@@ -375,7 +375,7 @@ <h1>AI4AI Survey: From Long-Horizon Agents to Recursive Self-Improvement</h1>
 
         <p>
           <span class="lang-en" lang="en">The survey covers the component capabilities that AI4AI requires and how they are measured today, gives a taxonomy for locating where the loop closes, analyzes reliable execution within the coupled model, harness, environment, and evaluation stack, unfolds in depth along the model and harness routes, and ends with the failures that bound recursive improvement.</span>
-          <span class="lang-zh" lang="zh-CN">这篇综述覆盖 AI4AI 需要的各种组件能力，以及它们目前的度量方式；给出一套定位自主范围的 taxonomy，把可靠执行放在 model、harness、environment、evaluation 四层耦合的框架下分析，并沿 model 与 harness 两条路线深度展开；最后总结限制递归改进的各类失败。</span>
+          <span class="lang-zh" lang="zh-CN">这篇综述覆盖 AI4AI 需要的各种组件能力，以及它们目前的度量方式；给出一套判断系统自主到哪一步的分类框架，把可靠执行放在模型、harness、环境、评测四层耦合的结构里分析，并沿 model 与 harness 两条路线深入展开；最后总结限制递归改进的各类失败。</span>
         </p>
 
         <p>
@@ -428,7 +428,7 @@ <h2 id="loop">
         </p>
         <p>
           <span class="lang-en" lang="en">Long-horizon is often equated with long context, many turns, or many tool calls. Those are only proxies. The survey builds its definition and metrics around four questions: what is improved, why long-horizon matters, how to measure it, and when the loop starts to recurse (RSI).</span>
-          <span class="lang-zh" lang="zh-CN">long-horizon 常被等同于长 context、多轮交互或大量 tool call，这些只是表象，长并不等于长程。综述围绕四个问题给出定义和度量：改进的是什么、为什么 long-horizon 重要、如何度量、以及何时开始递归修改（RSI）。</span>
+          <span class="lang-zh" lang="zh-CN">long-horizon 常被等同于长上下文、多轮交互或大量工具调用，这些只是表象，长并不等于长程。综述围绕四个问题给出定义和度量：改进的是什么、为什么 long-horizon 重要、如何度量、以及何时开始递归修改（RSI）。</span>
         </p>
         <figure class="framed plate">
           <img
@@ -451,7 +451,7 @@ <h2 id="closure">
         </p>
         <p>
           <span class="lang-en" lang="en">Of the taxonomy's five dimensions, closure does most of the discriminating work. It asks which stages of <code>(goal, plan, execute, feedback, repair)</code> the system owns.</span>
-          <span class="lang-zh" lang="zh-CN">taxonomy 的五个维度里，承担主要区分工作的是 closure：<code>(goal, plan, execute, feedback, repair)</code> 五个环节里，哪些由系统自主完成。</span>
+          <span class="lang-zh" lang="zh-CN">分类框架有五个维度，真正起区分作用的是 closure，也就是 <code>(goal, plan, execute, feedback, repair)</code> 五个环节里，哪些由系统自己完成。</span>
         </p>
 
         <div class="table-scroll">
@@ -489,7 +489,7 @@ <h2 id="closure">
                 </td>
                 <td>
                   <span class="lang-en" lang="en">humans fix the goal and the search space; the system closes execute only</span>
-                  <span class="lang-zh" lang="zh-CN">只有 execute 是系统自主的，目标和搜索空间由人定</span>
+                  <span class="lang-zh" lang="zh-CN">只有执行由系统自己完成，目标和搜索空间由人定</span>
                 </td>
               </tr>
               <tr>
@@ -504,7 +504,7 @@ <h2 id="closure">
                 </td>
                 <td>
                   <span class="lang-en" lang="en">the system closes plan and execute; feedback is mix under a human-written evaluator</span>
-                  <span class="lang-zh" lang="zh-CN">plan 与 execute 是系统自主的，feedback 在人写的 evaluator 下记为 mix</span>
+                  <span class="lang-zh" lang="zh-CN">规划和执行由系统自己完成，反馈依赖人写的评测器（记为 mix）</span>
                 </td>
               </tr>
               <tr>
@@ -519,7 +519,7 @@ <h2 id="closure">
                 </td>
                 <td>
                   <span class="lang-en" lang="en">all four execution stages close; the goal is still given by humans</span>
-                  <span class="lang-zh" lang="zh-CN">四个执行阶段都是系统自主的，goal 仍由人定</span>
+                  <span class="lang-zh" lang="zh-CN">四个执行阶段都由系统自己完成，目标仍由人定</span>
                 </td>
               </tr>
             </tbody>
@@ -541,7 +541,7 @@ <h2 id="closure">
           </li>
           <li>
             <span class="lang-en" lang="en"><strong>Agent edits its own code</strong> (Darwin Gödel Machine, Self-Harness): still judged by a fixed evaluator.</span>
-            <span class="lang-zh" lang="zh-CN"><strong>agent 改自己的代码</strong>（Darwin Gödel Machine、Self-Harness）：评判它的 evaluator 还是固定的。</span>
+            <span class="lang-zh" lang="zh-CN"><strong>agent 改自己的代码</strong>（Darwin Gödel Machine、Self-Harness）：评判它的评测器还是固定的。</span>
           </li>
           <li>
             <span class="lang-en" lang="en">Only when a pass feeds into the process that proposes the next improvement goal does the loop become <strong>truly recursive</strong>.</span>
@@ -564,7 +564,7 @@ <h2 id="horizon">
           />
           <figcaption>
             <span class="lang-en" lang="en">Model-side interventions by stage: plan yields supervision, execute grounds in tools and state, feedback becomes step-level credit, repair rewrites the training recipe.</span>
-            <span class="lang-zh" lang="zh-CN">model 侧干预：plan 产生监督信号，execute 依赖 tool 与 state，feedback 转为 step-level credit，repair 改写训练配方。</span>
+            <span class="lang-zh" lang="zh-CN">模型侧的干预：规划阶段产生监督信号，执行阶段落到工具与状态上，反馈阶段变成按步骤分配的信用，修复阶段改写训练配方。</span>
           </figcaption>
         </figure>
 
@@ -606,7 +606,7 @@ <h2 id="composition">
           </li>
           <li>
             <span class="lang-en" lang="en"><strong>Evidence and control.</strong> Gains can be bought with budget, so report a success rate together with its cost. A credible claim also needs model–harness attribution, contamination controls, and disclosed human oversight.</span>
-            <span class="lang-zh" lang="zh-CN"><strong>证据与控制。</strong>增益可以用预算买来，所以成功率要和成本一起报。可信的改进声明还需要 model–harness 归因、数据污染控制和人工监督的披露。</span>
+            <span class="lang-zh" lang="zh-CN"><strong>证据与控制。</strong>增益可以用预算买来，所以成功率要和成本一起报。可信的改进声明还需要说清模型和 harness 各自的贡献、控制数据污染，并披露人工监督的方式。</span>
           </li>
         </ul>
 
@@ -620,7 +620,7 @@ <h2 id="conclusion">
         </p>
         <p>
           <span class="lang-en" lang="en">Plan, execute, feedback, and repair are each already strong, and across the 35 audited systems execute is fully system-owned. <strong>No system picks its own goal, and none can judge what counts as “better” without a human.</strong></span>
-          <span class="lang-zh" lang="zh-CN">plan、execute、feedback、repair 每一项已经很强，盘点的 35 个系统里 execute 更是全部自主。<strong>但没有一个系统自己选目标，没有一个能不靠人判定“什么算更好”。</strong></span>
+          <span class="lang-zh" lang="zh-CN">规划、执行、反馈、修复每一项都已经很强，盘点的 35 个系统里，执行更是全部由系统自己完成。<strong>但没有一个系统自己选目标，没有一个能不靠人判定“什么算更好”。</strong></span>
         </p>
         <p>
           <span class="lang-en" lang="en">The eve ends the first time a system, with no human-specified goal or evaluation, has successors that reliably strengthen it. We hope this taxonomy and this audit help that moment arrive a little sooner.</span>