<?xml version="1.0" encoding="utf-8"?>
<rss version="2.0" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/">
    <channel>
        <title>NotionNext BLOG</title>
        <link>https://小鲸鱼.top/</link>
        <description>这是一个由NotionNext生成的站点</description>
        <lastBuildDate>Sun, 18 Jun 2023 18:34:24 GMT</lastBuildDate>
        <docs>https://validator.w3.org/feed/docs/rss2.html</docs>
        <generator>https://github.com/jpmonette/feed</generator>
        <language>zh-CN</language>
        <copyright>All rights reserved 2023, 小鲸鱼</copyright>
        <item>
            <title><![CDATA[模板使用说明]]></title>
            <link>https://小鲸鱼.top/article/guide</link>
            <guid>https://小鲸鱼.top/article/guide</guid>
            <pubDate>Fri, 02 Jul 2021 00:00:00 GMT</pubDate>
            <description><![CDATA[<- 点击标题OPEN按钮：查看此Notion数据库的使用方法]]></description>
            <content:encoded><![CDATA[<div id="container" class="mx-auto undefined"><main class="notion light-mode notion-page notion-block-0177447a54544a5cbf545b808ffccce3"><div class="notion-viewport"></div><div class="notion-collection-page-properties"></div><h2 class="notion-h notion-h1 notion-h-indent-0 notion-block-fb013e43d4724855acc8877b207d9893" data-id="fb013e43d4724855acc8877b207d9893"><span><div id="fb013e43d4724855acc8877b207d9893" class="notion-header-anchor"></div><a class="notion-hash-link" href="#fb013e43d4724855acc8877b207d9893" title="模板使用说明"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">模板使用说明</span></span></h2><div class="notion-text notion-block-93d1d694a1f541a1bc91d04209679a7a">您的NotionNext项目必须绑定一个Notion数据库才能使用。</div><div class="notion-text notion-block-866c1150500241b68505b91d2f64b31f">请复制该模板到您的Notion中，并按照模板格式创建文章：</div><div class="notion-row"><a target="_blank" rel="noopener noreferrer" class="notion-bookmark notion-block-11d0a043bff346aba5a246c097826c74" href="https://tanghh.notion.site/02ab3b8678004aa69e9e415905ef32a5?v=b7eb215720224ca5827bfaa5ef82cf2d"><div><div class="notion-bookmark-title">NOTION BLOG</div><div class="notion-bookmark-description">演示站点描述</div><div class="notion-bookmark-link"><div class="notion-bookmark-link-text">https://tanghh.notion.site/02ab3b8678004aa69e9e415905ef32a5?v=b7eb215720224ca5827bfaa5ef82cf2d</div></div></div><div class="notion-bookmark-image"><img style="object-fit:cover" src="https://tanghh.notion.site/image/https%3A%2F%2Fp1.qhimg.com%2Fbdm%2F1600_900_100%2Ft01e66d486b8e826270.jpg?table=collection&amp;id=4379bc14-5d22-453b-a153-12639616fc01&amp;spaceId=6c096b44-beb9-48ee-8f92-1efdde47f3a3&amp;width=2000&amp;userId=&amp;cache=v2" alt="NOTION BLOG" loading="lazy" decoding="async"/></div></a></div><div class="notion-blank notion-block-fd78cbcc085f4e1fa7855a3bd06e4725"> </div><div class="notion-sync-block notion-block-12cb65a227cc4407a1818adf6a529f30"><h3 class="notion-h notion-h2 notion-block-1ac468f9f4aa4e8c90923f762f980bbc" data-id="1ac468f9f4aa4e8c90923f762f980bbc"><span><div id="1ac468f9f4aa4e8c90923f762f980bbc" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1ac468f9f4aa4e8c90923f762f980bbc" title="数据库字段说明"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">数据库字段说明</span></span></h3><div class="notion-text notion-block-8c54d490dea34051aa1d03f7e5979c9e">Notion数据库中，每条数据都将有以下属性🤔：</div><table class="notion-simple-table notion-block-4f194d6f48eb40988a27aa7a6a99c834"><tbody><tr class="notion-simple-table-row notion-block-31ad79b35ce34acbac71e1f8674e795d"><td class="" style="width:116.99479675292969px"><div class="notion-simple-table-cell">属性</div></td><td class="" style="width:46.989585876464844px"><div class="notion-simple-table-cell">必填</div></td><td class="" style="width:239.00001525878906px"><div class="notion-simple-table-cell">说明</div></td><td class="" style="width:315.66668701171875px"><div class="notion-simple-table-cell">备注</div></td></tr><tr class="notion-simple-table-row notion-block-79f142dae7c24eda80737449cac6c197"><td class="" style="width:116.99479675292969px"><div class="notion-simple-table-cell"><code class="notion-inline-code">title</code></div></td><td class="" style="width:46.989585876464844px"><div class="notion-simple-table-cell"> 是</div></td><td class="" style="width:239.00001525878906px"><div class="notion-simple-table-cell"> 文章标题</div></td><td class="" style="width:315.66668701171875px"><div class="notion-simple-table-cell">ㅤ</div></td></tr><tr class="notion-simple-table-row notion-block-4438044f9067428e89d53fbc317c2231"><td class="" style="width:116.99479675292969px"><div class="notion-simple-table-cell"><code class="notion-inline-code">status</code></div></td><td class="" style="width:46.989585876464844px"><div class="notion-simple-table-cell">是</div></td><td class="" style="width:239.00001525878906px"><div class="notion-simple-table-cell">发布状态</div></td><td class="" style="width:315.66668701171875px"><div class="notion-simple-table-cell">（仅当状态为<code class="notion-inline-code">Published</code> 时会被 展示）</div></td></tr><tr class="notion-simple-table-row notion-block-2e8f56170cbe4d44b206ac1e8a6e2bb5"><td class="" style="width:116.99479675292969px"><div class="notion-simple-table-cell"><code class="notion-inline-code">type</code></div></td><td class="" style="width:46.989585876464844px"><div class="notion-simple-table-cell">是</div></td><td class="" style="width:239.00001525878906px"><div class="notion-simple-table-cell">页面类型 (博文<code class="notion-inline-code">Post</code> / 单页(<code class="notion-inline-code">Page</code>)</div></td><td class="" style="width:315.66668701171875px"><div class="notion-simple-table-cell">单页不会在博文列表显示 。</div></td></tr><tr class="notion-simple-table-row notion-block-cc08ec2f62914d4a8cf97a33560be8a2"><td class="" style="width:116.99479675292969px"><div class="notion-simple-table-cell"><code class="notion-inline-code">summary</code></div></td><td class="" style="width:46.989585876464844px"><div class="notion-simple-table-cell">否</div></td><td class="" style="width:239.00001525878906px"><div class="notion-simple-table-cell">内容摘要</div></td><td class="" style="width:315.66668701171875px"><div class="notion-simple-table-cell">搜索和简略显示会用到</div></td></tr><tr class="notion-simple-table-row notion-block-a5fe285452214ae8876400ae768f21f4"><td class="" style="width:116.99479675292969px"><div class="notion-simple-table-cell"><code class="notion-inline-code">date</code></div></td><td class="" style="width:46.989585876464844px"><div class="notion-simple-table-cell">否</div></td><td class="" style="width:239.00001525878906px"><div class="notion-simple-table-cell">发布日期</div></td><td class="" style="width:315.66668701171875px"><div class="notion-simple-table-cell">在V3.3.9之前的版本此项为必填。</div></td></tr><tr class="notion-simple-table-row notion-block-32785258daa24522b256450b76e56588"><td class="" style="width:116.99479675292969px"><div class="notion-simple-table-cell"><code class="notion-inline-code">category</code></div></td><td class="" style="width:46.989585876464844px"><div class="notion-simple-table-cell">否</div></td><td class="" style="width:239.00001525878906px"><div class="notion-simple-table-cell">文章分类</div></td><td class="" style="width:315.66668701171875px"><div class="notion-simple-table-cell">可以自定义</div></td></tr><tr class="notion-simple-table-row notion-block-6172a7239ac7495cafd8fee9cc026140"><td class="" style="width:116.99479675292969px"><div class="notion-simple-table-cell"><code class="notion-inline-code">tags</code></div></td><td class="" style="width:46.989585876464844px"><div class="notion-simple-table-cell">否</div></td><td class="" style="width:239.00001525878906px"><div class="notion-simple-table-cell">文章标签</div></td><td class="" style="width:315.66668701171875px"><div class="notion-simple-table-cell">可多个，建议不要太多</div></td></tr><tr class="notion-simple-table-row notion-block-577a4862779a4ad08496a71d10dc4500"><td class="" style="width:116.99479675292969px"><div class="notion-simple-table-cell"><code class="notion-inline-code">slug</code></div></td><td class="" style="width:46.989585876464844px"><div class="notion-simple-table-cell">否</div></td><td class="" style="width:239.00001525878906px"><div class="notion-simple-table-cell">文章短路径</div></td><td class="" style="width:315.66668701171875px"><div class="notion-simple-table-cell"> (每篇文章唯一，请勿 重复）</div></td></tr><tr class="notion-simple-table-row notion-block-0af15a5de56245a8822f7a4cd9fa9370"><td class="" style="width:116.99479675292969px"><div class="notion-simple-table-cell"><code class="notion-inline-code">icon</code></div></td><td class="" style="width:46.989585876464844px"><div class="notion-simple-table-cell">否</div></td><td class="" style="width:239.00001525878906px"><div class="notion-simple-table-cell">菜单栏图标(仅当<code class="notion-inline-code">Page</code>类型有效)</div></td><td class="" style="width:315.66668701171875px"><div class="notion-simple-table-cell">可以参考：<a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://fontawesome.com/v6/search">图标库地址</a></div></td></tr><tr class="notion-simple-table-row notion-block-03aefeeaaa57477d9291fcb2bef8e4a8"><td class="" style="width:116.99479675292969px"><div class="notion-simple-table-cell"><code class="notion-inline-code">password</code></div></td><td class="" style="width:46.989585876464844px"><div class="notion-simple-table-cell"> 否</div></td><td class="" style="width:239.00001525878906px"><div class="notion-simple-table-cell"> 文章加锁</div></td><td class="" style="width:315.66668701171875px"><div class="notion-simple-table-cell"> 需要输入密码才允许访问</div></td></tr></tbody></table><div class="notion-blank notion-block-9e670c3864f04073aad91a37be0baa06"> </div></div><h2 class="notion-h notion-h1 notion-h-indent-0 notion-block-224417a9b8bb4e37a578cd404d93600b" data-id="224417a9b8bb4e37a578cd404d93600b"><span><div id="224417a9b8bb4e37a578cd404d93600b" class="notion-header-anchor"></div><a class="notion-hash-link" href="#224417a9b8bb4e37a578cd404d93600b" title="更多说明帮助"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">更多说明帮助</span></span></h2><div class="notion-text notion-block-cc870bae83e0424a85c3b54bc35b0371">请访问教程</div><div class="notion-row"><a target="_blank" rel="noopener noreferrer" class="notion-bookmark notion-block-1b0d5599bcaf45e3ac2479d90bc1d60b" href="https://tangly1024.com/article/notion-next-guide"><div><div class="notion-bookmark-title">NotionNext 操作手册 | TANGLY’s BLOG</div><div class="notion-bookmark-description">介绍NotionNext的所有功能和常见问题说明</div><div class="notion-bookmark-link"><div class="notion-bookmark-link-icon"><img src="https://tangly1024.com/favicon.ico" alt="NotionNext 操作手册 | TANGLY’s BLOG" loading="lazy" decoding="async"/></div><div class="notion-bookmark-link-text">https://tangly1024.com/article/notion-next-guide</div></div></div><div class="notion-bookmark-image"><img style="object-fit:cover" src="https://images.unsplash.com/photo-1587620931276-d97f425f62b9?ixlib=rb-4.0.3&amp;q=80&amp;fm=jpg&amp;crop=entropy&amp;cs=tinysrgb" alt="NotionNext 操作手册 | TANGLY’s BLOG" loading="lazy" decoding="async"/></div></a></div></main></div>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[GPT-4最全攻略来袭，OpenAI官方发布，六个月攒下来的使用经验都在里面了]]></title>
            <link>https://小鲸鱼.top/article/openaigpt4</link>
            <guid>https://小鲸鱼.top/article/openaigpt4</guid>
            <pubDate>Sun, 18 Jun 2023 00:00:00 GMT</pubDate>
            <description><![CDATA[GPT-4官方推荐攻略，ChatGPT4.0，OpenAI]]></description>
            <content:encoded><![CDATA[<div id="container" class="mx-auto undefined"><main class="notion light-mode notion-page notion-block-b8c519968e554daf8a43dfb13ab2397e"><div class="notion-viewport"></div><div class="notion-collection-page-properties"></div><div class="notion-text notion-block-a4d8001b61f14f1ebad4ebb36fb07aba">GPT-4官方使用指南炸裂登场啦！</div><div class="notion-text notion-block-c9d93e5dd0ac499592e52edd1e67839d">你没听错，这次不需要自己做笔记了，OpenAI亲自帮你整理了一份。</div><div class="notion-text notion-block-89639920358046bc9e848feb3dd98108">据说汇聚了大伙儿6个月的使用经验，你、我、他的提示诀窍都融汇其中。</div><div class="notion-text notion-block-220da48f16dd429593559e8c3d84068b">虽然总结下来只有<b>六大策略</b>，但该有的细节可绝不含糊。</div><div class="notion-text notion-block-629dd08bd6d641138abb309aad3b78f3">不仅普通GPT-4用户可以在这份秘籍中get提示技巧，或许应用开发者也可以找到些许灵感。</div><div class="notion-text notion-block-06804a0f3be049d1aa8cec09ddaacdfa">网友们纷纷评论，给出了自己的“读后感”：</div><blockquote class="notion-quote notion-block-5407e26073994fd7bfe6f8005a1bbd14"><div>好有意思啊！总结来说，这些技巧的核心思想主要有两点。一是我们得写得更具体一些，给一些细节的提示。其次，对于那些复杂的任务，我们可以把它们拆分成一些小的提示来完成。</div></blockquote><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-3302cc90d59647fcb74986e4ea943db2"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column;height:100%"><img style="object-fit:cover" src="https://www.notion.so/image/https%3A%2F%2Fs3-us-west-2.amazonaws.com%2Fsecure.notion-static.com%2F95bfcd09-3011-4666-b6f7-b4faeb351791%2FUntitled.png?table=block&amp;id=3302cc90-d596-47fc-b749-86e4ea943db2" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-a8b6a20e20634109984a69adffb5f955">OpenAI表示，这份攻略目前仅针对GPT-4。（当然，你也可以在其它GPT模型上试试？）</div><div class="notion-text notion-block-3499c81e872946d9b7c8c15e9f1cddf4">赶紧瞧瞧，这份秘籍里究竟都有啥好东西。</div><h3 class="notion-h notion-h2 notion-h-indent-0 notion-block-58e0d3e1843c45d5b220946fe9cf04d3" data-id="58e0d3e1843c45d5b220946fe9cf04d3"><span><div id="58e0d3e1843c45d5b220946fe9cf04d3" class="notion-header-anchor"></div><a class="notion-hash-link" href="#58e0d3e1843c45d5b220946fe9cf04d3" title="6大干货技巧全在这"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title"><b>6大干货技巧全在这</b></span></span></h3><h4 class="notion-h notion-h3 notion-h-indent-1 notion-block-decfe77d3ce04e67898386936d1e2c93" data-id="decfe77d3ce04e67898386936d1e2c93"><span><div id="decfe77d3ce04e67898386936d1e2c93" class="notion-header-anchor"></div><a class="notion-hash-link" href="#decfe77d3ce04e67898386936d1e2c93" title="策略一：写清楚指令"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title"><b>策略一：写清楚指令</b></span></span></h4><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-c2ee5d7a8a344f3b9f550e5897450ebf"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:518px;max-width:100%;flex-direction:column"><img style="object-fit:cover" src="https://www.notion.so/image/https%3A%2F%2Fs3-us-west-2.amazonaws.com%2Fsecure.notion-static.com%2Fe29099c9-8e87-413c-a4ea-4c545ad6fbd6%2FUntitled.png?table=block&amp;id=c2ee5d7a-8a34-4f3b-9f55-0e5897450ebf" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-8a39d8ec78b04b1d8eaa280912bc6c99">要知道，模型可不会“读心术”，所以你得把你的要求明明白白地写出来。</div><div class="notion-text notion-block-de0f9a0264b147a9b0267a4b09d7926e">当模型输出变得太啰嗦时，你可以要求它回答简洁明了。相反地，如果输出太过简单，你可以毫不客气地要求它用专业水平来写。</div><div class="notion-text notion-block-f44ed168989e45fb88f4508de46ea13f">如果你对GPT输出的格式不满意，那就先给它展示你期望的格式，并要求它以同样的方式输出。</div><div class="notion-text notion-block-7d3d8c96e74e4feba849d79fde5c1a2c">总之，尽量别让GPT模型自己去猜你的意图，这样你得到的结果就更可能符合你的预期了。</div><div class="notion-text notion-block-ee6e8e34d865424bac0d122808f80a35"><b>实用技巧：</b></div><div class="notion-text notion-block-94194b3cf8cc4ca89af5f9bcb81aed45"><b>1、有细节才能得到更相关的答案</b></div><div class="notion-text notion-block-d4bc8eb2d0c34048a58b156a3942533a">为了使输出和输入具有强相关性，一切重要的细节信息，都可以喂给模型。</div><div class="notion-text notion-block-a1a499d78dbf4aec8b6b6687b1f31831">比如你想让GPT-4：总结会议记录</div><div class="notion-text notion-block-113aa627bf1143a69a1d381e7e53ee99">就可以尽可能在表述中增加细节：</div><blockquote class="notion-quote notion-block-c60f0943220a4293a9fc02f7e9adda55"><div>将会议记录总结成一段文字。然后编写一个Markdown列表，列出与会人员及其主要观点。最后，如果与会人员有关于下一步行动的建议，请列出来。</div></blockquote><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-2af9fe377f5542ae9b8eefd5aeb4af32"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column;height:100%"><img style="object-fit:cover" src="https://www.notion.so/image/https%3A%2F%2Fs3-us-west-2.amazonaws.com%2Fsecure.notion-static.com%2Fdf18480c-5702-4e34-aaf8-6c1d41e6db5b%2FUntitled.png?table=block&amp;id=2af9fe37-7f55-42ae-9b8e-efd5aeb4af32" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-73fd26218e5c4d54a5d895359e5916b1"><b>2、要求模型扮演特定角色</b></div><div class="notion-text notion-block-e9f0ffb29f47416faf64b8c2fca3d4c7">通过改变系统消息（system message），GPT-4会更容易扮演特定的角色，比在对话中提出要求的重视程度更高。</div><div class="notion-text notion-block-f3b69a0eee2745fe8f673fb52c6e4066">如规定它要回复一个文件，这份文件中的每个段落都要有好玩的评论：</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-c654c6ee0fc04f70a13fb447f33d076b"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column;height:100%"><img style="object-fit:cover" src="https://www.notion.so/image/https%3A%2F%2Fs3-us-west-2.amazonaws.com%2Fsecure.notion-static.com%2Fdf9f77ed-8bea-4acf-a22d-2f18ecb92170%2FUntitled.png?table=block&amp;id=c654c6ee-0fc0-4f70-a13f-b447f33d076b" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-0a07f2b1f6694d35a801f8e358b1afeb"><b>3、用分隔符清晰标示输入的不同部分</b></div><div class="notion-text notion-block-d7dd96602a154386aca34f9ef1e05ae7">用&quot;&quot;&quot;三重引号&quot;&quot;&quot;、&lt;XML标记&gt; 、节标题等分隔符标记出文本的不同部分，可以更便于模型进行不同的处理。在复杂的任务中，这种标记细节就显得格外重要。</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-32abe98295ad41c9a5d536719b6b03b8"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_ab615dabfe6d403591514ec6599b3a16@1743780481_oswg34116oswg1080oswg194_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-ade8c25483894a44add9a6fecb626b9f"><b>4、明确指定完成任务所需的步骤</b></div><div class="notion-text notion-block-a9f8d61b90a54c9296a79d032d3ef69e">有些任务按步骤进行效果更佳。因此，最好明确指定一系列步骤，这样模型就能更轻松地遵循这些步骤，并输出理想结果。比如在系统消息中设定按怎样的步骤进行回答。</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-ee6cf2a822414596b5878156e9873d26"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_d9fd6e314f6846f6a5ee30ffb70644a4@1743780481_oswg102857oswg1080oswg386_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-d13442d84ee546cd89856ebe1cd06c45"><b>5、提供示例</b></div><div class="notion-text notion-block-c6b5c3f5e51e4437a30aeededd8f9f11">如果你想让模型输出按照一种不是能够很好描述出来的特定样式，那你就可以提供示例。如提供示例后，只需要告诉它“教我耐心”，它就会按照示例的风格，将其描述得生动形象。</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-f9684500639948b9a5bb1924226acc50"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_8ef2438cd2854df086ff7b598804c510@1743780481_oswg96144oswg1080oswg368_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-225d58ab9ce94736a1e2629ee0b17065"><b>6、指定所需输出长度</b></div><div class="notion-text notion-block-7e6ed6171ac64cde858da8b0eae7eaaa">你还可以要求模型具体生成多少个单词、句子、段落、项目符号等。但是，在要求模型生成特定数量的单词/字的时候，它有可能不会那么精准。</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-844d7fe8189b45978fed70f1e319d981"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_103b29eb385841638aa35c811d5cf82e@1743780481_oswg36938oswg1080oswg193_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><h4 class="notion-h notion-h3 notion-h-indent-1 notion-block-60ea0c8355534c6c91206fb0387ac701" data-id="60ea0c8355534c6c91206fb0387ac701"><span><div id="60ea0c8355534c6c91206fb0387ac701" class="notion-header-anchor"></div><a class="notion-hash-link" href="#60ea0c8355534c6c91206fb0387ac701" title="策略二：提供参考文本"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title"><b>策略二：提供参考文本</b></span></span></h4><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-0394dd11605045dab2223afdf8e7e933"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_18aa4b7dc60d44dc9ceedf204aabdb7e@1743780481_oswg6938oswg502oswg90_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-750716413e75486b97b08a35179376ac">当涉及到深奥的话题、引用和URL等内容时，GPT模型可能会一本正经地胡说八道。</div><div class="notion-text notion-block-6ca2fda2a4a24be48b68eb1a4c9c2199">为GPT-4提供可以参考的文本，能够减少虚构性回答的出现，使回答的内容更加可靠。</div><div class="notion-text notion-block-29441554265e4a4e850bf25c21955a77"><b>实用技巧：</b></div><div class="notion-text notion-block-948792647b5047cf853a12ff5c65ae12"><b>1、让模型参照参考资料进行回答</b></div><div class="notion-text notion-block-afcf2d516f0e4cd89bffd2b96b788fda">如果我们能够向模型提供一些和问题有关的可信信息，就可以指示它用提供的信息来组织回答。</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-6715ea2d916b4bf2b7067dfd704397c6"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_bba07be1a0184e878b14a6ce1035c63e@1743780481_oswg77023oswg1080oswg284_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-087eab2e22fd48009a466ec09b98c286"><b>2、让模型引用参考资料进行回答</b></div><div class="notion-text notion-block-c4a24c7cd19945158ea1488463726da0">如果在上面的对话输入中已经补充了相关信息，那么我们还可以直接要求模型在回答中引用所提供的信息。</div><div class="notion-text notion-block-aa798baf8263476faf925ad40186efcf">这里要注意的是，可以通过编程，对让模型对输出中引用的部分进行验证注释。</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-25e65f796e1f4ae5bffa13e352803b5c"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_9fddf58fcc2b4860a5b679f6bf7d52c6@1743780481_oswg166718oswg1080oswg428_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><h4 class="notion-h notion-h3 notion-h-indent-1 notion-block-307031afc62a422aaa324125492b1ce8" data-id="307031afc62a422aaa324125492b1ce8"><span><div id="307031afc62a422aaa324125492b1ce8" class="notion-header-anchor"></div><a class="notion-hash-link" href="#307031afc62a422aaa324125492b1ce8" title="策略三：拆分复杂任务"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title"><b>策略三：拆分复杂任务</b></span></span></h4><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-c1f2a9606aad4d268e4795df4f52e911"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_f1bc7506295a49d5a6b269e656d5767d@1743780481_oswg17527oswg880oswg108_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-15f102ca2d0348549c684f30fbe5147a">相比之下，GPT-4在应对复杂任务时出错率更高。</div><div class="notion-text notion-block-239ad00f42a54ecb9d4f269571596af3">然而，我们可以采取一种巧妙的策略，将这些复杂任务重新拆解成一系列简单任务的工作流程。</div><div class="notion-text notion-block-252fb20c032c4757ae8f43732a1a16b9">这样一来，前面任务的输出就可以被用于构建后续任务的输入。</div><div class="notion-text notion-block-5b97fdd5d46e46c2aa3c179be366453f">就像在软件工程中将一个复杂系统分解为一组模块化组件一样，将任务分解成多个模块，也可以让模型的表现更好。</div><div class="notion-text notion-block-06003ce0ae3e44e0b88baf83999c37f1"><b>实用技巧：</b></div><div class="notion-text notion-block-ace160e2d235485686c8955c1967bbb7"><b>1、进行意图分类</b></div><div class="notion-text notion-block-417d42a72c55406ba170f0b25a2302a2">对于需要处理不同情况的大量具有独立性的任务，可以先对这些任务进行分类。</div><div class="notion-text notion-block-dd2cff93d543407f98c801f52c87cdd9">然后，根据分类来确定所需的指令。</div><div class="notion-text notion-block-df670c858aaf4fc9b2fff4331c28c70d">比如，对于客户服务应用程序，可以进行查询分类（计费、技术支持、账户管理、一般查询等）。</div><div class="notion-text notion-block-164b0dca5eb643359ab142228d9407a4">当用户提出：</div><blockquote class="notion-quote notion-block-e1679e065e5c4ee4bfc6529652612484"><div>我需要让我的互联网重新恢复正常。</div></blockquote><div class="notion-text notion-block-d32dd89f681d4356ac651484328f7168">根据用户查询的分类，可以锁定用户的具体诉求了，就可以向GPT-4提供一组更具体的指令，来进行下一步操作。</div><div class="notion-text notion-block-1dae1ee698be472ab93dfe69e109799d">例如，假设用户需要在“故障排除”方面寻求帮助。</div><div class="notion-text notion-block-95637e4f4e384be3a6738581f78394f5">就可以设定下一步的方案：</div><blockquote class="notion-quote notion-block-3166989160504cbf9fcbba09a39fe1b1"><div>要求用户检查路由器的所有电缆是否已连接……</div></blockquote><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-1edb7a6663e94501bc24d2ac6e4012e7"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_ae168c989d364a8e9379de24b7b29ba5@1743780481_oswg328133oswg1080oswg794_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-7c91bbb26d2646699f23c2a55a1fdbee"><b>2、对先前对话进行概括或筛选</b></div><div class="notion-text notion-block-6fe0858504464a30987b4ce649f1c90d">由于GPT-4的对话窗口是有限制的，上下文不能太长，不能在一个对话窗口中无限进行下去。</div><div class="notion-text notion-block-20c52198ee274d7b96f1fbfd9707f5eb">但也不是没有解决办法。</div><div class="notion-text notion-block-42648704e86a431ebb0c54bd5fa87469">方法之一是对先前的对话进行概括。一旦输入的文本长度达到预定的阈值，就可以触发一个查询，概括对话的一部分，被概括出来的这部分内容可以变成系统消息的一部分。</div><div class="notion-text notion-block-345608f7db674b6a95284d3961ae0adb">此外，可以在对话过程中就在后台对前面的对话进行概括。</div><div class="notion-text notion-block-498fcff5791a4f90a65fb1e68c61f99d">另一种方法是检索先前的对话，使用基于嵌入的搜索实现高效的知识检索。</div><div class="notion-text notion-block-e2df3c0da54b46e0855ce1855b1885e0"><b>3、逐段概括长文档，并递归构建完整概述</b></div><div class="notion-text notion-block-8f3aa12bf1cb4a369b8cc64ef6bede3f">还是文本过长的问题。</div><div class="notion-text notion-block-8605aa7369d34e359edea840e4580d1f">比如你要让GPT-4概括一本书，就可以使用一系列查询来概括这本书的每个部分。</div><div class="notion-text notion-block-02e31ec7b4e94436ac632936a1d1f85f">然后将部分概述连接起来进行总结，汇成一个总的答案。</div><div class="notion-text notion-block-8f8f4baa38eb42d08439af4bf49141d8">这个过程可以递归进行，直到整本书被概括。</div><div class="notion-text notion-block-374968b5e1b54e2d87576f15f01028bb">但是有些部分可能要借前面部分的信息才能理解后续部分，这里有一个技巧：</div><div class="notion-text notion-block-83abcb159a0e40f981d66b10645accbe">在概括当前内容时，将文本中当前内容之前的内容概述一起总结进来，进行概括。</div><div class="notion-text notion-block-bd9097eec82b49a29232bcb076b3f3cf">简单来说，用前面部分的“摘要”+当前部分，然后进行概括。</div><div class="notion-text notion-block-cab9691f0664401f9251c12699a04501">OpenAI之前还使用基于GPT-3训练的模型，对概括书籍的效果进行了研究。</div><h4 class="notion-h notion-h3 notion-h-indent-1 notion-block-c37f61aba83548aaa0bdcb6274660ac6" data-id="c37f61aba83548aaa0bdcb6274660ac6"><span><div id="c37f61aba83548aaa0bdcb6274660ac6" class="notion-header-anchor"></div><a class="notion-hash-link" href="#c37f61aba83548aaa0bdcb6274660ac6" title="策略四：给GPT时间“思考”"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title"><b>策略四：给GPT时间“思考”</b></span></span></h4><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-659bcdb3c4444fadadf5bde0017248d5"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_e53bf9f5f38141b6b774b043e439df89@1743780481_oswg10037oswg554oswg92_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-bbb9627d2f5848498a7171ef0f1d6299">如果让你计算17乘28，你可能不会立刻知道答案，但是可以通过一些时间计算出来。</div><div class="notion-text notion-block-87bebf35db5d4079959334e872e0223a">同样的道理，当GPT-4接收到问题时，它并不会花时间仔细思考，而是试图立刻给出答案，这样就可能导致推理出错。</div><div class="notion-text notion-block-f0788c256b4141b29ad149e93e43cfa2">因此，在让模型给出答案前，可以先要求它进行一系列的推理过程，帮助它通过推理来得出正确的答案。</div><div class="notion-text notion-block-f7d1297002204537b5c0ac62474daa92"><b>实用技巧：</b></div><div class="notion-text notion-block-9a487eb811c74f1a81b9711a2e0c55e2"><b>1、让模型制定解决方案</b></div><div class="notion-text notion-block-be7504822b7b4f28a66363d5e3724814">你可能有时候会发现，当我们明确指示模型在得出结论之前从基本原理出发进行推理时，我们可以获得更好的结果。</div><div class="notion-text notion-block-da4948364e1342eb8bda7b03aa07c967">比如说，假设我们希望模型评估学生解答数学问题的方案。</div><div class="notion-text notion-block-10b52657aa924b9da189af5af8e1dad6">最直接的方法是简单地询问模型学生的解答是否正确。</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-dc6590ae0d2a408582e29b079e3e181b"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_020861246e51463cb015614ad22cb422@1743780481_oswg215093oswg1080oswg691_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-8b206c8600d64c878b9bf8eb3aec8fdf">在上图中，GPT-4认为学生的方案是正确的。</div><div class="notion-text notion-block-5a2aff45d4214693834939661fa876e7">但实际上学生的方案是错误的。</div><div class="notion-text notion-block-cc2e01a1c6a543bf9f424e2a36ee40ae">这时候就可以通过提示模型生成自己的解决方案，来让模型成功注意到这一点。</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-08d10800353c4f4a853192a76cb95188"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_b3aee9eca64a4555907bab78c03b61b9@1743780481_oswg127817oswg1080oswg341_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-e672a15a2c8641c2ace8fc67b6a2302f">在生成了自己的解决方案，进行一遍推理过后，模型意识到之前学生的解决方案不正确。</div><div class="notion-text notion-block-7b6f9507b5d64afd99f7a065b842f13a"><b>2、隐藏推理过程</b></div><div class="notion-text notion-block-300b258acf9746afb47535f5e6a9888c">上面讲到了让模型进行推理，给出解决方案。</div><div class="notion-text notion-block-c0cd91617b1144a286ac34980598cb16">但在某些应用中，模型得出最终答案的推理过程不适合与用户共享。</div><div class="notion-text notion-block-3efce9d45ebd4e769d1fe1d065235dd9">比如，在作业辅导中，我们还是希望鼓励学生制定自己的解题方案，然后得出正确答案。但模型对学生解决方案的推理过程可能会向学生揭示答案。</div><div class="notion-text notion-block-a9d3362047154d8b8a380f8edeb5dec1">这时候我们就需要模型进行“内心独白”策略，让模型将输出中要对用户隐藏的部分放入结构化格式中。</div><div class="notion-text notion-block-c1b048a7327e42f2837cb2c70332b2e4">然后，在向用户呈现输出之前，对输出进行解析，并且仅使部分输出可见。</div><div class="notion-text notion-block-d87bff394a1c4ee3b70039567491c514">就像下面这个示例：</div><div class="notion-text notion-block-a19c4aae829e46e2b3e9b5483d319e35">先让模型制定自己的解决方案（因为学生的有可能是错的），然后与学生的解决方案进行对比。</div><div class="notion-text notion-block-62c7825bc5ac48a2a1eb2d39783f86ef">如果学生的答案中哪一步出错了，那就让模型针对这一步给出一点提示，而不是直接给学生完整的正确的解决方案。</div><div class="notion-text notion-block-44c0743355c24754bf3774c017480a02">如果学生还是错了，那就再进行上一步的提示。</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-57f5636229854d8f970d15e62923653b"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_816c9205e94b42728ba3d5380025fda3@1743780481_oswg241493oswg1080oswg735_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-fadd6b705b684a01bddc3b78d1c80ede">还可以使用“查询”策略，其中除了最后一步的查询以外，所有查询的输出都对用户隐藏。</div><div class="notion-text notion-block-9f2180a397504aed867d2ce8bd37e5e8">首先，我们可以要求模型自行解决问题。由于这个初始查询不需要学生的解决方案，因此可以省略掉。这也提供了额外的优势，即模型的解决方案不会受到学生解决方案偏见的影响。</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-082e10ec2c15486da79924d2d1332288"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_939da3278ad94087a0937ddbe4198360@1743780481_oswg20220oswg1080oswg130_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-054fd0cf5aa2445184ccf0ce6b790222">接下来，我们可以让模型使用所有可用信息来评估学生解决方案的正确性。</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-f73da344ce2247f9a0d16e09d08b0fd5"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_52daa0bd7558447991d3f54e3b919a3d@1743780481_oswg83547oswg1080oswg356_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-f5e9d6e460464bf3b4f0c67e5fc9b915">最后，我们可以让模型使用自己的分析来构建导师的角色。</div><blockquote class="notion-quote notion-block-d63a0dc86279429c86c47ab9ce2b3fd3"><div>你是一名数学导师。如果学生回答有误，请以不透露答案的方式向学生进行提示。如果学生答案无误，只需给他们一个鼓励性的评论。</div></blockquote><div class="notion-text notion-block-f930f5ef80ca4ddb9c55126e5b6e8521"><b>3、询问模型是否遗漏了内容</b></div><div class="notion-text notion-block-e27d68d34f2a4291bcc677dd413841b7">假设我们正在让GPT-4列出一个与特定问题相关的源文件摘录，在列出每个摘录之后，模型需要确定是继续写入下一个摘录，还是停止。</div><div class="notion-text notion-block-5feb3b1db01a4679ae1a17805dbaea0f">如果源文件很大，模型往往会过早地停止，未能列出所有相关的摘录。</div><div class="notion-text notion-block-c129d276480245449ea7bdbddb069266">在这种情况下，通常可以让模型进行后续查询，找到它在之前的处理中遗漏的摘录。</div><div class="notion-text notion-block-7d11ae1bc1f949c2a1fb20ed68323152">换而言之，模型生成的文本有可能很长，一次性生成不完，那么就可以让它进行查验，把遗漏的内容再补上。</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-c39d363d290c4e23be5792f7b983e0ee"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_d8bf0a75b5194f3192e76e40aa68ed6e@1743780481_oswg248025oswg1080oswg772_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><h4 class="notion-h notion-h3 notion-h-indent-1 notion-block-f2f288c67ebf46c9bf683dac4f6147d6" data-id="f2f288c67ebf46c9bf683dac4f6147d6"><span><div id="f2f288c67ebf46c9bf683dac4f6147d6" class="notion-header-anchor"></div><a class="notion-hash-link" href="#f2f288c67ebf46c9bf683dac4f6147d6" title="策略五：其它工具加持"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title"><b>策略五：其它工具加持</b></span></span></h4><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-bdbe2e87721947e4bc9649f542f73c51"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_2df1729f9ed7454aa649e70f68c5eb4b@1743780481_oswg8573oswg406oswg100_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-fc69a8a511b9465d83c04cb90c3bdba1">GPT-4虽然强大，但并非万能。</div><div class="notion-text notion-block-ac3f24a51d4c4e14b9150e28ef92e28a">我们可以借助其他工具来补充GPT-4的不足之处。</div><div class="notion-text notion-block-944ef8c0d28a4c37bae9f8ad00e1b2d7">比如，结合文本检索系统，或者利用代码执行引擎。</div><div class="notion-text notion-block-b271bcf813134a8088bf9074e6a4cc3f">在让GPT-4回答问题时，如果有一些任务可以由其他工具更可靠、更高效地完成，那么我们可以将这些任务交给它们来完成。这样既能发挥各自的优势，又能让GPT-4发挥最佳水平。</div><div class="notion-text notion-block-f70688dbac0246a9af53e3c38ee05558"><b>实用技巧：</b></div><div class="notion-text notion-block-69484a08555344abb4373f0a9e9b489b"><b>1、使用基于嵌入的搜索实现高效的知识检索</b></div><div class="notion-text notion-block-ca9e8454f3844420a96a606f4c27d941">这一技巧在上文中已经有所提及。</div><div class="notion-text notion-block-82ac219c21794199845a9bdd6377467f">若在模型的输入中提供额外的外部信息，有助于模型生成更好的回答。</div><div class="notion-text notion-block-1c345404d1614df7bdbb3a6f42983794">例如，如果用户询问关于一部特定电影的问题，将关于电影的信息（例如演员、导演等）添加到模型的输入中可能会很有用。</div><div class="notion-text notion-block-91bcff47d5ad43d2841718ec53a79fc9">嵌入可用于实现高效的知识检索，可以在模型运行时动态地将相关信息添加到模型的输入中。</div><div class="notion-text notion-block-8d0ab8414d4840c8bcaae568b5e8714f">文本嵌入是一种可以衡量文本字符串相关性的向量。相似或相关的字符串将比不相关的字符串更紧密地结合在一起。加上快速向量搜索算法的存在，意味着可以使用嵌入来实现高效的知识检索。</div><div class="notion-text notion-block-a9b2863c18284b4e8027d9c351e79882">特别的是，文本语料库可以分成多个部分，每个部分可以进行嵌入和存储。然后，给定一个查询，可以进行向量搜索以找到与查询最相关的语料库中的嵌入文本部分。</div><div class="notion-text notion-block-2a5f1b75b9474105a95929ea02abbf74"><b>2、使用代码执行进行更准确的计算或调用外部API</b></div><div class="notion-text notion-block-14b2ef2d32cf477b8bbd5f5613942375">不能仅依靠模型自身进行准确地计算。</div><div class="notion-text notion-block-2afb3c44ffd1409088caca1e6daba769">如果需要，可以指示模型编写和运行代码，而不是进行自主计算。</div><div class="notion-text notion-block-01b33bb77172483fb09b704c43cdf16a">可以指示模型将要运行的代码放入指定的格式中。在生成输出后，可以提取和运行代码。生成输出后，可以提取并运行代码。最后，如果需要，代码执行引擎（即Python解释器）的输出可以作为下一个输入。</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-bdd6354073504f349615a20d5ba415bb"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_9ba7ae79f9f54e509d3c95d2ba88f7f3@1743780481_oswg74010oswg1080oswg215_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-5d7f465ff9f541c49f8e0adb477dc7c8">代码执行的另一个很好的应用场景是调用外部API。</div><div class="notion-text notion-block-618fc7bec0104cfd8f6c75a5ef8c930e">如果将API的正确使用方式传达给模型，它可以编写使用该API的代码。</div><div class="notion-text notion-block-51147532cf8244e19bc484edaac658d2">可以通过向模型演示文档和/或代码示例来指导模型如何使用API。</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-abc5b1f0dd9646129992ab106cd3e63b"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_327a799d79ac4704a6ba492b08aa360a@1743780481_oswg91931oswg1080oswg305_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-ca3021f075164f508875799a3b603fbe">在这里OpenAI提出了特别警告⚠️：</div><blockquote class="notion-quote notion-block-bb7add093279413a89ad7edfb90cb5bf"><div>执行模型生成的代码在本质上来说并不安全，任何试图执行此操作的应用程序中都应采取预防措施。特别是，需要一个沙盒代码执行环境来限制不受信任的代码可能造成的危害。</div></blockquote><h4 class="notion-h notion-h3 notion-h-indent-1 notion-block-4d73dff076d0415b818e55bdce0b0705" data-id="4d73dff076d0415b818e55bdce0b0705"><span><div id="4d73dff076d0415b818e55bdce0b0705" class="notion-header-anchor"></div><a class="notion-hash-link" href="#4d73dff076d0415b818e55bdce0b0705" title="策略六：系统地测试更改"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title"><b>策略六：系统地测试更改</b></span></span></h4><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-52a6b1909dbc4acfacbd3dd982ec6261"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_2e23b945a50f48198d3b37b9abe953de@1743780481_oswg13398oswg596oswg106_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-207022d9103349209f0edea82feaca73">有时候很难确定一个改变是会让系统变得更好还是更差。</div><div class="notion-text notion-block-94f978801b554254b1805d405f6d0929">通过观察一些例子有可能会看出哪个更好，但是在样本数量较少的情况下，很难区分是真的得到了改进，还是只是随机运气。</div><div class="notion-text notion-block-64ad4f849b614b5ab211cf81475cbee5">也许这个“改变”能够提升某些输入的效果，但却会降低其它输入的效果。而评估程序(evaluation procedures，or “evals”)对于优化系统设计来说非常有用。好的评估有以下几个特点：</div><div class="notion-text notion-block-52bb7572c39c426a8f6b32dff202b086">1）代表现实世界的用法（或至少是多种用法）</div><div class="notion-text notion-block-489bed4371624b829df31bf8e4d04399">2）包含许多测试用例，可以获得更大的统计功效（参见下表）</div><div class="notion-text notion-block-fda9823eabe3455ab34fd984dc37df42">3）易于自动化或重复</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-ae8c99a9795f4cec939f6e0690ca88b8"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_2742e2a9d88e430aab90f735490ef23a@1743780481_oswg34588oswg1064oswg380_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-b9460e03663c41bba4bbc7c0e8eee0c2">对输出的评估可以是由计算机进行评估、人工评估，或者两者结合进行。计算机可以使用客观标准自动评估，也可以使用一些主观或模糊的标准，比如说用模型来评估模型。</div><div class="notion-text notion-block-2e5e42a082684ccbb0892bcc3be75e2c">OpenAI提供了一个开源软件框架——OpenAI Evals，提供了创建自动评估的工具。</div><div class="notion-text notion-block-54714bbc53644c2f9ec8811e57bd3cd3">当存在一系列质量同样高的输出时，基于模型的评估就会很有用。</div><div class="notion-text notion-block-b7ca4db001f14098a7540ab7e9e79f96"><b>实用技巧：</b></div><div class="notion-text notion-block-387679b5a2e74e3f8e688ded6d0ac49c"><b>1、参考黄金标准答案评估模型输出</b></div><div class="notion-text notion-block-8e07107173df4de4b9518f4c9cb73b05">假设已知问题的正确答案应参考一组特定的已知事实。</div><div class="notion-text notion-block-74b9c01cdf5c4090af2cf0882e131a8a">然后，我们可以询问模型答案中包含多少必需的事实。</div><div class="notion-text notion-block-941cef23be9645d189360f8983d127bf">例如，使用下面这个系统消息，</div><div class="notion-text notion-block-2e4661aae2e345f8bce48d094aff3ef4">给出必要的既定事实：</div><blockquote class="notion-quote notion-block-1e068d9d475a4b8299a61c19d961c801"><div>尼尔·阿姆斯特朗是第一个在月球上行走的人。</div><div class="notion-text notion-block-d241b9f7e3d04350b538ef329ed901db">尼尔·阿姆斯特朗第一次登上月球的日期是1969年7月21日。</div></blockquote><div class="notion-text notion-block-2ad8eff8abd04416b378f7b257807b4f">如果答案中包含既定给出的事实，模型会回答“是”。反之，模型会回答“否”，最后让模型统计有多少“是”的答案：</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-130a646c61c0458399ac746111ae6eab"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_ed55ebf27654470c851520a63b3e7939@1743780481_oswg225707oswg1080oswg624_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-6e5266ed226242efb8d2bf21158029e3">下面是包含两点既定事实的示例输入（既有事件，又有时间）：</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-b424b0519a4147aebf0a09c7266f11aa"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_c622cfeddaeb4f47b29d957193bdd60c@1743780481_oswg67134oswg1080oswg217_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-76f75fa75f1648c998941b5e9de505a2">仅满足一个既定事实的示例输入（没有时间）：</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-45c9bd405bad436fa0658901e760dc58"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_69d71e79a0734ac08f9658135213bd0a@1743780481_oswg58178oswg1080oswg223_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-8b9b6153d62c4f0a81f41b46bca04fb5">而下面这个示例输入，不包含任何一个既定事实：</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-a7a3d102bf77431082cf067d9f5aa92a"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_7d9f9078b4744083a6df707ff86521f4@1743780481_oswg71514oswg1080oswg288_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-8a635fb34f5745a3bd73637cac1a5758">这种基于模型的评估方法有许多可能的变化形式，需要跟踪候选答案与标准答案之间的重叠程度，并追踪候选答案是否与标准答案的有相矛盾的地方。</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-e94b07c84ffd469e9695247e2be57d49"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_600ce4b35364448e8ec524f7751793dd@1743780481_oswg182369oswg1080oswg449_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-977fc5563f0840b4a01226c3d8842cff">比如下面的这个示例输入，其中包含不合标准的答案，但与专家答案（标准答案）并不矛盾：</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-d386db1e01f64205aa0479470f21ba63"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_bbf675878d76450cb9bf8941696b2955@1743780481_oswg113803oswg1080oswg386_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-ab656e109a674c38ac7afd641360e0d5">下面是这个示例输入，其答案与专家答案直接矛盾（认为尼尔·阿姆斯特朗是第二个在月球上行走的人）：</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-f7fc5c5492ff4f5aab131fcc20b79cf3"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_d3e53ed3442143c0829216a6f4793e11@1743780481_oswg134550oswg1080oswg422_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><div class="notion-text notion-block-a33cc95e668e4809b31b857415f3e90c">最后一个是带有正确答案的示例输入，该输入还提供了比必要内容更多的详细信息（时间精确到了02:56，并指出了这是人类历史上的一项不朽成就）：</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-db88130ea1e64f069152ad5a19eaf902"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column"><img src="https://img.36krcdn.com/hsossms/20230612/v2_6d664e11c3174f0783e7aed524fbe732@1743780481_oswg150856oswg1080oswg459_img_000" alt="notion image" loading="lazy" decoding="async"/></div></figure><h4 class="notion-h notion-h3 notion-h-indent-1 notion-block-dbf49711249341db87c0333ab9dc376e" data-id="dbf49711249341db87c0333ab9dc376e"><span><div id="dbf49711249341db87c0333ab9dc376e" class="notion-header-anchor"></div><a class="notion-hash-link" href="#dbf49711249341db87c0333ab9dc376e" title="传送门"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title"><b>传送门</b></span></span></h4><div class="notion-text notion-block-49e3da6493104508bf019e9395f5f9bd">https://github.com/openai/evals（OpenAI Evals）</div><h4 class="notion-h notion-h3 notion-h-indent-1 notion-block-3e30ba992c3b4b92a3164a271e93e230" data-id="3e30ba992c3b4b92a3164a271e93e230"><span><div id="3e30ba992c3b4b92a3164a271e93e230" class="notion-header-anchor"></div><a class="notion-hash-link" href="#3e30ba992c3b4b92a3164a271e93e230" title="参考链接"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title"><b>参考链接</b></span></span></h4><div class="notion-text notion-block-aacc6a3f447c4d3e970c72ac1bea1bc0">[1]https://platform.openai.com/docs/guides/gpt-best-practices</div><div class="notion-text notion-block-b6ee30b05bef46aba978fa5cfe852b15">[2]https://www.reddit.com/r/OpenAI/comments/141yheo/openai_recently_added_a_gpt_best_practices_guide/</div><div class="notion-blank notion-block-41563a1d2eee49e8aa1f42369adecdd8"> </div><div class="notion-text notion-block-3afc888dfd25440aad90bd77bc044613">本文来自微信公众号<a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://mp.weixin.qq.com/s/u72wzup-5DCQrJdNNYJwDA">“量子位”（ID:QbitAI）</a>，作者：西风</div><div class="notion-blank notion-block-32bbf8968f1e4b568724f72ceb4c087d"> </div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-a51e05b4aedc407195cb58bc183010f3"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:100%;max-width:100%;flex-direction:column;height:100%"><img style="object-fit:cover" src="https://www.notion.so/image/https%3A%2F%2Fs3-us-west-2.amazonaws.com%2Fsecure.notion-static.com%2F2d1e1d1a-a44e-4dde-bfba-e461ee204671%2FUntitled.png?table=block&amp;id=a51e05b4-aedc-4071-95cb-58bc183010f3" alt="notion image" loading="lazy" decoding="async"/></div></figure></main></div>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[无slug文章]]></title>
            <link>https://小鲸鱼.top/article/33ddddd4-4ef8-48cf-b0cb-617b015feb38</link>
            <guid>https://小鲸鱼.top/article/33ddddd4-4ef8-48cf-b0cb-617b015feb38</guid>
            <pubDate>Fri, 02 Jul 2021 00:00:00 GMT</pubDate>
            <description><![CDATA[此文章的访问路径默认为notion的页面id]]></description>
            <content:encoded><![CDATA[<div id="container" class="mx-auto undefined"><main class="notion light-mode notion-page notion-block-33ddddd44ef848cfb0cb617b015feb38"><div class="notion-viewport"></div><div class="notion-collection-page-properties"></div><div class="notion-callout notion-gray_background_co notion-block-57cd2d6fe4264129bbaa199753491b18"><div class="notion-page-icon-inline notion-page-icon-span"><span class="notion-page-icon" role="img" aria-label="😀">😀</span></div><div class="notion-callout-text">这里写文章的前言：
一个简单的开头,简述这篇文章讨论的问题、目标、人物、背景是什么？并简述你给出的答案。<div class="notion-text notion-block-4ee08e6e4bed4c90a9b4bda882a1e2da">可以说说你的故事：阻碍、努力、结果成果，意外与转折。</div></div></div><div class="notion-blank notion-block-0845c41af74f4c0d8db7a9e17227bfc6"> </div><h2 class="notion-h notion-h1 notion-h-indent-0 notion-block-b4e98b22c36a4428b864f947c9578472" data-id="b4e98b22c36a4428b864f947c9578472"><span><div id="b4e98b22c36a4428b864f947c9578472" class="notion-header-anchor"></div><a class="notion-hash-link" href="#b4e98b22c36a4428b864f947c9578472" title="📝 主旨内容"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">📝 主旨内容</span></span></h2><h3 class="notion-h notion-h2 notion-h-indent-1 notion-block-4b8659cb849c46e794fe7385d7e79a2b" data-id="4b8659cb849c46e794fe7385d7e79a2b"><span><div id="4b8659cb849c46e794fe7385d7e79a2b" class="notion-header-anchor"></div><a class="notion-hash-link" href="#4b8659cb849c46e794fe7385d7e79a2b" title="观点1"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">观点1</span></span></h3><blockquote class="notion-quote notion-block-dc80ae2527a24427a29f3a25c608eda9"><div>引用的话语</div></blockquote><h3 class="notion-h notion-h2 notion-h-indent-1 notion-block-a2bbd721154b40ea8956012e5eb62133" data-id="a2bbd721154b40ea8956012e5eb62133"><span><div id="a2bbd721154b40ea8956012e5eb62133" class="notion-header-anchor"></div><a class="notion-hash-link" href="#a2bbd721154b40ea8956012e5eb62133" title="观点2"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">观点2</span></span></h3><blockquote class="notion-quote notion-block-ba1dcd8391054d5cbdf567198ed7ab19"><div>引用的话语</div></blockquote><h2 class="notion-h notion-h1 notion-h-indent-0 notion-block-66c3932adb4d43cbb6c2fc5700b0e4ea" data-id="66c3932adb4d43cbb6c2fc5700b0e4ea"><span><div id="66c3932adb4d43cbb6c2fc5700b0e4ea" class="notion-header-anchor"></div><a class="notion-hash-link" href="#66c3932adb4d43cbb6c2fc5700b0e4ea" title="🤗 总结归纳"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">🤗 总结归纳</span></span></h2><div class="notion-text notion-block-6d27868bd6c04231b69805549d1a2a47">总结文章的内容</div><h2 class="notion-h notion-h1 notion-h-indent-0 notion-block-33d40c56dc9c4753800e82987b6a325e" data-id="33d40c56dc9c4753800e82987b6a325e"><span><div id="33d40c56dc9c4753800e82987b6a325e" class="notion-header-anchor"></div><a class="notion-hash-link" href="#33d40c56dc9c4753800e82987b6a325e" title="📎 参考文章"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">📎 参考文章</span></span></h2><ul class="notion-list notion-list-disc notion-block-c5fadb5c4979427f8afd2ad53e9a8703"><li>一些引用</li></ul><ul class="notion-list notion-list-disc notion-block-87f0b567038249a0921a6c51f206ba96"><li>引用文章</li></ul><div class="notion-blank notion-block-1f2e59cf52be4d4eab5b34e6d4a04daf"> </div><div class="notion-callout notion-gray_background_co notion-block-0732fc0ad1c84991b4ddbfeed7744474"><div class="notion-page-icon-inline notion-page-icon-span"><span class="notion-page-icon" role="img" aria-label="💡">💡</span></div><div class="notion-callout-text">有关Notion安装或者使用上的问题，欢迎您在底部评论区留言，一起交流~</div></div></main></div>]]></content:encoded>
        </item>
    </channel>
</rss>