<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0"><channel><title><![CDATA[上下文污染比上下文不够更难排查]]></title><description><![CDATA[<p dir="auto">我们知识库问答经常答错，但不是没检索到，而是检索到太多相似旧文档。这个怎么治？</p>
]]></description><link>https://localaihub.com/topic/189/上下文污染比上下文不够更难排查</link><generator>RSS for Node</generator><lastBuildDate>Wed, 03 Jun 2026 19:24:01 GMT</lastBuildDate><atom:link href="https://localaihub.com/topic/189.rss" rel="self" type="application/rss+xml"/><pubDate>Wed, 13 May 2026 20:36:00 GMT</pubDate><ttl>60</ttl><item><title><![CDATA[Reply to 上下文污染比上下文不够更难排查 on Fri, 15 May 2026 00:30:00 GMT]]></title><description><![CDATA[<p dir="auto">对，别先调温度。温度不是清洁剂。</p>
]]></description><link>https://localaihub.com/post/2066</link><guid isPermaLink="true">https://localaihub.com/post/2066</guid><dc:creator><![CDATA[小风扇]]></dc:creator><pubDate>Fri, 15 May 2026 00:30:00 GMT</pubDate></item><item><title><![CDATA[Reply to 上下文污染比上下文不够更难排查 on Thu, 14 May 2026 23:34:00 GMT]]></title><description><![CDATA[<p dir="auto">所以先清文档状态，再改检索过滤，再审 prompt 示例。</p>
]]></description><link>https://localaihub.com/post/2065</link><guid isPermaLink="true">https://localaihub.com/post/2065</guid><dc:creator><![CDATA[不想写周报]]></dc:creator><pubDate>Thu, 14 May 2026 23:34:00 GMT</pubDate></item><item><title><![CDATA[Reply to 上下文污染比上下文不够更难排查 on Thu, 14 May 2026 20:34:00 GMT]]></title><description><![CDATA[<p dir="auto">权限污染也要小心。A 部门资料被塞进 B 用户上下文，模型不一定主动拒绝。</p>
]]></description><link>https://localaihub.com/post/2064</link><guid isPermaLink="true">https://localaihub.com/post/2064</guid><dc:creator><![CDATA[nora]]></dc:creator><pubDate>Thu, 14 May 2026 20:34:00 GMT</pubDate></item><item><title><![CDATA[Reply to 上下文污染比上下文不够更难排查 on Thu, 14 May 2026 18:58:00 GMT]]></title><description><![CDATA[<p dir="auto">示例也要版本化。业务规则变了，few-shot 不更新，比没示例更危险。</p>
]]></description><link>https://localaihub.com/post/2063</link><guid isPermaLink="true">https://localaihub.com/post/2063</guid><dc:creator><![CDATA[melo]]></dc:creator><pubDate>Thu, 14 May 2026 18:58:00 GMT</pubDate></item><item><title><![CDATA[Reply to 上下文污染比上下文不够更难排查 on Thu, 14 May 2026 18:05:00 GMT]]></title><description><![CDATA[<p dir="auto">我们有 few-shot 示例，可能有旧口径。</p>
]]></description><link>https://localaihub.com/post/2062</link><guid isPermaLink="true">https://localaihub.com/post/2062</guid><dc:creator><![CDATA[小蓝]]></dc:creator><pubDate>Thu, 14 May 2026 18:05:00 GMT</pubDate></item><item><title><![CDATA[Reply to 上下文污染比上下文不够更难排查 on Thu, 14 May 2026 15:13:00 GMT]]></title><description><![CDATA[<p dir="auto">还有系统提示污染。开发测试时塞的“示例答案”忘了删，线上模型会照着学。</p>
]]></description><link>https://localaihub.com/post/2061</link><guid isPermaLink="true">https://localaihub.com/post/2061</guid><dc:creator><![CDATA[阿航]]></dc:creator><pubDate>Thu, 14 May 2026 15:13:00 GMT</pubDate></item><item><title><![CDATA[Reply to 上下文污染比上下文不够更难排查 on Thu, 14 May 2026 13:42:00 GMT]]></title><description><![CDATA[<p dir="auto">能缓解一点，不能替代数据清理。强模型也会被坏上下文带偏。</p>
]]></description><link>https://localaihub.com/post/2060</link><guid isPermaLink="true">https://localaihub.com/post/2060</guid><dc:creator><![CDATA[Grace]]></dc:creator><pubDate>Thu, 14 May 2026 13:42:00 GMT</pubDate></item><item><title><![CDATA[Reply to 上下文污染比上下文不够更难排查 on Thu, 14 May 2026 12:03:00 GMT]]></title><description><![CDATA[<p dir="auto">用更强模型能解决吗？</p>
]]></description><link>https://localaihub.com/post/2059</link><guid isPermaLink="true">https://localaihub.com/post/2059</guid><dc:creator><![CDATA[普通网友A]]></dc:creator><pubDate>Thu, 14 May 2026 12:03:00 GMT</pubDate></item><item><title><![CDATA[Reply to 上下文污染比上下文不够更难排查 on Thu, 14 May 2026 09:41:00 GMT]]></title><description><![CDATA[<p dir="auto">我见过两个文件标题一样，一个是 2022，一个是 2025，模型直接混合回答。</p>
]]></description><link>https://localaihub.com/post/2058</link><guid isPermaLink="true">https://localaihub.com/post/2058</guid><dc:creator><![CDATA[半截薯条]]></dc:creator><pubDate>Thu, 14 May 2026 09:41:00 GMT</pubDate></item><item><title><![CDATA[Reply to 上下文污染比上下文不够更难排查 on Thu, 14 May 2026 06:55:00 GMT]]></title><description><![CDATA[<p dir="auto">只放进去还不够。检索层也要过滤，比如 status=published，effective_date &lt;= now。</p>
]]></description><link>https://localaihub.com/post/2057</link><guid isPermaLink="true">https://localaihub.com/post/2057</guid><dc:creator><![CDATA[陈一]]></dc:creator><pubDate>Thu, 14 May 2026 06:55:00 GMT</pubDate></item><item><title><![CDATA[Reply to 上下文污染比上下文不够更难排查 on Thu, 14 May 2026 04:16:00 GMT]]></title><description><![CDATA[<p dir="auto">那模型不知道新旧。把 source title、status、updated_at 放进去，并明确优先级。</p>
]]></description><link>https://localaihub.com/post/2056</link><guid isPermaLink="true">https://localaihub.com/post/2056</guid><dc:creator><![CDATA[林小北]]></dc:creator><pubDate>Thu, 14 May 2026 04:16:00 GMT</pubDate></item><item><title><![CDATA[Reply to 上下文污染比上下文不够更难排查 on Thu, 14 May 2026 03:20:00 GMT]]></title><description><![CDATA[<p dir="auto">有更新时间，但 prompt 没用。</p>
]]></description><link>https://localaihub.com/post/2055</link><guid isPermaLink="true">https://localaihub.com/post/2055</guid><dc:creator><![CDATA[不想写周报]]></dc:creator><pubDate>Thu, 14 May 2026 03:20:00 GMT</pubDate></item><item><title><![CDATA[Reply to 上下文污染比上下文不够更难排查 on Thu, 14 May 2026 00:46:00 GMT]]></title><description><![CDATA[<p dir="auto">先看文档治理。旧制度有没有下架？草稿有没有标状态？更新时间有没有进 metadata？</p>
]]></description><link>https://localaihub.com/post/2054</link><guid isPermaLink="true">https://localaihub.com/post/2054</guid><dc:creator><![CDATA[小吴]]></dc:creator><pubDate>Thu, 14 May 2026 00:46:00 GMT</pubDate></item><item><title><![CDATA[Reply to 上下文污染比上下文不够更难排查 on Wed, 13 May 2026 22:22:00 GMT]]></title><description><![CDATA[<p dir="auto">这就是上下文污染。召回里混进旧版本、重复文档、草稿，模型会把它们当同等级证据。</p>
]]></description><link>https://localaihub.com/post/2053</link><guid isPermaLink="true">https://localaihub.com/post/2053</guid><dc:creator><![CDATA[小风扇]]></dc:creator><pubDate>Wed, 13 May 2026 22:22:00 GMT</pubDate></item></channel></rss>