https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/08-academic-research-skills/04-deep-research/技术 × 产品 × 机器人Hugo 0.160.1 & FixIt v0.4.6-20260517062719-ed8ec5c2zh-CNruohuyang@163.com (Cygnus Yang)ruohuyang@163.com (Cygnus Yang)Sat, 23 May 2026 00:00:00 +0800https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/08-academic-research-skills/04-deep-research/07-13-agents-team/Sat, 23 May 2026 00:00:00 +0800ruohuyang@163.com (Cygnus Yang)https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/08-academic-research-skills/04-deep-research/07-13-agents-team/<p>Deep Research 是 ARS 的上游引擎——它将模糊的研究兴趣转化为可操作的 Research Question、可验证的证据体系和结构化的研究报告。这一过程不是由一个&quot;全知全能&quot;的 AI 完成的，而是由 13 个角色分工明确的 Agent 协作完成。</p> <p>本文将 13 个 Agent 按职责分组，逐一剖析它们的工作方式。</p> <h2 class="heading-element" id="agent-全景"><span>Agent 全景</span> <a href="#agent-%e5%85%a8%e6%99%af" class="heading-mark"> <svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg> </a> </h2><p>Deep Research 的 13 个 Agent 可以按职责分为五组：</p>https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/08-academic-research-skills/04-deep-research/08-seven-modes-in-action/Sat, 23 May 2026 00:00:00 +0800ruohuyang@163.com (Cygnus Yang)https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/08-academic-research-skills/04-deep-research/08-seven-modes-in-action/<p>Deep Research 提供 7 种研究模式，覆盖从&quot;我不知道该研究什么&quot;到&quot;帮我验证这篇论文的每一条引用&quot;的全光谱。选择正确的模式可以大幅降低 token 消耗，同时获得更精准的研究产出。</p> <h2 class="heading-element" id="模式一览"><span>模式一览</span> <a href="#%e6%a8%a1%e5%bc%8f%e4%b8%80%e8%a7%88" class="heading-mark"> <svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg> </a> </h2><table> <thead> <tr> <th>模式</th> <th>光谱</th> <th>输出长度</th> <th>监管强度</th> <th>一句话</th> </tr> </thead> <tbody> <tr> <td><code>socratic</code></td> <td>Originality</td> <td>研究计划摘要</td> <td>Very High</td> <td>&ldquo;我还不太确定研究方向&rdquo;</td> </tr> <tr> <td><code>full</code></td> <td>Balanced</td> <td>3,000–8,000 词</td> <td>High</td> <td>&ldquo;帮我做完整研究&rdquo;</td> </tr> <tr> <td><code>systematic-review</code></td> <td>Fidelity</td> <td>5,000–15,000 词 PRISMA</td> <td>Medium</td> <td>&ldquo;要做系统评价/Meta 分析&rdquo;</td> </tr> <tr> <td><code>lit-review</code></td> <td>Fidelity</td> <td>注释书目 + 综合</td> <td>Medium</td> <td>&ldquo;需要文献回顾&rdquo;</td> </tr> <tr> <td><code>fact-check</code></td> <td>Fidelity</td> <td>逐条验证报告</td> <td>Medium</td> <td>&ldquo;验证这些说法&rdquo;</td> </tr> <tr> <td><code>review</code></td> <td>Balanced</td> <td>评审报告</td> <td>High</td> <td>&ldquo;评估这篇论文&rdquo;</td> </tr> <tr> <td><code>quick</code></td> <td>Fidelity</td> <td>500–1,500 词简报</td> <td>Medium</td> <td>&ldquo;30 分钟快速了解&rdquo;</td> </tr> </tbody> </table> <p>光谱分类（Fidelity–Originality）来自 Lu 2026 Fig 1c 的分析框架：</p>