也省了，感兴趣的可以试试。 智谱官方提供的产品 产品形态，我现在理解有3种： a","https:\u002F\u002Fdump-1252523945.cos.ap-shanghai.myqcloud.com\u002Fimg\u002Fimage-20260830133209133.png","\u002Fnews\u002F915",[32,85],{},[85],1322482,12,{"id":329,"kind":4,"targetType":22,"targetId":330,"subtype":4,"typeLabel":24,"title":331,"subtitle":26,"summary":332,"coverUrl":333,"badgeText":29,"href":334,"sourceName":26,"meta":335,"metrics":336,"tags":337,"operationItemId":338,"sortOrder":339,"targetCode":38,"resolved":39},"NEWS_ARTICLE:914",914,"\"Mem2Evolve: Towards Self-Evolving Agents via Co-Evolutionary Capability Expansion and Experience Distillation\" 论文笔记","北航的工作 Mem2Evolve 发表于 ACL 2026，提出了一种全新的智能体自我进化范式，把"能力扩展"与"经验蒸馏"两个过程耦合起来，实现更稳定、更高效的智能体持续进化 背景 LLM 智能体在各类应用中已取得显著成功，研究者们不再满足于静态的任务特定系","https:\u002F\u002Fimg2024.cnblogs.com\u002Fblog\u002F3614753\u002F202608\u002F3614753-20260830145307727-30102612.png","\u002Fnews\u002F914",[32,85],{},[85],1322483,13,{"id":341,"kind":4,"targetType":22,"targetId":342,"subtype":4,"typeLabel":24,"title":343,"subtitle":26,"summary":344,"coverUrl":345,"badgeText":29,"href":346,"sourceName":26,"meta":347,"metrics":348,"tags":349,"operationItemId":350,"sortOrder":351,"targetCode":38,"resolved":39},"NEWS_ARTICLE:912",912,"面向 .NET 开发者的 RocketMQ 入门指南（四）：classic Remoting 客户端","目录前言Remoting 客户端职责一次请求的闭环Producer 发送流程消费模型的两个层次应用接口：Push 或 LitePull从 Pull 到 POPPull：Queue 与 OffsetOffset 水位Push：客户端负责接收与调度缓存与背压LitePull：后台预取与应用主动读取预取与","https:\u002F\u002Fimg2024.cnblogs.com\u002Fblog\u002F1201123\u002F202608\u002F1201123-20260830192359276-1598561466.png","\u002Fnews\u002F912",[32],{},[],1322485,15,{"id":353,"kind":4,"targetType":22,"targetId":354,"subtype":4,"typeLabel":24,"title":355,"subtitle":26,"summary":356,"coverUrl":357,"badgeText":29,"href":358,"sourceName":26,"meta":359,"metrics":361,"tags":362,"operationItemId":363,"sortOrder":364,"targetCode":38,"resolved":39},"NEWS_ARTICLE:911",911,"k8s安装metrics-server，以及相关报错排查","背景：有监控node节点pod的CPU等资源，实现自动扩缩容的需要下，需要安装metrics-server k8s版本：1.30.14（kubadm安装） 对应metrics-server版本：0.8x（本文使用0.8.1） 1、下载部署清单 wget https:\u002F\u002Fgithub.com\u002Fkube","https:\u002F\u002Fimg2024.cnblogs.com\u002Fblog\u002F1384371\u002F202608\u002F1384371-20260830185204348-546969162.png","\u002Fnews\u002F911",[32,360],"云原生与运维",{},[360],1322486,16,{"id":366,"kind":4,"targetType":22,"targetId":367,"subtype":4,"typeLabel":24,"title":368,"subtitle":26,"summary":369,"coverUrl":370,"badgeText":29,"href":371,"sourceName":26,"meta":372,"metrics":373,"tags":374,"operationItemId":375,"sortOrder":376,"targetCode":38,"resolved":39},"NEWS_ARTICLE:910",910,"MHS 三部曲（上）：别急着叫它「物理 MCP」——Anthropic 到底发布了什么","MHS 的本质不是'AI 控制硬件'，而是三件事的组合：窄腰接口（read\u002Fwrite）降低接入门槛，设备语义（状态\u002F操作\u002F标签）让机器能自我介绍，分层执行让模型待在该待的位置——实时控制环之外。","https:\u002F\u002Fimg2024.cnblogs.com\u002Fblog\u002F510\u002F202608\u002F510-20260830153745229-1536981088.jpg","\u002Fnews\u002F910",[32,85],{},[85],1322487,17,[378,383,388,393,398,403],{"rankNo":315,"score":91,"trendType":92,"trendValue":93,"reason":245,"card":379},{"id":305,"kind":4,"targetType":22,"targetId":306,"subtype":4,"typeLabel":24,"title":307,"subtitle":26,"summary":308,"coverUrl":309,"badgeText":29,"href":310,"sourceName":26,"meta":380,"metrics":381,"tags":382,"operationItemId":314,"sortOrder":315,"targetCode":38,"resolved":39},[32],{},[],{"rankNo":327,"score":91,"trendType":92,"trendValue":93,"reason":245,"card":384},{"id":317,"kind":4,"targetType":22,"targetId":318,"subtype":4,"typeLabel":24,"title":319,"subtitle":26,"summary":320,"coverUrl":321,"badgeText":29,"href":322,"sourceName":26,"meta":385,"metrics":386,"tags":387,"operationItemId":326,"sortOrder":327,"targetCode":38,"resolved":39},[32,85],{},[85],{"rankNo":339,"score":91,"trendType":92,"trendValue":93,"reason":245,"card":389},{"id":329,"kind":4,"targetType":22,"targetId":330,"subtype":4,"typeLabel":24,"title":331,"subtitle":26,"summary":332,"coverUrl":333,"badgeText":29,"href":334,"sourceName":26,"meta":390,"metrics":391,"tags":392,"operationItemId":338,"sortOrder":339,"targetCode":38,"resolved":39},[32,85],{},[85],{"rankNo":351,"score":91,"trendType":92,"trendValue":93,"reason":245,"card":394},{"id":341,"kind":4,"targetType":22,"targetId":342,"subtype":4,"typeLabel":24,"title":343,"subtitle":26,"summary":344,"coverUrl":345,"badgeText":29,"href":346,"sourceName":26,"meta":395,"metrics":396,"tags":397,"operationItemId":350,"sortOrder":351,"targetCode":38,"resolved":39},[32],{},[],{"rankNo":364,"score":91,"trendType":92,"trendValue":93,"reason":245,"card":399},{"id":353,"kind":4,"targetType":22,"targetId":354,"subtype":4,"typeLabel":24,"title":355,"subtitle":26,"summary":356,"coverUrl":357,"badgeText":29,"href":358,"sourceName":26,"meta":400,"metrics":401,"tags":402,"operationItemId":363,"sortOrder":364,"targetCode":38,"resolved":39},[32,360],{},[360],{"rankNo":376,"score":91,"trendType":92,"trendValue":93,"reason":245,"card":404},{"id":366,"kind":4,"targetType":22,"targetId":367,"subtype":4,"typeLabel":24,"title":368,"subtitle":26,"summary":369,"coverUrl":370,"badgeText":29,"href":371,"sourceName":26,"meta":405,"metrics":406,"tags":407,"operationItemId":375,"sortOrder":376,"targetCode":38,"resolved":39},[32,85],{},[85],{"blockCode":409,"blockType":410,"layoutType":411,"title":412,"subtitle":413,"itemLimit":124,"moreUrl":14,"cardVariant":414,"themeVariant":302,"rankCode":415,"rankName":416,"cards":417,"rankEntries":499},"news_hot_rail","RANK","RANK_RAIL","热度正在上升","最近一周的综合热点","NEWS_HOT_RAIL","news_hot","资讯综合热点榜",[418,428,438,443,448,458,463,473,478,488],{"id":419,"kind":4,"targetType":22,"targetId":420,"subtype":4,"typeLabel":24,"title":421,"subtitle":26,"summary":422,"coverUrl":29,"badgeText":29,"href":423,"sourceName":26,"meta":424,"metrics":425,"tags":426,"operationItemId":427,"sortOrder":76,"targetCode":38,"resolved":39},"NEWS_ARTICLE:860",860,"3句话，AI给我生成了一个galgame","demo和源码 demo地址 不支持移动端，可能需要1分钟的加载 demo仓库 plugin plugin我之前弄的时候很容易碰上路径问题，导致读取不到项目根目录下setting，如果要用的话得从Marketplace中获取后手工拷贝 agent & skills & script","\u002Fnews\u002F860",[32],{},[],1322418,{"id":429,"kind":4,"targetType":22,"targetId":430,"subtype":4,"typeLabel":24,"title":431,"subtitle":26,"summary":432,"coverUrl":29,"badgeText":29,"href":433,"sourceName":26,"meta":434,"metrics":435,"tags":436,"operationItemId":437,"sortOrder":13,"targetCode":38,"resolved":39},"NEWS_ARTICLE:859",859,"DDIA(六):编码格式的取舍与JSON战报体积优化方案","DDIA(六):编码格式的取舍与JSON战报体积优化方案 好家伙, 最近在研究战报的数据结构和体积压缩:于是去翻 DDIA 第 4 章,正好整章讲的就是这件事: 数据要跨进程、跨服务、落盘存档, 就得从内存结构体变成字节序列——这就是编码. 编码格式怎么选,直接影响包体大小、解析开销,以及未来改字段","\u002Fnews\u002F859",[32],{},[],1322419,{"id":41,"kind":4,"targetType":22,"targetId":42,"subtype":4,"typeLabel":24,"title":43,"subtitle":26,"summary":44,"coverUrl":45,"badgeText":29,"href":46,"sourceName":26,"meta":439,"metrics":440,"tags":441,"operationItemId":442,"sortOrder":196,"targetCode":38,"resolved":39},[32],{},[],1322420,{"id":53,"kind":4,"targetType":22,"targetId":54,"subtype":4,"typeLabel":24,"title":55,"subtitle":26,"summary":56,"coverUrl":57,"badgeText":29,"href":58,"sourceName":26,"meta":444,"metrics":445,"tags":446,"operationItemId":447,"sortOrder":208,"targetCode":38,"resolved":39},[32],{},[],1322421,{"id":449,"kind":4,"targetType":22,"targetId":450,"subtype":4,"typeLabel":24,"title":451,"subtitle":26,"summary":452,"coverUrl":29,"badgeText":29,"href":453,"sourceName":26,"meta":454,"metrics":455,"tags":456,"operationItemId":457,"sortOrder":220,"targetCode":38,"resolved":39},"NEWS_ARTICLE:856",856,"AI生产力：从效率到工作流重构","当初学专业的笔记还在吗？相信大部分人都没保留，而能力在不断积累和更新，关注和思考的角度在变化，在AI蒸馏我们的同时。也可以用AI的能力，去蒸馏一下魔幻的职场。","\u002Fnews\u002F856",[32,85],{},[85],1322422,{"id":65,"kind":4,"targetType":22,"targetId":66,"subtype":4,"typeLabel":24,"title":67,"subtitle":26,"summary":68,"coverUrl":69,"badgeText":29,"href":70,"sourceName":26,"meta":459,"metrics":460,"tags":461,"operationItemId":462,"sortOrder":231,"targetCode":38,"resolved":39},[32,72],{},[72],1322423,{"id":464,"kind":4,"targetType":22,"targetId":465,"subtype":4,"typeLabel":24,"title":466,"subtitle":26,"summary":467,"coverUrl":29,"badgeText":29,"href":468,"sourceName":26,"meta":469,"metrics":470,"tags":471,"operationItemId":472,"sortOrder":124,"targetCode":38,"resolved":39},"NEWS_ARTICLE:854",854,"分治：序列分治（CDQ）与点分治","分治：序列分治（CDQ）与点分治 一、分治思想概述 分治（Divide and Conquer）是算法设计中最核心的思想之一。它的基本策略是： 分（Divide）：将原问题划分为规模更小的子问题。 治（Conquer）：递归地求解子问题（若子问题足够小则直接求解）。 合（Combine）：将子问题的","\u002Fnews\u002F854",[32,33],{},[33],1322424,{"id":78,"kind":4,"targetType":22,"targetId":79,"subtype":4,"typeLabel":24,"title":80,"subtitle":26,"summary":81,"coverUrl":82,"badgeText":29,"href":83,"sourceName":26,"meta":474,"metrics":475,"tags":476,"operationItemId":477,"sortOrder":315,"targetCode":38,"resolved":39},[32,85],{},[85],1322425,{"id":479,"kind":4,"targetType":22,"targetId":480,"subtype":4,"typeLabel":24,"title":481,"subtitle":26,"summary":482,"coverUrl":29,"badgeText":29,"href":483,"sourceName":26,"meta":484,"metrics":485,"tags":486,"operationItemId":487,"sortOrder":327,"targetCode":38,"resolved":39},"NEWS_ARTICLE:852",852,"多智能体编排配置的工程化实践:权限、成本、上下文与可复现交付","一、权限最小化:两道白名单 第一道:任务工具白名单(编排层)。 主智能体(编排者)通过 permission 声明:Task 工具默认拒绝一切派发,只放行 10 个具名子智能体(planner、deep-worker、oracle、reviewer、consultant、ui-builder、exp","\u002Fnews\u002F852",[32],{},[],1322426,{"id":489,"kind":4,"targetType":22,"targetId":490,"subtype":4,"typeLabel":24,"title":491,"subtitle":26,"summary":492,"coverUrl":493,"badgeText":29,"href":494,"sourceName":26,"meta":495,"metrics":496,"tags":497,"operationItemId":498,"sortOrder":339,"targetCode":38,"resolved":39},"NEWS_ARTICLE:851",851,"多模态知识库，打通文本、图像与视频资源的协同实践与应用","本文介绍一个支持文本、图像、视频三类资源统一入库与检索的跨模态知识库系统。三类资源经统一向量化后进入同一向量空间，用户只需输入一句自然语言，即可一次性召回三种模态的相关资源，并按类型直观展示。","https:\u002F\u002Fimg2024.cnblogs.com\u002Fblog\u002F279374\u002F202609\u002F279374-20260902092931266-1302274433.png","\u002Fnews\u002F851",[32],{},[],1322427,[500,505,510,515,520,525,530,535,540,545],{"rankNo":76,"score":91,"trendType":92,"trendValue":93,"reason":94,"card":501},{"id":419,"kind":4,"targetType":22,"targetId":420,"subtype":4,"typeLabel":24,"title":421,"subtitle":26,"summary":422,"coverUrl":29,"badgeText":29,"href":423,"sourceName":26,"meta":502,"metrics":503,"tags":504,"operationItemId":427,"sortOrder":76,"targetCode":38,"resolved":39},[32],{},[],{"rankNo":13,"score":91,"trendType":92,"trendValue":93,"reason":94,"card":506},{"id":429,"kind":4,"targetType":22,"targetId":430,"subtype":4,"typeLabel":24,"title":431,"subtitle":26,"summary":432,"coverUrl":29,"badgeText":29,"href":433,"sourceName":26,"meta":507,"metrics":508,"tags":509,"operationItemId":437,"sortOrder":13,"targetCode":38,"resolved":39},[32],{},[],{"rankNo":196,"score":91,"trendType":92,"trendValue":93,"reason":94,"card":511},{"id":41,"kind":4,"targetType":22,"targetId":42,"subtype":4,"typeLabel":24,"title":43,"subtitle":26,"summary":44,"coverUrl":45,"badgeText":29,"href":46,"sourceName":26,"meta":512,"metrics":513,"tags":514,"operationItemId":442,"sortOrder":196,"targetCode":38,"resolved":39},[32],{},[],{"rankNo":208,"score":91,"trendType":92,"trendValue":93,"reason":94,"card":516},{"id":53,"kind":4,"targetType":22,"targetId":54,"subtype":4,"typeLabel":24,"title":55,"subtitle":26,"summary":56,"coverUrl":57,"badgeText":29,"href":58,"sourceName":26,"meta":517,"metrics":518,"tags":519,"operationItemId":447,"sortOrder":208,"targetCode":38,"resolved":39},[32],{},[],{"rankNo":220,"score":91,"trendType":92,"trendValue":93,"reason":94,"card":521},{"id":449,"kind":4,"targetType":22,"targetId":450,"subtype":4,"typeLabel":24,"title":451,"subtitle":26,"summary":452,"coverUrl":29,"badgeText":29,"href":453,"sourceName":26,"meta":522,"metrics":523,"tags":524,"operationItemId":457,"sortOrder":220,"targetCode":38,"resolved":39},[32,85],{},[85],{"rankNo":231,"score":91,"trendType":92,"trendValue":93,"reason":94,"card":526},{"id":65,"kind":4,"targetType":22,"targetId":66,"subtype":4,"typeLabel":24,"title":67,"subtitle":26,"summary":68,"coverUrl":69,"badgeText":29,"href":70,"sourceName":26,"meta":527,"metrics":528,"tags":529,"operationItemId":462,"sortOrder":231,"targetCode":38,"resolved":39},[32,72],{},[72],{"rankNo":124,"score":91,"trendType":92,"trendValue":93,"reason":94,"card":531},{"id":464,"kind":4,"targetType":22,"targetId":465,"subtype":4,"typeLabel":24,"title":466,"subtitle":26,"summary":467,"coverUrl":29,"badgeText":29,"href":468,"sourceName":26,"meta":532,"metrics":533,"tags":534,"operationItemId":472,"sortOrder":124,"targetCode":38,"resolved":39},[32,33],{},[33],{"rankNo":315,"score":91,"trendType":92,"trendValue":93,"reason":94,"card":536},{"id":78,"kind":4,"targetType":22,"targetId":79,"subtype":4,"typeLabel":24,"title":80,"subtitle":26,"summary":81,"coverUrl":82,"badgeText":29,"href":83,"sourceName":26,"meta":537,"metrics":538,"tags":539,"operationItemId":477,"sortOrder":315,"targetCode":38,"resolved":39},[32,85],{},[85],{"rankNo":327,"score":91,"trendType":92,"trendValue":93,"reason":94,"card":541},{"id":479,"kind":4,"targetType":22,"targetId":480,"subtype":4,"typeLabel":24,"title":481,"subtitle":26,"summary":482,"coverUrl":29,"badgeText":29,"href":483,"sourceName":26,"meta":542,"metrics":543,"tags":544,"operationItemId":487,"sortOrder":327,"targetCode":38,"resolved":39},[32],{},[],{"rankNo":339,"score":91,"trendType":92,"trendValue":93,"reason":94,"card":546},{"id":489,"kind":4,"targetType":22,"targetId":490,"subtype":4,"typeLabel":24,"title":491,"subtitle":26,"summary":492,"coverUrl":493,"badgeText":29,"href":494,"sourceName":26,"meta":547,"metrics":548,"tags":549,"operationItemId":498,"sortOrder":339,"targetCode":38,"resolved":39},[32],{},[],{"items":551,"total":686,"current":37,"size":687,"pages":13},[552,559,565,572,579,586,592,599,605,612,619,625,633,640,646,652,659,667,673,680],{"id":553,"dedupeKey":553,"kind":4,"title":554,"summary":555,"image":370,"href":556,"meta":557,"badge":24,"author":26,"tags":558},"NEWS_ARTICLE:827","MHS 三部曲（下）：谁允许 AI 行动？——权力、合规与中国厂商的答卷","我们总在等待一个像 ChatGPT 那样的机器人时刻。但具身智能真正的拐点，也许先发生在更不起眼的地方：一台陌生设备，第一次能把自己的能力、状态和边界完整告诉 AI；一个 Agent，第一次能把试出来的经验固化成可验证、可复用的机器技能。","\u002Fnews\u002F827","2026 · 人工智能",[85],{"id":560,"dedupeKey":560,"kind":4,"title":561,"summary":562,"image":29,"href":563,"meta":32,"badge":24,"author":26,"tags":564},"NEWS_ARTICLE:829","我不会美工，用 WorkBuddy 1 分钟做出 4 张海报","先说结果：下面这 4 张海报，从我发出指令到拿到图片，大约 1 分钟。 我不会美工，也没有打开 Photoshop。用的工具，是我之前用 WorkBuddy 做的一个海报生成器。这个生成器本身也没花几分钟，具体开发过程我在上一篇文章里写过。 这次我把海报生成器的网址、产品截图和要求一起发给 Work","\u002Fnews\u002F829",[],{"id":566,"dedupeKey":566,"kind":4,"title":567,"summary":568,"image":569,"href":570,"meta":32,"badge":24,"author":26,"tags":571},"NEWS_ARTICLE:828","一个人抵一个团队的时代，企业级 AI Agent 还需要做什么？","对于企业而言，真正需要的，不是一个“会聊天”的 Agent，而是一套能被多用户、多场景复用，且权限清晰、过程可审计、成本可控制的 Agent 能力体系。","https:\u002F\u002Fimg2024.cnblogs.com\u002Fblog\u002F2232255\u002F202609\u002F2232255-20260902173524728-659996478.png","\u002Fnews\u002F828",[],{"id":573,"dedupeKey":573,"kind":4,"title":574,"summary":575,"image":29,"href":576,"meta":577,"badge":24,"author":26,"tags":578},"NEWS_ARTICLE:832","用 runtime-async 写一个支持 async\u002Fawait 的轻量脚本引擎","起因：一个好奇 事情的开头很简单。 给 .NET 做过动态脚本的人大概都碰到过同一堵墙：表达式树也好、Reflection.Emit 也好，都很难支持 async\u002Fawait。原因不在语法，而在 await 的实现方式——C# 编译器要为每个 async 方法生成一个状态机结构体，把方法体切成若干片","\u002Fnews\u002F832","2026 · 软件开发",[72],{"id":580,"dedupeKey":580,"kind":4,"title":581,"summary":582,"image":583,"href":584,"meta":32,"badge":24,"author":26,"tags":585},"NEWS_ARTICLE:831","Agent Sandbox 规模化：JuiceFS 探索与实践","过去一段时间，我们在和云厂商以及 Agent 团队推进 Sandbox 落地时，除了要解决数据如何进入 Sandbox，还需要考虑任务所需的数据和执行过程中产生的数据，如何跨任务、跨阶段持续使用。Sandbox 可以随任务快速创建和销毁，但数据往往需要继续保留、共享和流转。 当 Sandbox 并发","https:\u002F\u002Fimg2024.cnblogs.com\u002Fblog\u002F2544292\u002F202609\u002F2544292-20260902171045357-1907499098.png","\u002Fnews\u002F831",[],{"id":587,"dedupeKey":587,"kind":4,"title":588,"summary":589,"image":29,"href":590,"meta":32,"badge":24,"author":26,"tags":591},"NEWS_ARTICLE:830","百智云长期记忆服务：长期记忆与上下文窗口的区别","很多人在接触 AI 长期记忆时，第一反应是：「这不就是把聊天记录存起来吗？」其实这是一个很常见的误解。今天我们把「长期记忆」和「上下文窗口」这两个概念彻底讲清楚。 一、什么是上下文窗口 上下文窗口（Context Window）是模型在单次推理时能「看到」的文本上限。它像一个临时的工作台：模型只能处","\u002Fnews\u002F830",[],{"id":593,"dedupeKey":593,"kind":4,"title":594,"summary":595,"image":596,"href":597,"meta":557,"badge":24,"author":26,"tags":598},"NEWS_ARTICLE:835","体验向量数据库 qdrant","作者:张富春(ahfuzhang)，转载时请注明作者和引用链接，谢谢！ cnblogs博客 zhihu Github 公众号:一本正经的瞎扯 背景 为了早点搞懂公司的百万行 C# 的祖传代码，我想在缺乏文档、缺乏帮手的情况下，先建立一个 企业知识库 来快速帮我理清头绪。 一开始，我是打算以 weav","https:\u002F\u002Fimg2022.cnblogs.com\u002Fblog\u002F1457949\u002F202202\u002F1457949-20220216153819145-1193738712.png","\u002Fnews\u002F835",[85],{"id":600,"dedupeKey":600,"kind":4,"title":601,"summary":602,"image":29,"href":603,"meta":32,"badge":24,"author":26,"tags":604},"NEWS_ARTICLE:833","如何利用AI技术实现漫画翻译","漫画作为图文结合的特色文化载体，承载着各国潮流文化与人文故事，但跨语言的文字壁垒，长期制约着漫画的传播与交流。传统漫画翻译高度依赖人工操作，需要人工框选文字、擦除原图文字、翻译文案、排版适配画风，流程繁琐、耗时费力，且极易出现排版错乱、画风违和、语义偏差等问题。 随着深度学习、计算机视觉与大模型技术","\u002Fnews\u002F833",[],{"id":606,"dedupeKey":606,"kind":4,"title":607,"summary":608,"image":609,"href":610,"meta":32,"badge":24,"author":26,"tags":611},"NEWS_ARTICLE:834","用 Apache SeaTunnel 同步 DynamoDB 到 Redis，一个配置文件就够","作者：Ricardo Ferreira 编译：Debra Chen 假设你把客户数据存储在 Amazon DynamoDB 中，作为应用的唯一数据源（single source of truth）。但如今你需要同样的数据放在 Redis 里，进行快速缓存、实时分析，或是为向量搜索推荐引擎赋能。举例来","https:\u002F\u002Fimg2024.cnblogs.com\u002Fblog\u002F3195851\u002F202609\u002F3195851-20260902145155710-759131565.jpg","\u002Fnews\u002F834",[],{"id":613,"dedupeKey":613,"kind":4,"title":614,"summary":615,"image":616,"href":617,"meta":32,"badge":24,"author":26,"tags":618},"NEWS_ARTICLE:836","自研一个团队文件共享服务是什么体验：文件共享服务的设计与实现解析","在研发与运维的日常里，"传个文件"这件看起来极小的事，往往会被无限放大——日志包在 A 同事的机器上、安装包在 B 同事的桌面上、配置脚本散在各个项目目录里；好不容易发到群里，过两天就过期了；更怕的是有人手滑把还在用的关键文件给删了。 自研了一套文件共享服务，把这些问题用一个网页","https:\u002F\u002Fimg2024.cnblogs.com\u002Fblog\u002F2198843\u002F202609\u002F2198843-20260902142051309-1510220729.png","\u002Fnews\u002F836",[],{"id":620,"dedupeKey":620,"kind":4,"title":621,"summary":622,"image":29,"href":623,"meta":32,"badge":24,"author":26,"tags":624},"NEWS_ARTICLE:837","很多B2B的生态合作，都是在带薪休假","有一种 B2B 生态工作，我把它叫作“带薪休假式生态”。 公司搭建了生态岗位，或者要求区域销售顺便拓展伙伴，管理层给了一个结果目标： 发展多少家伙伴，带来多少条线索，创造多少商机。 但应该找谁、为什么找、先找谁、每家公司投入多少精力、什么情况下继续、什么情况下停止，没有人说得清楚。 于是，生态人员只","\u002Fnews\u002F837",[],{"id":626,"dedupeKey":626,"kind":4,"title":627,"summary":628,"image":629,"href":630,"meta":631,"badge":24,"author":26,"tags":632},"NEWS_ARTICLE:838","AI 学习笔记：LLM 的微调实验","title: LLM 的微调实验 author: 凌杰 date: 2026-08-21 tags: LoRA, LLaMA-Factory, Qwen categories: 人工智能 [!NOTE] 笔记说明 这篇笔记对应的是《[[关于 AI 的学习路线图]]》一文中所规划的第三个学习阶段。其中","https:\u002F\u002Fimg2024.cnblogs.com\u002Fblog\u002F691082\u002F202609\u002F691082-20260902122904258-1285666483.png","\u002Fnews\u002F838","2026 · 综合技术",[33],{"id":634,"dedupeKey":634,"kind":4,"title":635,"summary":636,"image":29,"href":637,"meta":638,"badge":24,"author":26,"tags":639},"NEWS_ARTICLE:839","Nginx 1.30.4 编译安装与参数调优","Nginx 是一款轻量级、高性能、高并发的开源Web服务器、反向代理与负载均衡软件，凭借低资源占用、高稳定性的核心优势，成为当下互联网行业主流的服务部署组件，广泛应用于各类网站及业务系统的Web服务、反向代理、负载均衡场景。其基于宽松的类BSD开源协议发布，可免费商用。本文将完整展示 Nginx 1","\u002Fnews\u002F839","2026 · 云原生与运维",[360],{"id":641,"dedupeKey":641,"kind":4,"title":642,"summary":643,"image":29,"href":644,"meta":32,"badge":24,"author":26,"tags":645},"NEWS_ARTICLE:840","MiniPdf Office 转 PDF 支持 Rust 原生实现：3 MB 单文件部署、低内存","MiniPdf Office 转 PDF 支持 Rust 原生实现：3 MB 单文件部署、低内存 github link : https:\u002F\u002Fgithub.com\u002Fmini-software\u002FMiniPdf 有时候只想在 API、定时任务或 Container 里，把一份 Excel、Word 转成","\u002Fnews\u002F840",[],{"id":647,"dedupeKey":647,"kind":4,"title":648,"summary":649,"image":29,"href":650,"meta":32,"badge":24,"author":26,"tags":651},"NEWS_ARTICLE:841","AKF扩展立方体和AKF可用性立方体","很多人知道AKF扩展立方体是从《架构即未来》这本书开始。实际上akfpartners官方写过4篇关于AKF扩展立方体的文章，还有一篇介绍AKF可用性立方体。akfpartners官方在高可用、扩展性方面有很多专业技术文章，建议有空就翻翻看。 AKF扩展立方体和AKF可用性立方体都介绍了X轴、Y轴和Z","\u002Fnews\u002F841",[],{"id":653,"dedupeKey":653,"kind":4,"title":654,"summary":655,"image":656,"href":657,"meta":631,"badge":24,"author":26,"tags":658},"NEWS_ARTICLE:843","介绍一下常用的Token鉴权方案","本文梳理 Session‑Cookie、JWT、OAuth2.0、SSO 主流 Token 鉴权方案，对比各方案适用场景。重点讲解生产级 JWT 双 Token 架构，给出 RS256 非对称加密、Redis 黑名单、网关统一鉴权等 Java 实战代码。剖析 JWT 注销、并发刷新、令牌泄露等落地痛","https:\u002F\u002Fimg2024.cnblogs.com\u002Fblog\u002F739056\u002F202609\u002F739056-20260902114155245-1924311389.png","\u002Fnews\u002F843",[33],{"id":660,"dedupeKey":660,"kind":4,"title":661,"summary":662,"image":663,"href":664,"meta":665,"badge":24,"author":26,"tags":666},"NEWS_ARTICLE:842","几何分布：从“等一个结果”开始","你有没有过这样的经历？ 在游戏里抽卡，抽了多少次才终于出货？ 站在路边等公交车，第几辆来的才是你要坐的那一路？ 给客户打电话，打到第几个才终于接通？ 刷短视频时，刷了多少条才刷到自己想看的内容？ 这些场景背后，都藏着一个共同的概率模型——几何分布（Geometric Distribution）。 几","https:\u002F\u002Fimg2024.cnblogs.com\u002Fblog\u002F83005\u002F202609\u002F83005-20260902112922617-1277310049.png","\u002Fnews\u002F842","2026 · 数据与架构",[148],{"id":668,"dedupeKey":668,"kind":4,"title":669,"summary":670,"image":29,"href":671,"meta":32,"badge":24,"author":26,"tags":672},"NEWS_ARTICLE:844","数据库上K8s，运维为什么反而更累了？从StatefulSet到Operator的复盘","各位，我是老张。今天从一次实战出发，往下拆一层。 上个月陪一个朋友看他们的容器化改造。业务服务全部迁上K8s了，跑得挺顺，唯独数据库卡在原地。那哥们跟我吐槽：老张，代码都能容器化，怎么一到了数据库这儿，运维反而比以前更累了？我上去看了眼环境，果然，数据库还是手工搭的，跟K8s一点关系没有，扩容要半夜","\u002Fnews\u002F844",[],{"id":674,"dedupeKey":674,"kind":4,"title":675,"summary":676,"image":677,"href":678,"meta":32,"badge":24,"author":26,"tags":679},"NEWS_ARTICLE:846","好的 AI 办公应用，不是聊天框，而是能跑完流程","AI 办公应用的分水岭，不是聊天框写得多像人，而是能不能接入真实工作环境，把任务从输入推进到可验证结果。它需要本地执行、企业系统连接、审批审计、跨设备协作、低噪提醒和历史会话检索。","https:\u002F\u002Fimg2024.cnblogs.com\u002Fblog\u002F3798513\u002F202609\u002F3798513-20260902110101957-987446244.png","\u002Fnews\u002F846",[],{"id":681,"dedupeKey":681,"kind":4,"title":682,"summary":683,"image":29,"href":684,"meta":32,"badge":24,"author":26,"tags":685},"NEWS_ARTICLE:845","题解：P10027 梦境世界","题目。 考虑 DP。由于撤销操作会导致环路，所以不能直接 DP。 我们发现对于撤销操作肯定会回到某一个点，也就是存在一个点，走了若干步再撤回再走。这样就好转移了。 定义 \\(G_{i,j,k}\\) 代表从 \\((i,j)\\) 出发，拥有 \\(k\\) 瓶撤销药，到达终点的方案数。 同时，我们定义 \\(","\u002Fnews\u002F845",[],100,20]