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/05-academic-paper/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/05-academic-paper/09-12-agents-writing-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/05-academic-paper/09-12-agents-writing-team/<p>如果 Deep Research 是 ARS 的&quot;调研部&quot;，Academic Paper 就是&quot;出版部&quot;。它消费上游的研究产出——RQ Brief、Methodology Blueprint、注释书目、综合报告——然后通过 12 个角色分工明确的 Agent，将原始材料转化为符合学术出版标准的完整论文。</p> <p>本文逐一剖析这 12 个 Agent 的职责，以及它们如何通过 v3.6.8 的 Generator-Evaluator 合约防止自我欺骗。</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>12 个 Agent 按写作流程分为五组：</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/05-academic-paper/10-ten-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/05-academic-paper/10-ten-modes-in-action/<p>Academic Paper 是 ARS 中模式最多的 Skill——10 种模式覆盖了从&quot;帮我理清论文思路&quot;到&quot;生成 NeurIPS 的 AI 使用声明&quot;的完整写作生命周期。理解每种模式的定位和适用场景，是用好 ARS 的关键。</p> <h2 class="heading-element" id="模式全景"><span>模式全景</span> <a href="#%e6%a8%a1%e5%bc%8f%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><table> <thead> <tr> <th>#</th> <th>模式</th> <th>光谱</th> <th>监管</th> <th>核心场景</th> </tr> </thead> <tbody> <tr> <td>1</td> <td><code>plan</code></td> <td>Originality</td> <td>Very High</td> <td>&ldquo;我不知道怎么写&rdquo;</td> </tr> <tr> <td>2</td> <td><code>outline-only</code></td> <td>Balanced</td> <td>High</td> <td>&ldquo;先给我一个大纲&rdquo;</td> </tr> <tr> <td>3</td> <td><code>full</code></td> <td>Balanced</td> <td>High</td> <td>&ldquo;帮我写完整论文&rdquo;</td> </tr> <tr> <td>4</td> <td><code>abstract-only</code></td> <td>Fidelity</td> <td>Medium</td> <td>&ldquo;只需要摘要&rdquo;</td> </tr> <tr> <td>5</td> <td><code>lit-review</code></td> <td>Fidelity</td> <td>Medium</td> <td>&ldquo;把文献整理成综述论文&rdquo;</td> </tr> <tr> <td>6</td> <td><code>revision</code></td> <td>Fidelity</td> <td>High</td> <td>&ldquo;根据审稿意见修改&rdquo;</td> </tr> <tr> <td>7</td> <td><code>revision-coach</code></td> <td>Balanced</td> <td>Medium</td> <td>&ldquo;帮我解读审稿意见&rdquo;</td> </tr> <tr> <td>8</td> <td><code>format-convert</code></td> <td>Fidelity</td> <td>Low</td> <td>&ldquo;转成 LaTeX / 改引用格式&rdquo;</td> </tr> <tr> <td>9</td> <td><code>citation-check</code></td> <td>Fidelity</td> <td>Low</td> <td>&ldquo;检查引用格式和准确性&rdquo;</td> </tr> <tr> <td>10</td> <td><code>disclosure</code></td> <td>Fidelity</td> <td>Low</td> <td>&ldquo;生成 AI 使用声明&rdquo;</td> </tr> </tbody> </table> <p>监管强度从 Very High（用户主导对话 + 每步强制确认）到 Low（机械化/模板驱动，人类输入最少）。</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/05-academic-paper/11-style-calibration-quality/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/05-academic-paper/11-style-calibration-quality/<p>Academic Paper 的 12 个 Agent 可以生成结构完整、论证清晰的学术论文。但有两个问题对研究者至关重要：(1) 这篇论文听起来像<strong>我</strong>写的吗？(2) 这篇论文读起来像<strong>机器</strong>写的吗？</p> <p>ARS v2.9 引入的<strong>风格校准（Style Calibration）<strong>和</strong>写作品质检查（Writing Quality Check）<strong>分别回答了这两个问题。前者让 AI 学习你的声音，后者检测 AI 写作的常见模式。v3.3 引入的</strong>反泄漏协议（Anti-Leakage Protocol）<strong>和</strong>VLM 图表验证</strong>则保证了这些品质机制不会被 AI 自身破坏。</p> <h2 class="heading-element" id="风格校准让-ai-学你的声音"><span>风格校准：让 AI 学你的声音</span> <a href="#%e9%a3%8e%e6%a0%bc%e6%a0%a1%e5%87%86%e8%ae%a9-ai-%e5%ad%a6%e4%bd%a0%e7%9a%84%e5%a3%b0%e9%9f%b3" 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><h3 class="heading-element" id="原理"><span>原理</span> <a href="#%e5%8e%9f%e7%90%86" 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> </h3><p>风格校准是 Academic Paper intake 流程的 Step 10（可选步骤）。你提供 3 篇以上的既往论文，pipeline 会分析你的写作风格特征：</p>