[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"portal-settings:stajic:zh":3,"public-menus:all":38,"post:how-far-is-an-ai-claim-from-the-evidence-evidence-distance:zh":221,"related:post:how-far-is-an-ai-claim-from-the-evidence-evidence-distance:zh:1":1558},{"statusCode":4,"data":5,"message":37},200,{"tenantId":6,"lang":7,"defaultLang":8,"siteUrl":9,"contactEmail":10,"brandName":11,"logoUrl":12,"siteName":11,"siteDescription":13,"ogImage":10,"robotsIndex":14,"socialLinks":10,"reservedSlugs":10,"seoPolicy":15},"stajic","zh","de","https:\u002F\u002Fstajic.de",null,"Stajic Platform","\u002FLogo_Planet.svg","Stajic Portal",true,{"branding":16,"relatedContent":17,"crossDomainLinks":18},{"logoUrl":12},{"enabled":14},[19,22,25,28,31,34],{"url":20,"label":21,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Ffigure.rocks","figure.rocks",{"url":23,"label":24,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Floving.rocks","loving.rocks",{"url":26,"label":27,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.com","bazify.com",{"url":29,"label":30,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.de","bazify.de",{"url":32,"label":33,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.at","bazify.at",{"url":35,"label":36,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.ba","bazify.ba","Portal settings resolved",[39,45],{"id":40,"name":41,"location":42,"isActive":14,"isDefault":43,"items":44},1,"main-navigation","header",false,[],{"id":46,"name":47,"location":48,"isActive":14,"isDefault":14,"items":49},4,"main-menu","sidebar",[50,66,79,93,103,118,133],{"id":51,"title":52,"url":60,"target":61,"icon":62,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":64,"portfolioId":10,"children":65},"item-18",{"de":53,"en":54,"es":55,"fr":56,"it":54,"ru":57,"sr":58,"zh":59},"Startseite","Home","Inicio","Accueil","Главная","Почетна","首页","\u002Ffull-stack-web-developer-munich-performance-seo-and-maintainable-builds","_self","i-lucide-home","page",111,[],{"id":67,"title":68,"url":75,"target":61,"icon":76,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":77,"portfolioId":10,"children":78},"item-22",{"de":69,"en":69,"es":70,"fr":69,"it":71,"ru":72,"sr":73,"zh":74},"Vision","Visión","Visione","Видение","Визија","想象","\u002Fueber-uns-webdesign-muenchen-webaplikation","i-lucide-eye",113,[],{"id":80,"title":81,"url":89,"target":61,"icon":90,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":91,"portfolioId":10,"children":92},"item-19",{"de":82,"en":83,"es":84,"fr":83,"it":85,"ru":86,"sr":87,"zh":88},"Leistungen","Services","Servicios","Servizi","Услуги","Услуге","服务","\u002Fservices-dienstleistungen-muenchen","i-lucide-wrench",116,[],{"id":94,"title":95,"url":99,"target":61,"icon":100,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":101,"portfolioId":10,"children":102},"item-23",{"de":96,"en":96,"es":96,"fr":96,"it":96,"ru":97,"sr":97,"zh":98},"Blog","Блог","博客","\u002Fblog","i-lucide-book-open",112,[],{"id":104,"title":105,"url":114,"target":61,"icon":115,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":116,"portfolioId":10,"children":117},"item-32",{"de":106,"en":107,"es":108,"fr":109,"it":110,"ru":111,"sr":112,"zh":113},"Neue Technologien","New Technologies","Nuevas tecnologías","Nouvelles technologies","Nuove tecnologie","Новые технологии","Нове технологије","新技术！","\u002Fneue-webtechnologien","i-lucide-sparkles",122,[],{"id":119,"title":120,"url":129,"target":61,"icon":130,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":131,"portfolioId":10,"children":132},"item-20",{"de":121,"en":122,"es":123,"fr":124,"it":125,"ru":126,"sr":127,"zh":128},"Kontakt","Contact us!","Contacto","Contact","Contatto","Контакт","Контактирајте нас","联系我们！","\u002Fcontact","i-lucide-mail",115,[],{"id":134,"title":135,"url":144,"target":61,"icon":145,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":146,"portfolioId":10,"children":147},"item-21",{"de":136,"en":137,"es":138,"fr":139,"it":140,"ru":141,"sr":142,"zh":143},"Unsere Arbeit","Our Work","Nuestro trabajo","Nos réalisations","I nostri lavori","Наши работы","Наши радови","文件夹","\u002Fportfolio","i-lucide-briefcase",114,[148,161,175,191,206],{"id":149,"title":150,"url":144,"target":61,"icon":159,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":146,"portfolioId":10,"children":160},"item-24",{"de":151,"en":152,"es":153,"fr":154,"it":155,"ru":156,"sr":157,"zh":158},"Alle Projekte","All Projects","Todos los proyectos","Tous les projets","Tutti i progetti","Все проекты","Сви пројекти","所有项目","i-lucide-grid-3x3",[],{"id":162,"title":163,"url":171,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":174},"item-29",{"de":164,"en":165,"es":166,"fr":167,"it":168,"ru":169,"sr":170,"zh":143},"Local Roots, Global Reach","Local Roots - Global Reach","Empresa local ","Entreprise locale","Azienda locale","Местная компания","Локално предузеће глобално тржиште","\u002Fportfolio\u002Flocal-roots-global-reach-communication-media-systems-for-modern-business","i-lucide-globe-2","custom",[],{"id":176,"title":177,"url":186,"target":61,"icon":187,"isActive":14,"type":188,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":189,"children":190},"menu-2916d060-8681-482a-af9a-c107a8bc30d8",{"de":178,"en":179,"es":180,"fr":181,"it":182,"ru":183,"sr":184,"zh":185},"Von Rohdaten zu verifizierbarem KI-Kontext: Eine praktische Architektur für agentenbereite Informationen","From Raw Data to Verifiable AI Context:","De datos sin procesar a contexto de IA verificable: una arquitectura práctica para información lista para agentes","Des données brutes au contexte d'IA vérifiable : une architecture pratique pour des informations prêtes pour les agents","Dai dati grezzi a un contesto IA verificabile: Un'architettura pratica per informazioni pronte per gli agenti","От необработанных данных к верифицируемому контексту ИИ: практическая архитектура информации, готовой для агентов","Od sirovih podataka do proverljivog AI konteksta: Praktična arhitektura za informacije spremne za agente","从原始数据到可验证的 AI 上下文：面向智能体就绪信息的实用架构","\u002Fportfolio\u002Ffrom-raw-data-to-verifiable-ai-context-a-practical-architecture-for-agent-ready-information","i-lucide-database","portfolio",22,[],{"id":192,"title":193,"url":202,"target":61,"icon":203,"isActive":14,"type":188,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":204,"children":205},"menu-e2061cd7-ff53-49d5-a0e8-3d49090577c0",{"de":194,"en":195,"es":196,"fr":197,"it":198,"ru":199,"sr":200,"zh":201},"Präsenzplattformen — Unternehmenskommunikation","Presence Platforms — Enterprise Communication","Plataformas de presencia — Comunicación empresarial","Plateformes de présence — Communication d'entreprise","Presence Platforms — Comunicazione aziendale","Платформы присутствия — Корпоративные коммуникации","Presence platforme — Poslovna komunikacija","Presence 平台 — 企业通信","\u002Fportfolio\u002Fpresence-platforms-enterprise-communication-that-wins-without-competing","i-lucide-badge-percent",21,[],{"id":207,"title":208,"url":217,"target":61,"icon":218,"isActive":14,"type":188,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":219,"children":220},"menu-a781582e-e90d-418e-b8c9-b070bc37c37e",{"de":209,"en":210,"es":211,"fr":212,"it":213,"ru":214,"sr":215,"zh":216},"Maximierung des POS-ROI durch Konsistenz","Maximizing POS ROI Through Consistent","Maximizar el ROI del POS a través de constante","Maximiser le ROI du PDV grâce à la cohérence","Massimizzare il ROI del POS attraverso costante","Максимизация ROI POS за счет последовательного","Maksimizacija POS ROI-ja kroz dosledno","通过持续最大化 POS 投资回报率","\u002Fportfolio\u002Ffrom-vision-to-value-maximizing-pos-roi-through-consistent-ci-creation-and-strategic-branding","i-lucide-briefcase-business",11,[],{"statusCode":4,"data":222,"message":1557},{"id":223,"title":224,"slug":225,"content":226,"contentJson":227,"excerpt":741,"featuredImage":742,"featuredImageAlt":743,"featuredImageCaption":10,"featuredImageTitle":10,"featuredImageCopyright":10,"featuredImageAuthor":10,"featuredImageSourceUrl":10,"featuredImageLicense":10,"featuredImageIsAiGenerated":43,"status":744,"publishedAt":745,"createdAt":746,"updatedAt":747,"seoLocalePaths":748,"categories":757,"author":766,"translations":770},"497","AI主张与证据之间有多远？——证据距离","how-far-is-an-ai-claim-from-the-evidence-evidence-distance","{\"time\":1792097029426,\"blocks\":[{\"id\":\"fV7BD0fFRn\",\"type\":\"header\",\"data\":{\"text\":\"问题\",\"level\":2},\"tunes\":{}},{\"id\":\"tUBNklFmkB\",\"type\":\"paragraph\",\"data\":{\"text\":\"两个AI回答可能都引用了来源，但在这些来源对所提主张的支持有多直接上，仍可能大相径庭。我们该如何描述这种差异？\"},\"tunes\":{}},{\"id\":\"FwRn77XH9M\",\"type\":\"paragraph\",\"data\":{\"text\":\"本文使用术语证据距离来表示一项主张与支持它的可直接检验证据之间的距离。\"},\"tunes\":{}},{\"id\":\"YcbJYh_fW7\",\"type\":\"quote\",\"data\":{\"text\":\"证据距离是指将一项主张与能够直接验证它的证据分隔开的转换、假设和推理步骤的数量与强度。\",\"caption\":\"工作定义——此处引入的一个推理概念，并非标准化指标。\",\"alignment\":\"left\"},\"tunes\":{}},{\"id\":\"5KYvEA4v5o\",\"type\":\"tableOfContents\",\"data\":{\"title\":\"Contents\",\"minLevel\":2,\"maxLevel\":3},\"tunes\":{}},{\"id\":\"YxZslT77eK\",\"type\":\"header\",\"data\":{\"text\":\"这真正意味着什么\",\"level\":2},\"tunes\":{}},{\"id\":\"hZtJsMLV04\",\"type\":\"paragraph\",\"data\":{\"text\":\"引用证明某个来源被参考了。它并不自动证明该来源直接支持引用旁边的句子。\"},\"tunes\":{}},{\"id\":\"xcTtN7QaSQ\",\"type\":\"paragraph\",\"data\":{\"text\":\"在原始观察与最终AI回答之间，可能存在提取、解析、分块、检索、摘要、解释、聚合和推理。\"},\"tunes\":{}},{\"id\":\"wp6XTpoQAK\",\"type\":\"code\",\"data\":{\"code\":\"Primary evidence\\n   ↓\\nExtracted data\\n   ↓\\nRetrieved passage\\n   ↓\\nSummary\\n   ↓\\nInterpretation\\n   ↓\\nInference\\n   ↓\\nFinal claim\"},\"tunes\":{}},{\"id\":\"wRN6k2vvIy\",\"type\":\"paragraph\",\"data\":{\"text\":\"每一步都可能是合理的。但每一步也可能引入失真。\"},\"tunes\":{}},{\"id\":\"mo39IqnG40\",\"type\":\"header\",\"data\":{\"text\":\"最简单的例子\",\"level\":2},\"tunes\":{}},{\"id\":\"ysPGEKOYAD\",\"type\":\"paragraph\",\"data\":{\"text\":\"假设一个温度传感器报告38.2 °C。\"},\"tunes\":{}},{\"id\":\"AyD7F1En7R\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"主张\",\"与证据的关系\",\"示例性距离\"],[\"传感器报告38.2 °C。\",\"直接读数\",\"低\"],[\"系统运行温度比昨天更高。\",\"需要与昨天的读数进行比较\",\"中等\"],[\"冷却系统正在失效。\",\"需要因果推断和替代解释\",\"高\"]]},\"tunes\":{}},{\"id\":\"WmjjWyf_pZ\",\"type\":\"paragraph\",\"data\":{\"text\":\"原始传感器值直接支持第一个主张。它本身并不能证明第三个主张。\"},\"tunes\":{}},{\"id\":\"CzgUTnxkNa\",\"type\":\"code\",\"data\":{\"code\":\"Evidence exists\\n        ≠\\nEvidence directly supports every derived claim\"},\"tunes\":{}},{\"id\":\"Jrk0FmPH_u\",\"type\":\"header\",\"data\":{\"text\":\"这个例子在哪里不再适用\",\"level\":2},\"tunes\":{}},{\"id\":\"FCSg_sQyVr\",\"type\":\"paragraph\",\"data\":{\"text\":\"证据距离并不只是句子或处理步骤的计数。\"},\"tunes\":{}},{\"id\":\"Reo2XC69iu\",\"type\":\"paragraph\",\"data\":{\"text\":\"一个数学上有效的转换可能几乎完美地保留证据，而一个推测性解释可能造成大得多的认识论跳跃。\"},\"tunes\":{}},{\"id\":\"ixWx2M5On0\",\"type\":\"paragraph\",\"data\":{\"text\":\"因此，这个概念既包括转换距离，也包括推理距离。\"},\"tunes\":{}},{\"id\":\"eiC_FkuMl6\",\"type\":\"header\",\"data\":{\"text\":\"直接回答\",\"level\":2},\"tunes\":{}},{\"id\":\"iDIfyCXVdo\",\"type\":\"paragraph\",\"data\":{\"text\":\"当一项主张所依赖的转换、假设、较弱的来源或比底层证据直接支持的更强的推断越多时，证据距离就会增加。\"},\"tunes\":{}},{\"id\":\"7nDss8_HDy\",\"type\":\"code\",\"data\":{\"code\":\"Short distance:\\nsource → claim\\n\\nLonger distance:\\nsource → extraction → summary → assumption → inference → claim\"},\"tunes\":{}},{\"id\":\"YufEvqPPRl\",\"type\":\"paragraph\",\"data\":{\"text\":\"更长的距离并不会自动使一项主张为假。它意味着推理链需要更多的验证。\"},\"tunes\":{}},{\"id\":\"fTJJ76BOtg\",\"type\":\"header\",\"data\":{\"text\":\"为什么这对人工智能很重要\",\"level\":2},\"tunes\":{}},{\"id\":\"EnJwVhC4ut\",\"type\":\"paragraph\",\"data\":{\"text\":\"大语言模型特别擅长在信息片段之间生成流畅的衔接。这种能力很有用，但流畅性可能掩盖结论已经偏离直接检索到的证据有多远。\"},\"tunes\":{}},{\"id\":\"PaHPCJcGtq\",\"type\":\"paragraph\",\"data\":{\"text\":\"因此，一个答案可能看起来来源充分，却包含那些只有通过若干未明说的推断步骤才得到支持的主张。\"},\"tunes\":{}},{\"id\":\"Hr7cDI03fu\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Ffrom-research-protocol-to-a-general-ai-reasoning-framework\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"从研究协议到通用人工智能推理框架\",\"description\":\"可靠的推理要求明确区分证据、假设、假说和结论。\"}},\"tunes\":{}},{\"id\":\"qesGvLUoeT\",\"type\":\"header\",\"data\":{\"text\":\"证据距离的四个组成部分\",\"level\":2},\"tunes\":{}},{\"id\":\"EqY_fbqszH\",\"type\":\"header\",\"data\":{\"text\":\"1. 转换距离\",\"level\":3},\"tunes\":{}},{\"id\":\"23n9wvhPdO\",\"type\":\"paragraph\",\"data\":{\"text\":\"在到达模型之前，原始证据被转换、总结、规范化或重新格式化了多少次？\"},\"tunes\":{}},{\"id\":\"O6QQhmlTxz\",\"type\":\"header\",\"data\":{\"text\":\"2. 推断距离\",\"level\":3},\"tunes\":{}},{\"id\":\"UiYNguu1uN\",\"type\":\"paragraph\",\"data\":{\"text\":\"在证据支持最终结论之前，需要多少推理步骤？\"},\"tunes\":{}},{\"id\":\"g05wqDQgkG\",\"type\":\"header\",\"data\":{\"text\":\"3. 来源距离\",\"level\":3},\"tunes\":{}},{\"id\":\"NN7WM-rh4D\",\"type\":\"paragraph\",\"data\":{\"text\":\"检索到的陈述是否仍可追溯到可识别的原始来源，还是已经经过了来源不明的摘要和副本？\"},\"tunes\":{}},{\"id\":\"pO4BwFbnwm\",\"type\":\"header\",\"data\":{\"text\":\"4. 适用性距离\",\"level\":3},\"tunes\":{}},{\"id\":\"_aLkKwrmAF\",\"type\":\"paragraph\",\"data\":{\"text\":\"需要多少推理才能确定该证据适用于这个确切的版本、用户、司法管辖区、时间段或情况？\"},\"tunes\":{}},{\"id\":\"9AYQBFdRlr\",\"type\":\"header\",\"data\":{\"text\":\"RAG 中的证据距离\",\"level\":2},\"tunes\":{}},{\"id\":\"Jy7wXw4Tk6\",\"type\":\"paragraph\",\"data\":{\"text\":\"RAG可以缩短一个重要距离：模型与外部知识之间的距离。\"},\"tunes\":{}},{\"id\":\"7-qI9PFyFh\",\"type\":\"paragraph\",\"data\":{\"text\":\"它不会自动缩短检索到的段落与最终主张之间的距离。\"},\"tunes\":{}},{\"id\":\"qFBuWPyRVD\",\"type\":\"code\",\"data\":{\"code\":\"RAG solves:\\nmodel ↔ external information\\n\\nRAG does not automatically solve:\\nevidence ↔ justified conclusion\"},\"tunes\":{}},{\"id\":\"2N3t-_hLkJ\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fwhere-does-an-llm-get-its-data-rag-data-sources-in-python\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"LLM从哪里获取数据？Python中的RAG数据源\",\"description\":\"检索将外部信息带入上下文。证据距离则探讨这些信息与最终产生的主张之间发生了什么。\"}},\"tunes\":{}},{\"id\":\"0btFOLsc9z\",\"type\":\"header\",\"data\":{\"text\":\"与检索触发器的联系\",\"level\":2},\"tunes\":{}},{\"id\":\"fgBAbw_jMu\",\"type\":\"paragraph\",\"data\":{\"text\":\"检索触发器询问模型内部知识是否足够。\"},\"tunes\":{}},{\"id\":\"ekw_MX8pcq\",\"type\":\"paragraph\",\"data\":{\"text\":\"证据距离则提出一个更晚的问题：在证据被检索到之后，它如何直接支持系统想要主张的内容？\"},\"tunes\":{}},{\"id\":\"vUELqPH-2O\",\"type\":\"code\",\"data\":{\"code\":\"Need evidence?\\n   ↓\\nRetrieval Trigger\\n   ↓\\nRetrieve\\n   ↓\\nHow directly does evidence support the claim?\\n   ↓\\nEvidence Distance\"},\"tunes\":{}},{\"id\":\"aUnS1BYdWO\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fwhen-should-an-ai-stop-trusting-its-own-knowledge-the-retrieval-trigger\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"AI何时应停止信任自己的知识？——检索触发器\",\"description\":\"触发器决定何时需要证据。证据距离评估最终主张在多大程度上超出了该证据。\"}},\"tunes\":{}},{\"id\":\"7BAWlmyNGr\",\"type\":\"header\",\"data\":{\"text\":\"一个实用的诊断方法\",\"level\":2},\"tunes\":{}},{\"id\":\"g3PGrPE0tb\",\"type\":\"paragraph\",\"data\":{\"text\":\"对于AI生成的重要主张，逆向重建支持链。\"},\"tunes\":{}},{\"id\":\"dp2cnXHuB6\",\"type\":\"code\",\"data\":{\"code\":\"CLAIM\\n  ↑\\nWhich inference produced it?\\n  ↑\\nWhich retrieved evidence supports that inference?\\n  ↑\\nWhich source produced that evidence?\\n  ↑\\nCan the original source still be inspected?\"},\"tunes\":{}},{\"id\":\"XRTZZAdMwT\",\"type\":\"paragraph\",\"data\":{\"text\":\"如果链条断裂，系统就遇到了证据可追溯性问题。\"},\"tunes\":{}},{\"id\":\"sVc8b_RmA3\",\"type\":\"paragraph\",\"data\":{\"text\":\"如果链条保持完整但需要几个强假设，证据距离仍然很高。\"},\"tunes\":{}},{\"id\":\"6ZDtz05TS6\",\"type\":\"header\",\"data\":{\"text\":\"示例：产品数据\",\"level\":2},\"tunes\":{}},{\"id\":\"wzANu_1l77\",\"type\":\"paragraph\",\"data\":{\"text\":\"假设制造商规格说明某台笔记本电脑重1.48公斤。\"},\"tunes\":{}},{\"id\":\"t3Y0dGscSV\",\"type\":\"code\",\"data\":{\"code\":\"Manufacturer specification\\n→ 1.48 kg\\n→ \\\"The laptop weighs 1.48 kg\\\"\\n\\nLow Evidence Distance\"},\"tunes\":{}},{\"id\":\"mhp0ZezeBc\",\"type\":\"paragraph\",\"data\":{\"text\":\"现在考虑：\"},\"tunes\":{}},{\"id\":\"n_YAUzfPsV\",\"type\":\"code\",\"data\":{\"code\":\"Manufacturer specification\\n→ 1.48 kg\\n→ comparison with three competitors\\n→ assumption about travel behavior\\n→ \\\"This is the best laptop for frequent travelers\\\"\\n\\nMuch higher Evidence Distance\"},\"tunes\":{}},{\"id\":\"Y3fMatFi8v\",\"type\":\"paragraph\",\"data\":{\"text\":\"规范是真实的。建议距离规范仍有若干推理步骤。\"},\"tunes\":{}},{\"id\":\"qHYUCCt15Y\",\"type\":\"header\",\"data\":{\"text\":\"示例：科学证据\",\"level\":2},\"tunes\":{}},{\"id\":\"RBVqyYOIA3\",\"type\":\"paragraph\",\"data\":{\"text\":\"一篇论文可能报告观察到的关联。AI 回答可能错误地将其压缩为因果陈述。\"},\"tunes\":{}},{\"id\":\"cLSh6aFNKL\",\"type\":\"code\",\"data\":{\"code\":\"Observed association\\n        ↓\\nInterpretation\\n        ↓\\nUnstated causal assumption\\n        ↓\\n\\\"X causes Y\\\"\"},\"tunes\":{}},{\"id\":\"cxcNCdEY9X\",\"type\":\"paragraph\",\"data\":{\"text\":\"问题不在于没有证据存在。问题在于最终陈述跨越的证据距离大于研究直接支持的范围。\"},\"tunes\":{}},{\"id\":\"JaogISlecp\",\"type\":\"header\",\"data\":{\"text\":\"与智能体就绪上下文的联系\",\"level\":2},\"tunes\":{}},{\"id\":\"naOjS6Pghm\",\"type\":\"paragraph\",\"data\":{\"text\":\"结构化信息对象可以通过保留来源标识、单位、时间戳、冲突和有效性来减少可避免的证据距离，而不是迫使模型从扁平化文本中重建它们。\"},\"tunes\":{}},{\"id\":\"nLTYHo549E\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fportfolio\u002Ffrom-raw-data-to-verifiable-ai-context-a-practical-architecture-for-agent-ready-information\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"从原始数据到可验证的 AI 上下文：面向智能体就绪信息的实用架构\",\"description\":\"在检索之前保留来源和有效性，可减少后续所需的无支持重建步骤。\"}},\"tunes\":{}},{\"id\":\"i1wIDx777S\",\"type\":\"header\",\"data\":{\"text\":\"一个简单的工作量表\",\"level\":2},\"tunes\":{}},{\"id\":\"W4UfU7VTbA\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"级别\",\"描述\"],[\"ED0\",\"直接报告的来源事实\"],[\"ED1\",\"简单的确定性转换或直接比较\"],[\"ED2\",\"一个有意义的解释步骤\"],[\"ED3\",\"多个假设或组合来源\"],[\"ED4\",\"超出直接证据的强因果、预测或规范性结论\"]]},\"tunes\":{}},{\"id\":\"Auxd-zADbs\",\"type\":\"paragraph\",\"data\":{\"text\":\"这个量表是说明性的，不是科学标准。其目的是强制提出一个明确的问题：我们必须在证据和主张之间添加多少内容？\"},\"tunes\":{}},{\"id\":\"d9i0nljc3B\",\"type\":\"header\",\"data\":{\"text\":\"误解\",\"level\":2},\"tunes\":{}},{\"id\":\"tRUSdnqaxg\",\"type\":\"paragraph\",\"data\":{\"text\":\"有引用意味着证据距离低。不。引用可能位于一个大幅超出所引来源所确立内容的主张旁边。\"},\"tunes\":{}},{\"id\":\"xBaooxyjdK\",\"type\":\"paragraph\",\"data\":{\"text\":\"证据距离长意味着主张错误。不。复杂结论可以得到良好支持，但它们需要更强的推理链。\"},\"tunes\":{}},{\"id\":\"M6TP00KpM-\",\"type\":\"paragraph\",\"data\":{\"text\":\"一手来源消除推断。不。一手证据仍可能需要解释。\"},\"tunes\":{}},{\"id\":\"cbeehGVfaZ\",\"type\":\"paragraph\",\"data\":{\"text\":\"更多来源会自动减少距离。同一个无支持主张的十次薄弱重复并不会创造直接证据。\"},\"tunes\":{}},{\"id\":\"yilYDEjtXm\",\"type\":\"header\",\"data\":{\"text\":\"边缘情况：派生数据\",\"level\":2},\"tunes\":{}},{\"id\":\"Rie0jjOB7i\",\"type\":\"paragraph\",\"data\":{\"text\":\"一些衍生值比原始的人类描述更可靠。从已验证交易中正确计算出的总计值可能高度可靠，即使它与原始记录之间隔着一层转换。\"},\"tunes\":{}},{\"id\":\"H_RqySF8UH\",\"type\":\"paragraph\",\"data\":{\"text\":\"因此，证据距离衡量的是认知上的分离程度，而不是对未经改动数据的简单偏好。\"},\"tunes\":{}},{\"id\":\"h_4pF3TWew\",\"type\":\"header\",\"data\":{\"text\":\"局限性\",\"level\":2},\"tunes\":{}},{\"id\":\"HVlueqwNtS\",\"type\":\"paragraph\",\"data\":{\"text\":\"证据距离是一种概念性诊断工具，而不是通用的量化评分。\"},\"tunes\":{}},{\"id\":\"TgG8F5GAz_\",\"type\":\"paragraph\",\"data\":{\"text\":\"不同领域需要对推断进行不同的处理。数学证明、临床证据、软件日志、市场分析和历史解释无法共享一个通用的数值阈值。\"},\"tunes\":{}},{\"id\":\"jhKQ6wrTSH\",\"type\":\"header\",\"data\":{\"text\":\"什么会改变这个答案？\",\"level\":2},\"tunes\":{}},{\"id\":\"fwUqP2Hles\",\"type\":\"paragraph\",\"data\":{\"text\":\"未来的AI系统可能会生成显式的溯源图、机器可检查的主张-证据链接以及带类型的推理轨迹。这可能使证据距离的某些部分可以自动测量。\"},\"tunes\":{}},{\"id\":\"NqwtKBCt9m\",\"type\":\"paragraph\",\"data\":{\"text\":\"根本问题依然存在：即使每一个中间句子听起来都合理，一个主张也可能离其证据越来越远。\"},\"tunes\":{}},{\"id\":\"yZaNUCADQE\",\"type\":\"header\",\"data\":{\"text\":\"结论\",\"level\":2},\"tunes\":{}},{\"id\":\"32JDcomCKE\",\"type\":\"paragraph\",\"data\":{\"text\":\"重要的区别不仅仅在于有来源的答案和无来源的答案之间。\"},\"tunes\":{}},{\"id\":\"eY8fObIJSm\",\"type\":\"paragraph\",\"data\":{\"text\":\"而在于那些仍然接近可直接检验证据的主张，与那些依赖于越来越长的解释和假设链条的主张之间。\"},\"tunes\":{}},{\"id\":\"OS3wrEHB8d\",\"type\":\"code\",\"data\":{\"code\":\"Evidence\\n   ↓\\nEvidence Distance\\n   ↓\\nReasoning\\n   ↓\\nAnswer Validity Boundary\"},\"tunes\":{}},{\"id\":\"-0ka7WOwvJ\",\"type\":\"paragraph\",\"data\":{\"text\":\"证据距离为我们提供了一种描述该空间的语言。\"},\"tunes\":{}},{\"id\":\"FspoypQluI\",\"type\":\"header\",\"data\":{\"text\":\"主要来源\",\"level\":2},\"tunes\":{}},{\"id\":\"GAKfU4izj5\",\"type\":\"paragraph\",\"data\":{\"text\":\"Patrick Lewis 等人，\u003Ca href=\\\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2005.11401\\\" target=\\\"_blank\\\">Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks\u003C\u002Fa>（2020年）。\"},\"tunes\":{}},{\"id\":\"2A6CV-DqBs\",\"type\":\"paragraph\",\"data\":{\"text\":\"Nelson F. Liu 等人，\u003Ca href=\\\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2307.03172\\\" target=\\\"_blank\\\">Lost in the Middle: How Language Models Use Long Contexts\u003C\u002Fa>（2023年）。\"},\"tunes\":{}},{\"id\":\"DvWlEyMpVu\",\"type\":\"paragraph\",\"data\":{\"text\":\"W3C，\u003Ca href=\\\"https:\u002F\u002Fwww.w3.org\u002FTR\u002Fprov-overview\u002F\\\" target=\\\"_blank\\\">PROV Overview\u003C\u002Fa>。一种用于表示溯源关系的标准化模型。\"},\"tunes\":{}}],\"version\":\"2.31.6\"}",{"time":228,"blocks":229,"version":740},1792097029426,[230,236,242,247,255,262,267,272,277,283,288,293,298,320,325,330,335,340,345,350,355,360,365,370,375,380,385,396,401,406,411,416,421,426,431,436,441,446,451,456,461,470,475,480,485,490,499,504,509,514,519,524,529,534,539,544,549,554,559,564,569,574,579,584,592,597,620,625,630,635,640,645,650,655,660,665,670,675,680,685,690,695,700,705,710,715,720,725,730,735],{"id":231,"data":232,"type":42,"tunes":235},"fV7BD0fFRn",{"text":233,"level":234},"问题",2,{},{"id":237,"data":238,"type":240,"tunes":241},"tUBNklFmkB",{"text":239},"两个AI回答可能都引用了来源，但在这些来源对所提主张的支持有多直接上，仍可能大相径庭。我们该如何描述这种差异？","paragraph",{},{"id":243,"data":244,"type":240,"tunes":246},"FwRn77XH9M",{"text":245},"本文使用术语证据距离来表示一项主张与支持它的可直接检验证据之间的距离。",{},{"id":248,"data":249,"type":253,"tunes":254},"YcbJYh_fW7",{"text":250,"caption":251,"alignment":252},"证据距离是指将一项主张与能够直接验证它的证据分隔开的转换、假设和推理步骤的数量与强度。","工作定义——此处引入的一个推理概念，并非标准化指标。","left","quote",{},{"id":256,"data":257,"type":260,"tunes":261},"5KYvEA4v5o",{"title":258,"maxLevel":259,"minLevel":234},"Contents",3,"tableOfContents",{},{"id":263,"data":264,"type":42,"tunes":266},"YxZslT77eK",{"text":265,"level":234},"这真正意味着什么",{},{"id":268,"data":269,"type":240,"tunes":271},"hZtJsMLV04",{"text":270},"引用证明某个来源被参考了。它并不自动证明该来源直接支持引用旁边的句子。",{},{"id":273,"data":274,"type":240,"tunes":276},"xcTtN7QaSQ",{"text":275},"在原始观察与最终AI回答之间，可能存在提取、解析、分块、检索、摘要、解释、聚合和推理。",{},{"id":278,"data":279,"type":281,"tunes":282},"wp6XTpoQAK",{"code":280},"Primary evidence\n   ↓\nExtracted data\n   ↓\nRetrieved passage\n   ↓\nSummary\n   ↓\nInterpretation\n   ↓\nInference\n   ↓\nFinal claim","code",{},{"id":284,"data":285,"type":240,"tunes":287},"wRN6k2vvIy",{"text":286},"每一步都可能是合理的。但每一步也可能引入失真。",{},{"id":289,"data":290,"type":42,"tunes":292},"mo39IqnG40",{"text":291,"level":234},"最简单的例子",{},{"id":294,"data":295,"type":240,"tunes":297},"ysPGEKOYAD",{"text":296},"假设一个温度传感器报告38.2 °C。",{},{"id":299,"data":300,"type":318,"tunes":319},"AyD7F1En7R",{"content":301,"stretched":43,"withHeadings":14},[302,306,310,314],[303,304,305],"主张","与证据的关系","示例性距离",[307,308,309],"传感器报告38.2 °C。","直接读数","低",[311,312,313],"系统运行温度比昨天更高。","需要与昨天的读数进行比较","中等",[315,316,317],"冷却系统正在失效。","需要因果推断和替代解释","高","table",{},{"id":321,"data":322,"type":240,"tunes":324},"WmjjWyf_pZ",{"text":323},"原始传感器值直接支持第一个主张。它本身并不能证明第三个主张。",{},{"id":326,"data":327,"type":281,"tunes":329},"CzgUTnxkNa",{"code":328},"Evidence exists\n        ≠\nEvidence directly supports every derived claim",{},{"id":331,"data":332,"type":42,"tunes":334},"Jrk0FmPH_u",{"text":333,"level":234},"这个例子在哪里不再适用",{},{"id":336,"data":337,"type":240,"tunes":339},"FCSg_sQyVr",{"text":338},"证据距离并不只是句子或处理步骤的计数。",{},{"id":341,"data":342,"type":240,"tunes":344},"Reo2XC69iu",{"text":343},"一个数学上有效的转换可能几乎完美地保留证据，而一个推测性解释可能造成大得多的认识论跳跃。",{},{"id":346,"data":347,"type":240,"tunes":349},"ixWx2M5On0",{"text":348},"因此，这个概念既包括转换距离，也包括推理距离。",{},{"id":351,"data":352,"type":42,"tunes":354},"eiC_FkuMl6",{"text":353,"level":234},"直接回答",{},{"id":356,"data":357,"type":240,"tunes":359},"iDIfyCXVdo",{"text":358},"当一项主张所依赖的转换、假设、较弱的来源或比底层证据直接支持的更强的推断越多时，证据距离就会增加。",{},{"id":361,"data":362,"type":281,"tunes":364},"7nDss8_HDy",{"code":363},"Short distance:\nsource → claim\n\nLonger distance:\nsource → extraction → summary → assumption → inference → claim",{},{"id":366,"data":367,"type":240,"tunes":369},"YufEvqPPRl",{"text":368},"更长的距离并不会自动使一项主张为假。它意味着推理链需要更多的验证。",{},{"id":371,"data":372,"type":42,"tunes":374},"fTJJ76BOtg",{"text":373,"level":234},"为什么这对人工智能很重要",{},{"id":376,"data":377,"type":240,"tunes":379},"EnJwVhC4ut",{"text":378},"大语言模型特别擅长在信息片段之间生成流畅的衔接。这种能力很有用，但流畅性可能掩盖结论已经偏离直接检索到的证据有多远。",{},{"id":381,"data":382,"type":240,"tunes":384},"PaHPCJcGtq",{"text":383},"因此，一个答案可能看起来来源充分，却包含那些只有通过若干未明说的推断步骤才得到支持的主张。",{},{"id":386,"data":387,"type":394,"tunes":395},"Hr7cDI03fu",{"link":388,"meta":389},"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Ffrom-research-protocol-to-a-general-ai-reasoning-framework",{"image":390,"title":392,"description":393},{"url":391},"","从研究协议到通用人工智能推理框架","可靠的推理要求明确区分证据、假设、假说和结论。","linkTool",{},{"id":397,"data":398,"type":42,"tunes":400},"qesGvLUoeT",{"text":399,"level":234},"证据距离的四个组成部分",{},{"id":402,"data":403,"type":42,"tunes":405},"EqY_fbqszH",{"text":404,"level":259},"1. 转换距离",{},{"id":407,"data":408,"type":240,"tunes":410},"23n9wvhPdO",{"text":409},"在到达模型之前，原始证据被转换、总结、规范化或重新格式化了多少次？",{},{"id":412,"data":413,"type":42,"tunes":415},"O6QQhmlTxz",{"text":414,"level":259},"2. 推断距离",{},{"id":417,"data":418,"type":240,"tunes":420},"UiYNguu1uN",{"text":419},"在证据支持最终结论之前，需要多少推理步骤？",{},{"id":422,"data":423,"type":42,"tunes":425},"g05wqDQgkG",{"text":424,"level":259},"3. 来源距离",{},{"id":427,"data":428,"type":240,"tunes":430},"NN7WM-rh4D",{"text":429},"检索到的陈述是否仍可追溯到可识别的原始来源，还是已经经过了来源不明的摘要和副本？",{},{"id":432,"data":433,"type":42,"tunes":435},"pO4BwFbnwm",{"text":434,"level":259},"4. 适用性距离",{},{"id":437,"data":438,"type":240,"tunes":440},"_aLkKwrmAF",{"text":439},"需要多少推理才能确定该证据适用于这个确切的版本、用户、司法管辖区、时间段或情况？",{},{"id":442,"data":443,"type":42,"tunes":445},"9AYQBFdRlr",{"text":444,"level":234},"RAG 中的证据距离",{},{"id":447,"data":448,"type":240,"tunes":450},"Jy7wXw4Tk6",{"text":449},"RAG可以缩短一个重要距离：模型与外部知识之间的距离。",{},{"id":452,"data":453,"type":240,"tunes":455},"7-qI9PFyFh",{"text":454},"它不会自动缩短检索到的段落与最终主张之间的距离。",{},{"id":457,"data":458,"type":281,"tunes":460},"qFBuWPyRVD",{"code":459},"RAG solves:\nmodel ↔ external information\n\nRAG does not automatically solve:\nevidence ↔ justified conclusion",{},{"id":462,"data":463,"type":394,"tunes":469},"2N3t-_hLkJ",{"link":464,"meta":465},"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fwhere-does-an-llm-get-its-data-rag-data-sources-in-python",{"image":466,"title":467,"description":468},{"url":391},"LLM从哪里获取数据？Python中的RAG数据源","检索将外部信息带入上下文。证据距离则探讨这些信息与最终产生的主张之间发生了什么。",{},{"id":471,"data":472,"type":42,"tunes":474},"0btFOLsc9z",{"text":473,"level":234},"与检索触发器的联系",{},{"id":476,"data":477,"type":240,"tunes":479},"fgBAbw_jMu",{"text":478},"检索触发器询问模型内部知识是否足够。",{},{"id":481,"data":482,"type":240,"tunes":484},"ekw_MX8pcq",{"text":483},"证据距离则提出一个更晚的问题：在证据被检索到之后，它如何直接支持系统想要主张的内容？",{},{"id":486,"data":487,"type":281,"tunes":489},"vUELqPH-2O",{"code":488},"Need evidence?\n   ↓\nRetrieval Trigger\n   ↓\nRetrieve\n   ↓\nHow directly does evidence support the claim?\n   ↓\nEvidence Distance",{},{"id":491,"data":492,"type":394,"tunes":498},"aUnS1BYdWO",{"link":493,"meta":494},"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fwhen-should-an-ai-stop-trusting-its-own-knowledge-the-retrieval-trigger",{"image":495,"title":496,"description":497},{"url":391},"AI何时应停止信任自己的知识？——检索触发器","触发器决定何时需要证据。证据距离评估最终主张在多大程度上超出了该证据。",{},{"id":500,"data":501,"type":42,"tunes":503},"7BAWlmyNGr",{"text":502,"level":234},"一个实用的诊断方法",{},{"id":505,"data":506,"type":240,"tunes":508},"g3PGrPE0tb",{"text":507},"对于AI生成的重要主张，逆向重建支持链。",{},{"id":510,"data":511,"type":281,"tunes":513},"dp2cnXHuB6",{"code":512},"CLAIM\n  ↑\nWhich inference produced it?\n  ↑\nWhich retrieved evidence supports that inference?\n  ↑\nWhich source produced that evidence?\n  ↑\nCan the original source still be inspected?",{},{"id":515,"data":516,"type":240,"tunes":518},"XRTZZAdMwT",{"text":517},"如果链条断裂，系统就遇到了证据可追溯性问题。",{},{"id":520,"data":521,"type":240,"tunes":523},"sVc8b_RmA3",{"text":522},"如果链条保持完整但需要几个强假设，证据距离仍然很高。",{},{"id":525,"data":526,"type":42,"tunes":528},"6ZDtz05TS6",{"text":527,"level":234},"示例：产品数据",{},{"id":530,"data":531,"type":240,"tunes":533},"wzANu_1l77",{"text":532},"假设制造商规格说明某台笔记本电脑重1.48公斤。",{},{"id":535,"data":536,"type":281,"tunes":538},"t3Y0dGscSV",{"code":537},"Manufacturer specification\n→ 1.48 kg\n→ \"The laptop weighs 1.48 kg\"\n\nLow Evidence Distance",{},{"id":540,"data":541,"type":240,"tunes":543},"mhp0ZezeBc",{"text":542},"现在考虑：",{},{"id":545,"data":546,"type":281,"tunes":548},"n_YAUzfPsV",{"code":547},"Manufacturer specification\n→ 1.48 kg\n→ comparison with three competitors\n→ assumption about travel behavior\n→ \"This is the best laptop for frequent travelers\"\n\nMuch higher Evidence Distance",{},{"id":550,"data":551,"type":240,"tunes":553},"Y3fMatFi8v",{"text":552},"规范是真实的。建议距离规范仍有若干推理步骤。",{},{"id":555,"data":556,"type":42,"tunes":558},"qHYUCCt15Y",{"text":557,"level":234},"示例：科学证据",{},{"id":560,"data":561,"type":240,"tunes":563},"RBVqyYOIA3",{"text":562},"一篇论文可能报告观察到的关联。AI 回答可能错误地将其压缩为因果陈述。",{},{"id":565,"data":566,"type":281,"tunes":568},"cLSh6aFNKL",{"code":567},"Observed association\n        ↓\nInterpretation\n        ↓\nUnstated causal assumption\n        ↓\n\"X causes Y\"",{},{"id":570,"data":571,"type":240,"tunes":573},"cxcNCdEY9X",{"text":572},"问题不在于没有证据存在。问题在于最终陈述跨越的证据距离大于研究直接支持的范围。",{},{"id":575,"data":576,"type":42,"tunes":578},"JaogISlecp",{"text":577,"level":234},"与智能体就绪上下文的联系",{},{"id":580,"data":581,"type":240,"tunes":583},"naOjS6Pghm",{"text":582},"结构化信息对象可以通过保留来源标识、单位、时间戳、冲突和有效性来减少可避免的证据距离，而不是迫使模型从扁平化文本中重建它们。",{},{"id":585,"data":586,"type":394,"tunes":591},"nLTYHo549E",{"link":587,"meta":588},"https:\u002F\u002Fstajic.de\u002Fzh\u002Fportfolio\u002Ffrom-raw-data-to-verifiable-ai-context-a-practical-architecture-for-agent-ready-information",{"image":589,"title":185,"description":590},{"url":391},"在检索之前保留来源和有效性，可减少后续所需的无支持重建步骤。",{},{"id":593,"data":594,"type":42,"tunes":596},"i1wIDx777S",{"text":595,"level":234},"一个简单的工作量表",{},{"id":598,"data":599,"type":318,"tunes":619},"W4UfU7VTbA",{"content":600,"stretched":43,"withHeadings":14},[601,604,607,610,613,616],[602,603],"级别","描述",[605,606],"ED0","直接报告的来源事实",[608,609],"ED1","简单的确定性转换或直接比较",[611,612],"ED2","一个有意义的解释步骤",[614,615],"ED3","多个假设或组合来源",[617,618],"ED4","超出直接证据的强因果、预测或规范性结论",{},{"id":621,"data":622,"type":240,"tunes":624},"Auxd-zADbs",{"text":623},"这个量表是说明性的，不是科学标准。其目的是强制提出一个明确的问题：我们必须在证据和主张之间添加多少内容？",{},{"id":626,"data":627,"type":42,"tunes":629},"d9i0nljc3B",{"text":628,"level":234},"误解",{},{"id":631,"data":632,"type":240,"tunes":634},"tRUSdnqaxg",{"text":633},"有引用意味着证据距离低。不。引用可能位于一个大幅超出所引来源所确立内容的主张旁边。",{},{"id":636,"data":637,"type":240,"tunes":639},"xBaooxyjdK",{"text":638},"证据距离长意味着主张错误。不。复杂结论可以得到良好支持，但它们需要更强的推理链。",{},{"id":641,"data":642,"type":240,"tunes":644},"M6TP00KpM-",{"text":643},"一手来源消除推断。不。一手证据仍可能需要解释。",{},{"id":646,"data":647,"type":240,"tunes":649},"cbeehGVfaZ",{"text":648},"更多来源会自动减少距离。同一个无支持主张的十次薄弱重复并不会创造直接证据。",{},{"id":651,"data":652,"type":42,"tunes":654},"yilYDEjtXm",{"text":653,"level":234},"边缘情况：派生数据",{},{"id":656,"data":657,"type":240,"tunes":659},"Rie0jjOB7i",{"text":658},"一些衍生值比原始的人类描述更可靠。从已验证交易中正确计算出的总计值可能高度可靠，即使它与原始记录之间隔着一层转换。",{},{"id":661,"data":662,"type":240,"tunes":664},"H_RqySF8UH",{"text":663},"因此，证据距离衡量的是认知上的分离程度，而不是对未经改动数据的简单偏好。",{},{"id":666,"data":667,"type":42,"tunes":669},"h_4pF3TWew",{"text":668,"level":234},"局限性",{},{"id":671,"data":672,"type":240,"tunes":674},"HVlueqwNtS",{"text":673},"证据距离是一种概念性诊断工具，而不是通用的量化评分。",{},{"id":676,"data":677,"type":240,"tunes":679},"TgG8F5GAz_",{"text":678},"不同领域需要对推断进行不同的处理。数学证明、临床证据、软件日志、市场分析和历史解释无法共享一个通用的数值阈值。",{},{"id":681,"data":682,"type":42,"tunes":684},"jhKQ6wrTSH",{"text":683,"level":234},"什么会改变这个答案？",{},{"id":686,"data":687,"type":240,"tunes":689},"fwUqP2Hles",{"text":688},"未来的AI系统可能会生成显式的溯源图、机器可检查的主张-证据链接以及带类型的推理轨迹。这可能使证据距离的某些部分可以自动测量。",{},{"id":691,"data":692,"type":240,"tunes":694},"NqwtKBCt9m",{"text":693},"根本问题依然存在：即使每一个中间句子听起来都合理，一个主张也可能离其证据越来越远。",{},{"id":696,"data":697,"type":42,"tunes":699},"yZaNUCADQE",{"text":698,"level":234},"结论",{},{"id":701,"data":702,"type":240,"tunes":704},"32JDcomCKE",{"text":703},"重要的区别不仅仅在于有来源的答案和无来源的答案之间。",{},{"id":706,"data":707,"type":240,"tunes":709},"eY8fObIJSm",{"text":708},"而在于那些仍然接近可直接检验证据的主张，与那些依赖于越来越长的解释和假设链条的主张之间。",{},{"id":711,"data":712,"type":281,"tunes":714},"OS3wrEHB8d",{"code":713},"Evidence\n   ↓\nEvidence Distance\n   ↓\nReasoning\n   ↓\nAnswer Validity Boundary",{},{"id":716,"data":717,"type":240,"tunes":719},"-0ka7WOwvJ",{"text":718},"证据距离为我们提供了一种描述该空间的语言。",{},{"id":721,"data":722,"type":42,"tunes":724},"FspoypQluI",{"text":723,"level":234},"主要来源",{},{"id":726,"data":727,"type":240,"tunes":729},"GAKfU4izj5",{"text":728},"Patrick Lewis 等人，\u003Ca href=\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2005.11401\" target=\"_blank\">Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks\u003C\u002Fa>（2020年）。",{},{"id":731,"data":732,"type":240,"tunes":734},"2A6CV-DqBs",{"text":733},"Nelson F. Liu 等人，\u003Ca href=\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2307.03172\" target=\"_blank\">Lost in the Middle: How Language Models Use Long Contexts\u003C\u002Fa>（2023年）。",{},{"id":736,"data":737,"type":240,"tunes":739},"DvWlEyMpVu",{"text":738},"W3C，\u003Ca href=\"https:\u002F\u002Fwww.w3.org\u002FTR\u002Fprov-overview\u002F\" target=\"_blank\">PROV Overview\u003C\u002Fa>。一种用于表示溯源关系的标准化模型。",{},"2.31.6","并非所有被支持的答案都与其证据同样接近。证据距离是一种实用的方式，用来描述一个AI主张与可直接检验的证据之间相隔了多少次转换、摘要、假设和推理步骤。","\u002Fuploads\u002F2026\u002F10\u002Fhow-far-is-an-ai-claim-from-the-evidence-evidence-distance-1792095863209-3c1mog.webp","how-far-is-an-ai-claim-from-the-evidence-evidence-distance-1792095863209-3c1mog","PUBLISHED","2026-10-15T00:23:00.000Z","2026-10-15T20:23:08.634Z","2026-10-15T20:43:52.893Z",{"en":749,"de":750,"sr":751,"es":752,"fr":753,"it":754,"ru":755,"zh":756},"\u002Fblog\u002Fhow-far-is-an-ai-claim-from-the-evidence-evidence-distance","\u002Fde\u002Fblog\u002Fhow-far-is-an-ai-claim-from-the-evidence-evidence-distance","\u002Fsr\u002Fblog\u002Fhow-far-is-an-ai-claim-from-the-evidence-evidence-distance","\u002Fes\u002Fblog\u002Fhow-far-is-an-ai-claim-from-the-evidence-evidence-distance","\u002Ffr\u002Fblog\u002Fhow-far-is-an-ai-claim-from-the-evidence-evidence-distance","\u002Fit\u002Fblog\u002Fhow-far-is-an-ai-claim-from-the-evidence-evidence-distance","\u002Fru\u002Fblog\u002Fhow-far-is-an-ai-claim-from-the-evidence-evidence-distance","\u002Fzh\u002Fblog\u002Fhow-far-is-an-ai-claim-from-the-evidence-evidence-distance",[758,762],{"id":759,"name":760,"slug":761},68,"风险、控制与证据","risks-and-controls",{"id":763,"name":764,"slug":765},49,"控制与证据","controls",{"id":767,"login":768,"email":391,"displayName":769},"20","rooth8233","Aleksandar Stajić",[771,1264],{"lang":772,"title":773,"content":774,"contentJson":775,"excerpt":1263},"en","How Far Is an AI Claim From the Evidence? — Evidence Distance","{\"time\":1792096795538,\"blocks\":[{\"id\":\"g-2Ciyj8LJ\",\"type\":\"header\",\"data\":{\"text\":\"Question\",\"level\":2},\"tunes\":{}},{\"id\":\"SQWD7UraiX\",\"type\":\"paragraph\",\"data\":{\"text\":\"Two AI answers can both cite sources and still differ greatly in how directly those sources support the claims being made. How can we describe that difference?\"},\"tunes\":{}},{\"id\":\"Sj1fx2865j\",\"type\":\"paragraph\",\"data\":{\"text\":\"This article uses the term Evidence Distance for the distance between a claim and the directly inspectable evidence that supports it.\"},\"tunes\":{}},{\"id\":\"d5th-A-mq5\",\"type\":\"quote\",\"data\":{\"text\":\"Evidence Distance is the number and strength of transformations, assumptions and inferential steps separating a claim from the evidence that can directly verify it.\",\"caption\":\"Working definition — a reasoning concept introduced here, not a standardized metric.\",\"alignment\":\"left\"},\"tunes\":{}},{\"id\":\"X81jXC72FC\",\"type\":\"tableOfContents\",\"data\":{\"title\":\"Contents\",\"minLevel\":2,\"maxLevel\":3},\"tunes\":{}},{\"id\":\"_pZcnPEoil\",\"type\":\"header\",\"data\":{\"text\":\"What This Really Means\",\"level\":2},\"tunes\":{}},{\"id\":\"F5odO7xyha\",\"type\":\"paragraph\",\"data\":{\"text\":\"A citation proves that a source was referenced. It does not automatically prove that the source directly supports the sentence beside the citation.\"},\"tunes\":{}},{\"id\":\"8Gqlg-ArJS\",\"type\":\"paragraph\",\"data\":{\"text\":\"Between an original observation and a final AI answer there may be extraction, parsing, chunking, retrieval, summarization, interpretation, aggregation and inference.\"},\"tunes\":{}},{\"id\":\"lXkL5LV9ab\",\"type\":\"code\",\"data\":{\"code\":\"Primary evidence\\n   ↓\\nExtracted data\\n   ↓\\nRetrieved passage\\n   ↓\\nSummary\\n   ↓\\nInterpretation\\n   ↓\\nInference\\n   ↓\\nFinal claim\"},\"tunes\":{}},{\"id\":\"Z-8GPQIpCI\",\"type\":\"paragraph\",\"data\":{\"text\":\"Every step can be legitimate. But every step can also introduce distortion.\"},\"tunes\":{}},{\"id\":\"aXMwTGZ8Ia\",\"type\":\"header\",\"data\":{\"text\":\"Simplest Example\",\"level\":2},\"tunes\":{}},{\"id\":\"bZlU7KD3M2\",\"type\":\"paragraph\",\"data\":{\"text\":\"Suppose a temperature sensor reports 38.2 °C.\"},\"tunes\":{}},{\"id\":\"SP87UUPaU3\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"Claim\",\"Relationship to evidence\",\"Illustrative distance\"],[\"The sensor reports 38.2 °C.\",\"Direct reading\",\"Low\"],[\"The system is running hotter than yesterday.\",\"Requires comparison with yesterday's reading\",\"Moderate\"],[\"The cooling system is failing.\",\"Requires causal inference and alternative explanations\",\"High\"]]},\"tunes\":{}},{\"id\":\"eYar059d69\",\"type\":\"paragraph\",\"data\":{\"text\":\"The original sensor value directly supports the first claim. It does not by itself prove the third.\"},\"tunes\":{}},{\"id\":\"eCnFlByvHt\",\"type\":\"code\",\"data\":{\"code\":\"Evidence exists\\n        ≠\\nEvidence directly supports every derived claim\"},\"tunes\":{}},{\"id\":\"aXGkK6BpkF\",\"type\":\"header\",\"data\":{\"text\":\"Where the Example Stops Working\",\"level\":2},\"tunes\":{}},{\"id\":\"lAwb9QZ_yB\",\"type\":\"paragraph\",\"data\":{\"text\":\"Evidence Distance is not simply a count of sentences or processing steps.\"},\"tunes\":{}},{\"id\":\"-SzYko9oIH\",\"type\":\"paragraph\",\"data\":{\"text\":\"One mathematically valid transformation may preserve evidence almost perfectly, while a single speculative interpretation may create a much larger epistemic jump.\"},\"tunes\":{}},{\"id\":\"XOQloBndLw\",\"type\":\"paragraph\",\"data\":{\"text\":\"The concept therefore includes both distance by transformation and distance by inference.\"},\"tunes\":{}},{\"id\":\"EOXbS4TAOR\",\"type\":\"header\",\"data\":{\"text\":\"Direct Answer\",\"level\":2},\"tunes\":{}},{\"id\":\"HcVL8vGM74\",\"type\":\"paragraph\",\"data\":{\"text\":\"Evidence Distance increases when a claim depends on more transformations, more assumptions, weaker provenance or stronger inference than the underlying evidence directly warrants.\"},\"tunes\":{}},{\"id\":\"_f7WXjUOsS\",\"type\":\"code\",\"data\":{\"code\":\"Short distance:\\nsource → claim\\n\\nLonger distance:\\nsource → extraction → summary → assumption → inference → claim\"},\"tunes\":{}},{\"id\":\"zUEC-cDhqm\",\"type\":\"paragraph\",\"data\":{\"text\":\"A longer distance does not automatically make a claim false. It means the reasoning chain requires more validation.\"},\"tunes\":{}},{\"id\":\"ix7TUGNz_W\",\"type\":\"header\",\"data\":{\"text\":\"Why This Matters for AI\",\"level\":2},\"tunes\":{}},{\"id\":\"wrAy5PLtLB\",\"type\":\"paragraph\",\"data\":{\"text\":\"LLMs are particularly good at producing fluent bridges between pieces of information. That capability is useful, but fluency can hide how far the conclusion has moved beyond the directly retrieved evidence.\"},\"tunes\":{}},{\"id\":\"jJseLuTWdv\",\"type\":\"paragraph\",\"data\":{\"text\":\"An answer can therefore look well sourced while containing claims whose support exists only through several unstated inferential steps.\"},\"tunes\":{}},{\"id\":\"ox8fok9dzs\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fstajic.de\u002Fblog\u002Ffrom-research-protocol-to-a-general-ai-reasoning-framework\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"From Research Protocol to a General AI Reasoning Framework\",\"description\":\"Reliable reasoning requires explicit separation of evidence, assumptions, hypotheses and conclusions.\"}},\"tunes\":{}},{\"id\":\"6ipse6YG1U\",\"type\":\"header\",\"data\":{\"text\":\"Four Components of Evidence Distance\",\"level\":2},\"tunes\":{}},{\"id\":\"P4xQ7o96dR\",\"type\":\"header\",\"data\":{\"text\":\"1. Transformation Distance\",\"level\":3},\"tunes\":{}},{\"id\":\"dvN8TUNEYG\",\"type\":\"paragraph\",\"data\":{\"text\":\"How many times has the original evidence been converted, summarized, normalized or reformatted before reaching the model?\"},\"tunes\":{}},{\"id\":\"dXofACF9_x\",\"type\":\"header\",\"data\":{\"text\":\"2. Inferential Distance\",\"level\":3},\"tunes\":{}},{\"id\":\"kkEzV3ydId\",\"type\":\"paragraph\",\"data\":{\"text\":\"How many reasoning steps are required before the evidence supports the final conclusion?\"},\"tunes\":{}},{\"id\":\"PPYB0RuX5_\",\"type\":\"header\",\"data\":{\"text\":\"3. Provenance Distance\",\"level\":3},\"tunes\":{}},{\"id\":\"cAVBYebJr7\",\"type\":\"paragraph\",\"data\":{\"text\":\"Can the retrieved statement still be traced to an identifiable original source, or has it passed through summaries and copies whose origin is unclear?\"},\"tunes\":{}},{\"id\":\"254N_99Z7X\",\"type\":\"header\",\"data\":{\"text\":\"4. Applicability Distance\",\"level\":3},\"tunes\":{}},{\"id\":\"to3KWv5WSK\",\"type\":\"paragraph\",\"data\":{\"text\":\"How much reasoning is required to establish that the evidence applies to this exact version, user, jurisdiction, time period or situation?\"},\"tunes\":{}},{\"id\":\"3WnVPQ6wqw\",\"type\":\"header\",\"data\":{\"text\":\"Evidence Distance in RAG\",\"level\":2},\"tunes\":{}},{\"id\":\"r8s-75EBhA\",\"type\":\"paragraph\",\"data\":{\"text\":\"RAG can shorten one important distance: the distance between a model and external knowledge.\"},\"tunes\":{}},{\"id\":\"I_8h52A3Yc\",\"type\":\"paragraph\",\"data\":{\"text\":\"It does not automatically shorten the distance between a retrieved passage and the final claim.\"},\"tunes\":{}},{\"id\":\"jDDRUu2FUK\",\"type\":\"code\",\"data\":{\"code\":\"RAG solves:\\nmodel ↔ external information\\n\\nRAG does not automatically solve:\\nevidence ↔ justified conclusion\"},\"tunes\":{}},{\"id\":\"rm14qIVB1M\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fwhere-does-an-llm-get-its-data-rag-data-sources-in-python\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"Where Does an LLM Get Its Data? RAG Data Sources in Python\",\"description\":\"Retrieval brings external information into context. Evidence Distance asks what happens between that information and the claim eventually produced.\"}},\"tunes\":{}},{\"id\":\"CRUZDXYloa\",\"type\":\"header\",\"data\":{\"text\":\"Connection to the Retrieval Trigger\",\"level\":2},\"tunes\":{}},{\"id\":\"-TPiYJ569c\",\"type\":\"paragraph\",\"data\":{\"text\":\"The Retrieval Trigger asks whether internal model knowledge is sufficient.\"},\"tunes\":{}},{\"id\":\"_VjLi5fs4x\",\"type\":\"paragraph\",\"data\":{\"text\":\"Evidence Distance asks a later question: after evidence has been retrieved, how directly does it support what the system wants to claim?\"},\"tunes\":{}},{\"id\":\"EImPw6i617\",\"type\":\"code\",\"data\":{\"code\":\"Need evidence?\\n   ↓\\nRetrieval Trigger\\n   ↓\\nRetrieve\\n   ↓\\nHow directly does evidence support the claim?\\n   ↓\\nEvidence Distance\"},\"tunes\":{}},{\"id\":\"ZqyhB2bFtR\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fwhen-should-an-ai-stop-trusting-its-own-knowledge-the-retrieval-trigger\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"When Should an AI Stop Trusting Its Own Knowledge? — The Retrieval Trigger\",\"description\":\"The trigger decides when evidence is needed. Evidence Distance evaluates how far the eventual claim moves beyond that evidence.\"}},\"tunes\":{}},{\"id\":\"ST2vnm6vwF\",\"type\":\"header\",\"data\":{\"text\":\"A Practical Diagnostic\",\"level\":2},\"tunes\":{}},{\"id\":\"qyoHcDcTKb\",\"type\":\"paragraph\",\"data\":{\"text\":\"For an important AI-generated claim, reconstruct the support chain backwards.\"},\"tunes\":{}},{\"id\":\"e3fDmLk0fe\",\"type\":\"code\",\"data\":{\"code\":\"CLAIM\\n  ↑\\nWhich inference produced it?\\n  ↑\\nWhich retrieved evidence supports that inference?\\n  ↑\\nWhich source produced that evidence?\\n  ↑\\nCan the original source still be inspected?\"},\"tunes\":{}},{\"id\":\"iQ60c7NRaE\",\"type\":\"paragraph\",\"data\":{\"text\":\"If the chain breaks, the system has encountered an evidence-traceability problem.\"},\"tunes\":{}},{\"id\":\"YQAKlNZ2gM\",\"type\":\"paragraph\",\"data\":{\"text\":\"If the chain remains intact but requires several strong assumptions, the Evidence Distance is still high.\"},\"tunes\":{}},{\"id\":\"clN7jFkUBf\",\"type\":\"header\",\"data\":{\"text\":\"Example: Product Data\",\"level\":2},\"tunes\":{}},{\"id\":\"CRF-A9O7Ot\",\"type\":\"paragraph\",\"data\":{\"text\":\"Suppose a manufacturer specification says a laptop weighs 1.48 kg.\"},\"tunes\":{}},{\"id\":\"JF3z6aF2q_\",\"type\":\"code\",\"data\":{\"code\":\"Manufacturer specification\\n→ 1.48 kg\\n→ \\\"The laptop weighs 1.48 kg\\\"\\n\\nLow Evidence Distance\"},\"tunes\":{}},{\"id\":\"vTMOOgkLPb\",\"type\":\"paragraph\",\"data\":{\"text\":\"Now consider:\"},\"tunes\":{}},{\"id\":\"Tuhfzor1yp\",\"type\":\"code\",\"data\":{\"code\":\"Manufacturer specification\\n→ 1.48 kg\\n→ comparison with three competitors\\n→ assumption about travel behavior\\n→ \\\"This is the best laptop for frequent travelers\\\"\\n\\nMuch higher Evidence Distance\"},\"tunes\":{}},{\"id\":\"JHFRGIgUJD\",\"type\":\"paragraph\",\"data\":{\"text\":\"The specification is real. The recommendation is still several reasoning steps away from it.\"},\"tunes\":{}},{\"id\":\"2lKERCrnW1\",\"type\":\"header\",\"data\":{\"text\":\"Example: Scientific Evidence\",\"level\":2},\"tunes\":{}},{\"id\":\"HA8DBYwUuK\",\"type\":\"paragraph\",\"data\":{\"text\":\"A paper may report an observed association. An AI answer can incorrectly compress that into a causal statement.\"},\"tunes\":{}},{\"id\":\"D9C2dVB7j_\",\"type\":\"code\",\"data\":{\"code\":\"Observed association\\n        ↓\\nInterpretation\\n        ↓\\nUnstated causal assumption\\n        ↓\\n\\\"X causes Y\\\"\"},\"tunes\":{}},{\"id\":\"CKNjK25pkL\",\"type\":\"paragraph\",\"data\":{\"text\":\"The problem is not that no evidence exists. The problem is that the final statement crosses a larger evidential distance than the study directly supports.\"},\"tunes\":{}},{\"id\":\"sR430N3cDd\",\"type\":\"header\",\"data\":{\"text\":\"Connection to Agent-Ready Context\",\"level\":2},\"tunes\":{}},{\"id\":\"v77YyG55ko\",\"type\":\"paragraph\",\"data\":{\"text\":\"Structured information objects can reduce avoidable Evidence Distance by preserving source identity, units, timestamps, conflicts and validity instead of forcing the model to reconstruct them from flattened text.\"},\"tunes\":{}},{\"id\":\"EX8darCIjd\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fstajic.de\u002Fportfolio\u002Ffrom-raw-data-to-verifiable-ai-context-a-practical-architecture-for-agent-ready-information\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"From Raw Data to Verifiable AI Context: A Practical Architecture for Agent-Ready Information\",\"description\":\"Preserving provenance and validity before retrieval reduces the number of unsupported reconstruction steps required later.\"}},\"tunes\":{}},{\"id\":\"dsfuoURj1e\",\"type\":\"header\",\"data\":{\"text\":\"A Simple Working Scale\",\"level\":2},\"tunes\":{}},{\"id\":\"BLi7Yth64O\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"Level\",\"Description\"],[\"ED0\",\"Directly reported source fact\"],[\"ED1\",\"Simple deterministic transformation or direct comparison\"],[\"ED2\",\"One meaningful interpretive step\"],[\"ED3\",\"Multiple assumptions or combined sources\"],[\"ED4\",\"Strong causal, predictive or prescriptive conclusion beyond direct evidence\"]]},\"tunes\":{}},{\"id\":\"8y8bzCcHBs\",\"type\":\"paragraph\",\"data\":{\"text\":\"This scale is illustrative, not a scientific standard. Its purpose is to force an explicit question: How much did we have to add between the evidence and the claim?\"},\"tunes\":{}},{\"id\":\"aFK0YiAlYO\",\"type\":\"header\",\"data\":{\"text\":\"Misconceptions\",\"level\":2},\"tunes\":{}},{\"id\":\"07KSk6xDx-\",\"type\":\"paragraph\",\"data\":{\"text\":\"A citation means Evidence Distance is low. No. A citation can sit beside a claim that substantially exceeds what the cited source establishes.\"},\"tunes\":{}},{\"id\":\"t_cf-Zfan9\",\"type\":\"paragraph\",\"data\":{\"text\":\"Long Evidence Distance means the claim is wrong. No. Complex conclusions can be well supported, but they require a stronger reasoning chain.\"},\"tunes\":{}},{\"id\":\"Pu9PRX67Q6\",\"type\":\"paragraph\",\"data\":{\"text\":\"Primary sources eliminate inference. No. Primary evidence can still require interpretation.\"},\"tunes\":{}},{\"id\":\"oF1xNsOHHq\",\"type\":\"paragraph\",\"data\":{\"text\":\"More sources automatically reduce distance. Ten weak repetitions of the same unsupported claim do not create direct evidence.\"},\"tunes\":{}},{\"id\":\"IkolK5baph\",\"type\":\"header\",\"data\":{\"text\":\"Edge Case: Derived Data\",\"level\":2},\"tunes\":{}},{\"id\":\"jRH36fwdR3\",\"type\":\"paragraph\",\"data\":{\"text\":\"Some derived values are more reliable than raw human descriptions. A correctly calculated total from verified transactions can be highly dependable even though it is one transformation away from the original records.\"},\"tunes\":{}},{\"id\":\"SXbdKiNDst\",\"type\":\"paragraph\",\"data\":{\"text\":\"Evidence Distance therefore measures epistemic separation, not a simplistic preference for untouched data.\"},\"tunes\":{}},{\"id\":\"mUbvdnFChM\",\"type\":\"header\",\"data\":{\"text\":\"Limitations\",\"level\":2},\"tunes\":{}},{\"id\":\"N_EskxFOep\",\"type\":\"paragraph\",\"data\":{\"text\":\"Evidence Distance is a conceptual diagnostic, not a universal quantitative score.\"},\"tunes\":{}},{\"id\":\"nW_-E08So_\",\"type\":\"paragraph\",\"data\":{\"text\":\"Different domains require different treatment of inference. Mathematical proof, clinical evidence, software logs, market analysis and historical interpretation cannot share one universal numerical threshold.\"},\"tunes\":{}},{\"id\":\"_Bd95wXkri\",\"type\":\"header\",\"data\":{\"text\":\"What Would Change This Answer?\",\"level\":2},\"tunes\":{}},{\"id\":\"OqCaMc2_OX\",\"type\":\"paragraph\",\"data\":{\"text\":\"Future AI systems may generate explicit provenance graphs, machine-checkable claim-evidence links and typed reasoning traces. That could make parts of Evidence Distance measurable automatically.\"},\"tunes\":{}},{\"id\":\"58Hf3O9-Iz\",\"type\":\"paragraph\",\"data\":{\"text\":\"The underlying problem remains: a claim can move farther from its evidence even while every intermediate sentence sounds plausible.\"},\"tunes\":{}},{\"id\":\"8WOq5oE_e_\",\"type\":\"header\",\"data\":{\"text\":\"Conclusion\",\"level\":2},\"tunes\":{}},{\"id\":\"1pr1TfwnKc\",\"type\":\"paragraph\",\"data\":{\"text\":\"The important distinction is not merely between sourced and unsourced answers.\"},\"tunes\":{}},{\"id\":\"54CysT_NXa\",\"type\":\"paragraph\",\"data\":{\"text\":\"It is between claims that remain close to directly inspectable evidence and claims that depend on increasingly long chains of interpretation and assumption.\"},\"tunes\":{}},{\"id\":\"Qrjo3WEUSL\",\"type\":\"code\",\"data\":{\"code\":\"Evidence\\n   ↓\\nEvidence Distance\\n   ↓\\nReasoning\\n   ↓\\nAnswer Validity Boundary\"},\"tunes\":{}},{\"id\":\"6ir8aIAfd3\",\"type\":\"paragraph\",\"data\":{\"text\":\"Evidence Distance gives us a language for describing that space.\"},\"tunes\":{}},{\"id\":\"vPhupXcbeN\",\"type\":\"header\",\"data\":{\"text\":\"Primary Sources\",\"level\":2},\"tunes\":{}},{\"id\":\"BaO4JmWRQL\",\"type\":\"paragraph\",\"data\":{\"text\":\"Patrick Lewis et al., \u003Ca href=\\\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2005.11401\\\" target=\\\"_blank\\\">Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks\u003C\u002Fa> (2020).\"},\"tunes\":{}},{\"id\":\"eg4Z4G6Cgs\",\"type\":\"paragraph\",\"data\":{\"text\":\"Nelson F. Liu et al., \u003Ca href=\\\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2307.03172\\\" target=\\\"_blank\\\">Lost in the Middle: How Language Models Use Long Contexts\u003C\u002Fa> (2023).\"},\"tunes\":{}},{\"id\":\"BJH7zUd4T8\",\"type\":\"paragraph\",\"data\":{\"text\":\"W3C, \u003Ca href=\\\"https:\u002F\u002Fwww.w3.org\u002FTR\u002Fprov-overview\u002F\\\" target=\\\"_blank\\\">PROV Overview\u003C\u002Fa>. A standardized model for representing provenance relationships.\"},\"tunes\":{}}],\"version\":\"2.31.6\"}",{"time":776,"blocks":777,"version":740},1792096795538,[778,783,788,793,799,803,808,813,818,822,827,832,837,858,863,867,872,877,882,887,892,897,901,906,911,916,921,930,935,940,945,950,955,960,965,970,975,980,985,990,994,1003,1008,1013,1018,1022,1031,1036,1041,1045,1050,1055,1060,1065,1069,1074,1078,1083,1088,1093,1097,1102,1107,1112,1121,1126,1144,1149,1154,1159,1164,1169,1174,1179,1184,1189,1194,1199,1204,1209,1214,1219,1224,1229,1234,1238,1243,1248,1253,1258],{"id":779,"data":780,"type":42,"tunes":782},"g-2Ciyj8LJ",{"text":781,"level":234},"Question",{},{"id":784,"data":785,"type":240,"tunes":787},"SQWD7UraiX",{"text":786},"Two AI answers can both cite sources and still differ greatly in how directly those sources support the claims being made. How can we describe that difference?",{},{"id":789,"data":790,"type":240,"tunes":792},"Sj1fx2865j",{"text":791},"This article uses the term Evidence Distance for the distance between a claim and the directly inspectable evidence that supports it.",{},{"id":794,"data":795,"type":253,"tunes":798},"d5th-A-mq5",{"text":796,"caption":797,"alignment":252},"Evidence Distance is the number and strength of transformations, assumptions and inferential steps separating a claim from the evidence that can directly verify it.","Working definition — a reasoning concept introduced here, not a standardized metric.",{},{"id":800,"data":801,"type":260,"tunes":802},"X81jXC72FC",{"title":258,"maxLevel":259,"minLevel":234},{},{"id":804,"data":805,"type":42,"tunes":807},"_pZcnPEoil",{"text":806,"level":234},"What This Really Means",{},{"id":809,"data":810,"type":240,"tunes":812},"F5odO7xyha",{"text":811},"A citation proves that a source was referenced. It does not automatically prove that the source directly supports the sentence beside the citation.",{},{"id":814,"data":815,"type":240,"tunes":817},"8Gqlg-ArJS",{"text":816},"Between an original observation and a final AI answer there may be extraction, parsing, chunking, retrieval, summarization, interpretation, aggregation and inference.",{},{"id":819,"data":820,"type":281,"tunes":821},"lXkL5LV9ab",{"code":280},{},{"id":823,"data":824,"type":240,"tunes":826},"Z-8GPQIpCI",{"text":825},"Every step can be legitimate. But every step can also introduce distortion.",{},{"id":828,"data":829,"type":42,"tunes":831},"aXMwTGZ8Ia",{"text":830,"level":234},"Simplest Example",{},{"id":833,"data":834,"type":240,"tunes":836},"bZlU7KD3M2",{"text":835},"Suppose a temperature sensor reports 38.2 °C.",{},{"id":838,"data":839,"type":318,"tunes":857},"SP87UUPaU3",{"content":840,"stretched":43,"withHeadings":14},[841,845,849,853],[842,843,844],"Claim","Relationship to evidence","Illustrative distance",[846,847,848],"The sensor reports 38.2 °C.","Direct reading","Low",[850,851,852],"The system is running hotter than yesterday.","Requires comparison with yesterday's reading","Moderate",[854,855,856],"The cooling system is failing.","Requires causal inference and alternative explanations","High",{},{"id":859,"data":860,"type":240,"tunes":862},"eYar059d69",{"text":861},"The original sensor value directly supports the first claim. It does not by itself prove the third.",{},{"id":864,"data":865,"type":281,"tunes":866},"eCnFlByvHt",{"code":328},{},{"id":868,"data":869,"type":42,"tunes":871},"aXGkK6BpkF",{"text":870,"level":234},"Where the Example Stops Working",{},{"id":873,"data":874,"type":240,"tunes":876},"lAwb9QZ_yB",{"text":875},"Evidence Distance is not simply a count of sentences or processing steps.",{},{"id":878,"data":879,"type":240,"tunes":881},"-SzYko9oIH",{"text":880},"One mathematically valid transformation may preserve evidence almost perfectly, while a single speculative interpretation may create a much larger epistemic jump.",{},{"id":883,"data":884,"type":240,"tunes":886},"XOQloBndLw",{"text":885},"The concept therefore includes both distance by transformation and distance by inference.",{},{"id":888,"data":889,"type":42,"tunes":891},"EOXbS4TAOR",{"text":890,"level":234},"Direct Answer",{},{"id":893,"data":894,"type":240,"tunes":896},"HcVL8vGM74",{"text":895},"Evidence Distance increases when a claim depends on more transformations, more assumptions, weaker provenance or stronger inference than the underlying evidence directly warrants.",{},{"id":898,"data":899,"type":281,"tunes":900},"_f7WXjUOsS",{"code":363},{},{"id":902,"data":903,"type":240,"tunes":905},"zUEC-cDhqm",{"text":904},"A longer distance does not automatically make a claim false. It means the reasoning chain requires more validation.",{},{"id":907,"data":908,"type":42,"tunes":910},"ix7TUGNz_W",{"text":909,"level":234},"Why This Matters for AI",{},{"id":912,"data":913,"type":240,"tunes":915},"wrAy5PLtLB",{"text":914},"LLMs are particularly good at producing fluent bridges between pieces of information. That capability is useful, but fluency can hide how far the conclusion has moved beyond the directly retrieved evidence.",{},{"id":917,"data":918,"type":240,"tunes":920},"jJseLuTWdv",{"text":919},"An answer can therefore look well sourced while containing claims whose support exists only through several unstated inferential steps.",{},{"id":922,"data":923,"type":394,"tunes":929},"ox8fok9dzs",{"link":924,"meta":925},"https:\u002F\u002Fstajic.de\u002Fblog\u002Ffrom-research-protocol-to-a-general-ai-reasoning-framework",{"image":926,"title":927,"description":928},{"url":391},"From Research Protocol to a General AI Reasoning Framework","Reliable reasoning requires explicit separation of evidence, assumptions, hypotheses and conclusions.",{},{"id":931,"data":932,"type":42,"tunes":934},"6ipse6YG1U",{"text":933,"level":234},"Four Components of Evidence Distance",{},{"id":936,"data":937,"type":42,"tunes":939},"P4xQ7o96dR",{"text":938,"level":259},"1. Transformation Distance",{},{"id":941,"data":942,"type":240,"tunes":944},"dvN8TUNEYG",{"text":943},"How many times has the original evidence been converted, summarized, normalized or reformatted before reaching the model?",{},{"id":946,"data":947,"type":42,"tunes":949},"dXofACF9_x",{"text":948,"level":259},"2. Inferential Distance",{},{"id":951,"data":952,"type":240,"tunes":954},"kkEzV3ydId",{"text":953},"How many reasoning steps are required before the evidence supports the final conclusion?",{},{"id":956,"data":957,"type":42,"tunes":959},"PPYB0RuX5_",{"text":958,"level":259},"3. Provenance Distance",{},{"id":961,"data":962,"type":240,"tunes":964},"cAVBYebJr7",{"text":963},"Can the retrieved statement still be traced to an identifiable original source, or has it passed through summaries and copies whose origin is unclear?",{},{"id":966,"data":967,"type":42,"tunes":969},"254N_99Z7X",{"text":968,"level":259},"4. Applicability Distance",{},{"id":971,"data":972,"type":240,"tunes":974},"to3KWv5WSK",{"text":973},"How much reasoning is required to establish that the evidence applies to this exact version, user, jurisdiction, time period or situation?",{},{"id":976,"data":977,"type":42,"tunes":979},"3WnVPQ6wqw",{"text":978,"level":234},"Evidence Distance in RAG",{},{"id":981,"data":982,"type":240,"tunes":984},"r8s-75EBhA",{"text":983},"RAG can shorten one important distance: the distance between a model and external knowledge.",{},{"id":986,"data":987,"type":240,"tunes":989},"I_8h52A3Yc",{"text":988},"It does not automatically shorten the distance between a retrieved passage and the final claim.",{},{"id":991,"data":992,"type":281,"tunes":993},"jDDRUu2FUK",{"code":459},{},{"id":995,"data":996,"type":394,"tunes":1002},"rm14qIVB1M",{"link":997,"meta":998},"https:\u002F\u002Fstajic.de\u002Fblog\u002Fwhere-does-an-llm-get-its-data-rag-data-sources-in-python",{"image":999,"title":1000,"description":1001},{"url":391},"Where Does an LLM Get Its Data? RAG Data Sources in Python","Retrieval brings external information into context. Evidence Distance asks what happens between that information and the claim eventually produced.",{},{"id":1004,"data":1005,"type":42,"tunes":1007},"CRUZDXYloa",{"text":1006,"level":234},"Connection to the Retrieval Trigger",{},{"id":1009,"data":1010,"type":240,"tunes":1012},"-TPiYJ569c",{"text":1011},"The Retrieval Trigger asks whether internal model knowledge is sufficient.",{},{"id":1014,"data":1015,"type":240,"tunes":1017},"_VjLi5fs4x",{"text":1016},"Evidence Distance asks a later question: after evidence has been retrieved, how directly does it support what the system wants to claim?",{},{"id":1019,"data":1020,"type":281,"tunes":1021},"EImPw6i617",{"code":488},{},{"id":1023,"data":1024,"type":394,"tunes":1030},"ZqyhB2bFtR",{"link":1025,"meta":1026},"https:\u002F\u002Fstajic.de\u002Fblog\u002Fwhen-should-an-ai-stop-trusting-its-own-knowledge-the-retrieval-trigger",{"image":1027,"title":1028,"description":1029},{"url":391},"When Should an AI Stop Trusting Its Own Knowledge? — The Retrieval Trigger","The trigger decides when evidence is needed. Evidence Distance evaluates how far the eventual claim moves beyond that evidence.",{},{"id":1032,"data":1033,"type":42,"tunes":1035},"ST2vnm6vwF",{"text":1034,"level":234},"A Practical Diagnostic",{},{"id":1037,"data":1038,"type":240,"tunes":1040},"qyoHcDcTKb",{"text":1039},"For an important AI-generated claim, reconstruct the support chain backwards.",{},{"id":1042,"data":1043,"type":281,"tunes":1044},"e3fDmLk0fe",{"code":512},{},{"id":1046,"data":1047,"type":240,"tunes":1049},"iQ60c7NRaE",{"text":1048},"If the chain breaks, the system has encountered an evidence-traceability problem.",{},{"id":1051,"data":1052,"type":240,"tunes":1054},"YQAKlNZ2gM",{"text":1053},"If the chain remains intact but requires several strong assumptions, the Evidence Distance is still high.",{},{"id":1056,"data":1057,"type":42,"tunes":1059},"clN7jFkUBf",{"text":1058,"level":234},"Example: Product Data",{},{"id":1061,"data":1062,"type":240,"tunes":1064},"CRF-A9O7Ot",{"text":1063},"Suppose a manufacturer specification says a laptop weighs 1.48 kg.",{},{"id":1066,"data":1067,"type":281,"tunes":1068},"JF3z6aF2q_",{"code":537},{},{"id":1070,"data":1071,"type":240,"tunes":1073},"vTMOOgkLPb",{"text":1072},"Now consider:",{},{"id":1075,"data":1076,"type":281,"tunes":1077},"Tuhfzor1yp",{"code":547},{},{"id":1079,"data":1080,"type":240,"tunes":1082},"JHFRGIgUJD",{"text":1081},"The specification is real. The recommendation is still several reasoning steps away from it.",{},{"id":1084,"data":1085,"type":42,"tunes":1087},"2lKERCrnW1",{"text":1086,"level":234},"Example: Scientific Evidence",{},{"id":1089,"data":1090,"type":240,"tunes":1092},"HA8DBYwUuK",{"text":1091},"A paper may report an observed association. An AI answer can incorrectly compress that into a causal statement.",{},{"id":1094,"data":1095,"type":281,"tunes":1096},"D9C2dVB7j_",{"code":567},{},{"id":1098,"data":1099,"type":240,"tunes":1101},"CKNjK25pkL",{"text":1100},"The problem is not that no evidence exists. The problem is that the final statement crosses a larger evidential distance than the study directly supports.",{},{"id":1103,"data":1104,"type":42,"tunes":1106},"sR430N3cDd",{"text":1105,"level":234},"Connection to Agent-Ready Context",{},{"id":1108,"data":1109,"type":240,"tunes":1111},"v77YyG55ko",{"text":1110},"Structured information objects can reduce avoidable Evidence Distance by preserving source identity, units, timestamps, conflicts and validity instead of forcing the model to reconstruct them from flattened text.",{},{"id":1113,"data":1114,"type":394,"tunes":1120},"EX8darCIjd",{"link":1115,"meta":1116},"https:\u002F\u002Fstajic.de\u002Fportfolio\u002Ffrom-raw-data-to-verifiable-ai-context-a-practical-architecture-for-agent-ready-information",{"image":1117,"title":1118,"description":1119},{"url":391},"From Raw Data to Verifiable AI Context: A Practical Architecture for Agent-Ready Information","Preserving provenance and validity before retrieval reduces the number of unsupported reconstruction steps required later.",{},{"id":1122,"data":1123,"type":42,"tunes":1125},"dsfuoURj1e",{"text":1124,"level":234},"A Simple Working Scale",{},{"id":1127,"data":1128,"type":318,"tunes":1143},"BLi7Yth64O",{"content":1129,"stretched":43,"withHeadings":14},[1130,1133,1135,1137,1139,1141],[1131,1132],"Level","Description",[605,1134],"Directly reported source fact",[608,1136],"Simple deterministic transformation or direct comparison",[611,1138],"One meaningful interpretive step",[614,1140],"Multiple assumptions or combined sources",[617,1142],"Strong causal, predictive or prescriptive conclusion beyond direct evidence",{},{"id":1145,"data":1146,"type":240,"tunes":1148},"8y8bzCcHBs",{"text":1147},"This scale is illustrative, not a scientific standard. Its purpose is to force an explicit question: How much did we have to add between the evidence and the claim?",{},{"id":1150,"data":1151,"type":42,"tunes":1153},"aFK0YiAlYO",{"text":1152,"level":234},"Misconceptions",{},{"id":1155,"data":1156,"type":240,"tunes":1158},"07KSk6xDx-",{"text":1157},"A citation means Evidence Distance is low. No. A citation can sit beside a claim that substantially exceeds what the cited source establishes.",{},{"id":1160,"data":1161,"type":240,"tunes":1163},"t_cf-Zfan9",{"text":1162},"Long Evidence Distance means the claim is wrong. No. Complex conclusions can be well supported, but they require a stronger reasoning chain.",{},{"id":1165,"data":1166,"type":240,"tunes":1168},"Pu9PRX67Q6",{"text":1167},"Primary sources eliminate inference. No. Primary evidence can still require interpretation.",{},{"id":1170,"data":1171,"type":240,"tunes":1173},"oF1xNsOHHq",{"text":1172},"More sources automatically reduce distance. Ten weak repetitions of the same unsupported claim do not create direct evidence.",{},{"id":1175,"data":1176,"type":42,"tunes":1178},"IkolK5baph",{"text":1177,"level":234},"Edge Case: Derived Data",{},{"id":1180,"data":1181,"type":240,"tunes":1183},"jRH36fwdR3",{"text":1182},"Some derived values are more reliable than raw human descriptions. A correctly calculated total from verified transactions can be highly dependable even though it is one transformation away from the original records.",{},{"id":1185,"data":1186,"type":240,"tunes":1188},"SXbdKiNDst",{"text":1187},"Evidence Distance therefore measures epistemic separation, not a simplistic preference for untouched data.",{},{"id":1190,"data":1191,"type":42,"tunes":1193},"mUbvdnFChM",{"text":1192,"level":234},"Limitations",{},{"id":1195,"data":1196,"type":240,"tunes":1198},"N_EskxFOep",{"text":1197},"Evidence Distance is a conceptual diagnostic, not a universal quantitative score.",{},{"id":1200,"data":1201,"type":240,"tunes":1203},"nW_-E08So_",{"text":1202},"Different domains require different treatment of inference. Mathematical proof, clinical evidence, software logs, market analysis and historical interpretation cannot share one universal numerical threshold.",{},{"id":1205,"data":1206,"type":42,"tunes":1208},"_Bd95wXkri",{"text":1207,"level":234},"What Would Change This Answer?",{},{"id":1210,"data":1211,"type":240,"tunes":1213},"OqCaMc2_OX",{"text":1212},"Future AI systems may generate explicit provenance graphs, machine-checkable claim-evidence links and typed reasoning traces. That could make parts of Evidence Distance measurable automatically.",{},{"id":1215,"data":1216,"type":240,"tunes":1218},"58Hf3O9-Iz",{"text":1217},"The underlying problem remains: a claim can move farther from its evidence even while every intermediate sentence sounds plausible.",{},{"id":1220,"data":1221,"type":42,"tunes":1223},"8WOq5oE_e_",{"text":1222,"level":234},"Conclusion",{},{"id":1225,"data":1226,"type":240,"tunes":1228},"1pr1TfwnKc",{"text":1227},"The important distinction is not merely between sourced and unsourced answers.",{},{"id":1230,"data":1231,"type":240,"tunes":1233},"54CysT_NXa",{"text":1232},"It is between claims that remain close to directly inspectable evidence and claims that depend on increasingly long chains of interpretation and assumption.",{},{"id":1235,"data":1236,"type":281,"tunes":1237},"Qrjo3WEUSL",{"code":713},{},{"id":1239,"data":1240,"type":240,"tunes":1242},"6ir8aIAfd3",{"text":1241},"Evidence Distance gives us a language for describing that space.",{},{"id":1244,"data":1245,"type":42,"tunes":1247},"vPhupXcbeN",{"text":1246,"level":234},"Primary Sources",{},{"id":1249,"data":1250,"type":240,"tunes":1252},"BaO4JmWRQL",{"text":1251},"Patrick Lewis et al., \u003Ca href=\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2005.11401\" target=\"_blank\">Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks\u003C\u002Fa> (2020).",{},{"id":1254,"data":1255,"type":240,"tunes":1257},"eg4Z4G6Cgs",{"text":1256},"Nelson F. Liu et al., \u003Ca href=\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2307.03172\" target=\"_blank\">Lost in the Middle: How Language Models Use Long Contexts\u003C\u002Fa> (2023).",{},{"id":1259,"data":1260,"type":240,"tunes":1262},"BJH7zUd4T8",{"text":1261},"W3C, \u003Ca href=\"https:\u002F\u002Fwww.w3.org\u002FTR\u002Fprov-overview\u002F\" target=\"_blank\">PROV Overview\u003C\u002Fa>. A standardized model for representing provenance relationships.",{},"Not all supported answers are equally close to their evidence. Evidence Distance is a practical way to describe how many transformations, summaries, assumptions and inferential steps separate an AI claim from directly inspectable evidence.",{"lang":7,"title":224,"content":226,"contentJson":1265,"excerpt":741},{"time":228,"blocks":1266,"version":740},[1267,1270,1273,1276,1279,1282,1285,1288,1291,1294,1297,1300,1303,1311,1314,1317,1320,1323,1326,1329,1332,1335,1338,1341,1344,1347,1350,1355,1358,1361,1364,1367,1370,1373,1376,1379,1382,1385,1388,1391,1394,1399,1402,1405,1408,1411,1416,1419,1422,1425,1428,1431,1434,1437,1440,1443,1446,1449,1452,1455,1458,1461,1464,1467,1472,1475,1485,1488,1491,1494,1497,1500,1503,1506,1509,1512,1515,1518,1521,1524,1527,1530,1533,1536,1539,1542,1545,1548,1551,1554],{"id":231,"data":1268,"type":42,"tunes":1269},{"text":233,"level":234},{},{"id":237,"data":1271,"type":240,"tunes":1272},{"text":239},{},{"id":243,"data":1274,"type":240,"tunes":1275},{"text":245},{},{"id":248,"data":1277,"type":253,"tunes":1278},{"text":250,"caption":251,"alignment":252},{},{"id":256,"data":1280,"type":260,"tunes":1281},{"title":258,"maxLevel":259,"minLevel":234},{},{"id":263,"data":1283,"type":42,"tunes":1284},{"text":265,"level":234},{},{"id":268,"data":1286,"type":240,"tunes":1287},{"text":270},{},{"id":273,"data":1289,"type":240,"tunes":1290},{"text":275},{},{"id":278,"data":1292,"type":281,"tunes":1293},{"code":280},{},{"id":284,"data":1295,"type":240,"tunes":1296},{"text":286},{},{"id":289,"data":1298,"type":42,"tunes":1299},{"text":291,"level":234},{},{"id":294,"data":1301,"type":240,"tunes":1302},{"text":296},{},{"id":299,"data":1304,"type":318,"tunes":1310},{"content":1305,"stretched":43,"withHeadings":14},[1306,1307,1308,1309],[303,304,305],[307,308,309],[311,312,313],[315,316,317],{},{"id":321,"data":1312,"type":240,"tunes":1313},{"text":323},{},{"id":326,"data":1315,"type":281,"tunes":1316},{"code":328},{},{"id":331,"data":1318,"type":42,"tunes":1319},{"text":333,"level":234},{},{"id":336,"data":1321,"type":240,"tunes":1322},{"text":338},{},{"id":341,"data":1324,"type":240,"tunes":1325},{"text":343},{},{"id":346,"data":1327,"type":240,"tunes":1328},{"text":348},{},{"id":351,"data":1330,"type":42,"tunes":1331},{"text":353,"level":234},{},{"id":356,"data":1333,"type":240,"tunes":1334},{"text":358},{},{"id":361,"data":1336,"type":281,"tunes":1337},{"code":363},{},{"id":366,"data":1339,"type":240,"tunes":1340},{"text":368},{},{"id":371,"data":1342,"type":42,"tunes":1343},{"text":373,"level":234},{},{"id":376,"data":1345,"type":240,"tunes":1346},{"text":378},{},{"id":381,"data":1348,"type":240,"tunes":1349},{"text":383},{},{"id":386,"data":1351,"type":394,"tunes":1354},{"link":388,"meta":1352},{"image":1353,"title":392,"description":393},{"url":391},{},{"id":397,"data":1356,"type":42,"tunes":1357},{"text":399,"level":234},{},{"id":402,"data":1359,"type":42,"tunes":1360},{"text":404,"level":259},{},{"id":407,"data":1362,"type":240,"tunes":1363},{"text":409},{},{"id":412,"data":1365,"type":42,"tunes":1366},{"text":414,"level":259},{},{"id":417,"data":1368,"type":240,"tunes":1369},{"text":419},{},{"id":422,"data":1371,"type":42,"tunes":1372},{"text":424,"level":259},{},{"id":427,"data":1374,"type":240,"tunes":1375},{"text":429},{},{"id":432,"data":1377,"type":42,"tunes":1378},{"text":434,"level":259},{},{"id":437,"data":1380,"type":240,"tunes":1381},{"text":439},{},{"id":442,"data":1383,"type":42,"tunes":1384},{"text":444,"level":234},{},{"id":447,"data":1386,"type":240,"tunes":1387},{"text":449},{},{"id":452,"data":1389,"type":240,"tunes":1390},{"text":454},{},{"id":457,"data":1392,"type":281,"tunes":1393},{"code":459},{},{"id":462,"data":1395,"type":394,"tunes":1398},{"link":464,"meta":1396},{"image":1397,"title":467,"description":468},{"url":391},{},{"id":471,"data":1400,"type":42,"tunes":1401},{"text":473,"level":234},{},{"id":476,"data":1403,"type":240,"tunes":1404},{"text":478},{},{"id":481,"data":1406,"type":240,"tunes":1407},{"text":483},{},{"id":486,"data":1409,"type":281,"tunes":1410},{"code":488},{},{"id":491,"data":1412,"type":394,"tunes":1415},{"link":493,"meta":1413},{"image":1414,"title":496,"description":497},{"url":391},{},{"id":500,"data":1417,"type":42,"tunes":1418},{"text":502,"level":234},{},{"id":505,"data":1420,"type":240,"tunes":1421},{"text":507},{},{"id":510,"data":1423,"type":281,"tunes":1424},{"code":512},{},{"id":515,"data":1426,"type":240,"tunes":1427},{"text":517},{},{"id":520,"data":1429,"type":240,"tunes":1430},{"text":522},{},{"id":525,"data":1432,"type":42,"tunes":1433},{"text":527,"level":234},{},{"id":530,"data":1435,"type":240,"tunes":1436},{"text":532},{},{"id":535,"data":1438,"type":281,"tunes":1439},{"code":537},{},{"id":540,"data":1441,"type":240,"tunes":1442},{"text":542},{},{"id":545,"data":1444,"type":281,"tunes":1445},{"code":547},{},{"id":550,"data":1447,"type":240,"tunes":1448},{"text":552},{},{"id":555,"data":1450,"type":42,"tunes":1451},{"text":557,"level":234},{},{"id":560,"data":1453,"type":240,"tunes":1454},{"text":562},{},{"id":565,"data":1456,"type":281,"tunes":1457},{"code":567},{},{"id":570,"data":1459,"type":240,"tunes":1460},{"text":572},{},{"id":575,"data":1462,"type":42,"tunes":1463},{"text":577,"level":234},{},{"id":580,"data":1465,"type":240,"tunes":1466},{"text":582},{},{"id":585,"data":1468,"type":394,"tunes":1471},{"link":587,"meta":1469},{"image":1470,"title":185,"description":590},{"url":391},{},{"id":593,"data":1473,"type":42,"tunes":1474},{"text":595,"level":234},{},{"id":598,"data":1476,"type":318,"tunes":1484},{"content":1477,"stretched":43,"withHeadings":14},[1478,1479,1480,1481,1482,1483],[602,603],[605,606],[608,609],[611,612],[614,615],[617,618],{},{"id":621,"data":1486,"type":240,"tunes":1487},{"text":623},{},{"id":626,"data":1489,"type":42,"tunes":1490},{"text":628,"level":234},{},{"id":631,"data":1492,"type":240,"tunes":1493},{"text":633},{},{"id":636,"data":1495,"type":240,"tunes":1496},{"text":638},{},{"id":641,"data":1498,"type":240,"tunes":1499},{"text":643},{},{"id":646,"data":1501,"type":240,"tunes":1502},{"text":648},{},{"id":651,"data":1504,"type":42,"tunes":1505},{"text":653,"level":234},{},{"id":656,"data":1507,"type":240,"tunes":1508},{"text":658},{},{"id":661,"data":1510,"type":240,"tunes":1511},{"text":663},{},{"id":666,"data":1513,"type":42,"tunes":1514},{"text":668,"level":234},{},{"id":671,"data":1516,"type":240,"tunes":1517},{"text":673},{},{"id":676,"data":1519,"type":240,"tunes":1520},{"text":678},{},{"id":681,"data":1522,"type":42,"tunes":1523},{"text":683,"level":234},{},{"id":686,"data":1525,"type":240,"tunes":1526},{"text":688},{},{"id":691,"data":1528,"type":240,"tunes":1529},{"text":693},{},{"id":696,"data":1531,"type":42,"tunes":1532},{"text":698,"level":234},{},{"id":701,"data":1534,"type":240,"tunes":1535},{"text":703},{},{"id":706,"data":1537,"type":240,"tunes":1538},{"text":708},{},{"id":711,"data":1540,"type":281,"tunes":1541},{"code":713},{},{"id":716,"data":1543,"type":240,"tunes":1544},{"text":718},{},{"id":721,"data":1546,"type":42,"tunes":1547},{"text":723,"level":234},{},{"id":726,"data":1549,"type":240,"tunes":1550},{"text":728},{},{"id":731,"data":1552,"type":240,"tunes":1553},{"text":733},{},{"id":736,"data":1555,"type":240,"tunes":1556},{"text":738},{},"Post erfolgreich abgerufen",{"items":1559,"source":1616,"manualIds":1617,"manualMatchedIds":1618},[1560,1567,1574,1581,1588,1595,1602,1609],{"id":1561,"slug":1562,"title":1563,"excerpt":1564,"featuredImage":1565,"publishedAt":1566},"491","ai-governance-models-data-permissions-risk-and-auditability","AI治理：模型、数据、权限、风险与可审计性","AI治理定义了在模型、提供商、数据、权限、风险、评估及整个生命周期中，谁可以批准、操作、更改和审计AI系统。","\u002Fuploads\u002F2026\u002F10\u002Fai-governance-models-data-permissions-risk-and-auditability-1791485901301-g60xyu.webp","2026-10-08T12:56:00.000Z",{"id":1568,"slug":1569,"title":1570,"excerpt":1571,"featuredImage":1572,"publishedAt":1573},"381","enterprise-grade-multi-tenant-architecture-for-an-international-platform","企业级多租户架构，适用于国际平台","Loving Rocks 是一款企业级婚礼平台，采用真正的多租户架构设计，实现租户间数据库隔离，并内置国际化支持，以确保全球可扩展性、安全性及长期运营稳定性。","\u002Fuploads\u002F2026\u002F01\u002Fenterprise-grade-multi-tenant-architecture-for-an-international-platform-1769789121298-b6v7ak.webp","2026-01-30T12:04:00.000Z",{"id":1575,"slug":1576,"title":1577,"excerpt":1578,"featuredImage":1579,"publishedAt":1580},"481","generative-ai-explained-models-retrieval-tools-and-applications-are-not-the-same-thing","生成式人工智能解析：模型、检索、工具与应用并非同一回事","生成式AI不仅仅是一个模型。了解模型、检索、工具、上下文、运行时和应用程序如何在生产AI系统中协同工作。","\u002Fuploads\u002F2026\u002F10\u002Fgenerative-ai-explained-models-retrieval-tools-and-applications-are-not-the-same-thing-1791475411822-pp0dvz.webp","2026-10-08T12:00:00.000Z",{"id":1582,"slug":1583,"title":1584,"excerpt":1585,"featuredImage":1586,"publishedAt":1587},"498","can-evidence-survive-the-ai-pipeline-the-evidence-preservation-chain","证据能否在人工智能流水线中存续？——证据保全链","证据在源头可能是正确的，但在经过解析、规范化、分块、检索和上下文组装之后，仍可能变得不可靠。证据保全链追踪的是身份、来源、范围和含义是否在整个AI流水线中得以保留。","\u002Fuploads\u002F2026\u002F10\u002Fcan-evidence-survive-the-ai-pipeline-the-evidence-preservation-chain-1792096396174-urdg9z.webp","2026-10-15T16:31:00.000Z",{"id":1589,"slug":1590,"title":1591,"excerpt":1592,"featuredImage":1593,"publishedAt":1594},"482","adr-vs-nfr-architecture-decisions-and-system-quality-are-not-the-same-thing","ADR 与 NFR：架构决策与系统质量并非同一回事","ADR 与 NFR 解析：了解系统质量需求如何驱动架构决策、ADR 如何记录权衡取舍，以及为什么验证保持独立。","\u002Fuploads\u002F2026\u002F10\u002Fadr-vs-nfr-architecture-decisions-and-system-quality-are-not-the-same-thing-1791475921511-6zgen1.webp","2026-10-08T12:11:00.000Z",{"id":1596,"slug":1597,"title":1598,"excerpt":1599,"featuredImage":1600,"publishedAt":1601},"494","air-gapped-ai-how-ai-systems-work-without-internet-or-cloud-access","气隙AI：AI系统如何在没有互联网或云访问的情况下工作","气隙AI在隔离的安全域内运行模型、RAG和AI应用，无需互联网或云依赖。了解模型、数据、更新和工具如何离线运行。","\u002Fuploads\u002F2026\u002F10\u002Fair-gapped-ai-how-ai-systems-work-without-internet-or-cloud-access-1791487983978-e6xqf0.webp","2026-10-08T11:32:00.000Z",{"id":1603,"slug":1604,"title":1605,"excerpt":1606,"featuredImage":1607,"publishedAt":1608},"492","mcp-explained-what-it-connects-what-it-does-not-do-and-where-it-fits","MCP 解析：它连接什么、不做什么以及它适用于何处","模型上下文协议通过标准的客户端-服务器边界，将AI应用程序连接到外部工具、资源和提示。了解MCP能做什么、不能做什么，以及它在智能体架构中的定位。","\u002Fuploads\u002F2026\u002F10\u002Fmcp-explained-what-it-connects-what-it-does-not-do-and-where-it-fits-1791486640275-7ub1cq.webp","2026-10-08T15:09:00.000Z",{"id":1610,"slug":1611,"title":1612,"excerpt":1613,"featuredImage":1614,"publishedAt":1615},"495","sovereign-ai-control-of-models-data-infrastructure-and-dependencies","主权人工智能：模型、数据、基础设施与依赖关系的控制","主权人工智能关乎对模型、数据、基础设施、软件、运营和战略依赖的有效控制——而不仅仅是人工智能模型托管在哪里。","\u002Fuploads\u002F2026\u002F10\u002Fsovereign-ai-control-of-models-data-infrastructure-and-dependencies-1791488833132-niy85x.webp","2026-10-08T15:45:00.000Z","fallback",[],[]]