Feature/a ibasic

docs/basics/hallucination.md

@@ -5,11 +5,11 @@
 
 # 为什么模型会胡说
 
-<div markdown="1" style="position:relative;overflow:hidden;border:1px solid rgba(255,112,67,0.28);border-radius:1rem;padding:1.25rem 1.35rem;margin:0.8rem 0 1.5rem;background:linear-gradient(135deg,rgba(255,112,67,0.14),rgba(63,81,181,0.08) 46%,rgba(0,188,212,0.10));box-shadow:0 0.65rem 1.8rem rgba(0,0,0,0.10);">
-<div style="position:absolute;right:-3.8rem;top:-4.2rem;width:12rem;height:12rem;border-radius:50%;background:radial-gradient(circle,rgba(255,112,67,0.26),rgba(255,112,67,0));"></div>
-<div style="position:absolute;left:-4rem;bottom:-5rem;width:14rem;height:14rem;border-radius:50%;background:radial-gradient(circle,rgba(33,150,243,0.18),rgba(33,150,243,0));"></div>
+<div markdown="1" style="position:relative;overflow:hidden;border:1px solid rgba(244,67,54,0.30);border-radius:1rem;padding:1.25rem 1.35rem;margin:0.8rem 0 1.5rem;background:linear-gradient(135deg,rgba(244,67,54,0.13),rgba(255,112,67,0.10) 46%,rgba(156,39,176,0.10));box-shadow:0 0.65rem 1.8rem rgba(0,0,0,0.10);">
+<div style="position:absolute;right:-3.8rem;top:-4.2rem;width:12rem;height:12rem;border-radius:50%;background:radial-gradient(circle,rgba(244,67,54,0.23),rgba(244,67,54,0));"></div>
+<div style="position:absolute;left:-4rem;bottom:-5rem;width:14rem;height:14rem;border-radius:50%;background:radial-gradient(circle,rgba(156,39,176,0.16),rgba(156,39,176,0));"></div>
 <div style="position:relative;z-index:1;">
-<span style="display:inline-block;padding:0.18rem 0.55rem;border-radius:999px;background:rgba(255,112,67,0.16);color:#e65100;font-size:0.78rem;font-weight:700;letter-spacing:0.02em;">AI 基础 · 第 10 站</span>
+<span style="display:inline-block;padding:0.18rem 0.55rem;border-radius:999px;background:rgba(244,67,54,0.15);color:#b71c1c;font-size:0.78rem;font-weight:700;letter-spacing:0.02em;">AI 基础 · 第 10 站</span>
 
 <strong>AI 幻觉</strong>指模型生成看起来合理、语气很自信，实际却错误、虚构、无法验证或与材料冲突的内容。它最麻烦的地方在于：错得很像真的。
 
@@ -277,31 +277,31 @@
 ??? warning "误区 1：AI 幻觉就是模型撒谎"
 
     撒谎需要意图。模型没有人的动机，它是在根据上下文和概率生成文本。
 
     更准确的理解是：模型在证据不足、目标不清或检索有噪音时，仍然把答案补全成了自然语言。
 
 ??? warning "误区 2：模型越大，幻觉越少"
 
     模型变大通常会提升很多能力，但真实性不一定线性提升。TruthfulQA 的早期实验就发现，更大的模型可能更会模仿训练数据中的流行错误说法。
 
     对事实任务来说，模型能力、训练目标、对齐方式、检索质量和核查流程都重要。
 
 ??? warning "误区 3：接入 RAG 后就不会幻觉"
 
     RAG 能显著改善事实任务，但检索结果可能错，知识库可能旧，切块可能丢上下文，模型也可能没有忠实使用证据。
 
     好的 RAG 系统需要评估检索、引用和生成三段链路。
 
 ??? warning "误区 4：temperature 调成 0 就能消灭幻觉"
 
     低温让输出更稳定，但稳定不等于正确。
 
     如果模型缺少事实依据，低温可能只是让它更稳定地重复同一个错误。
 
 ??? warning "误区 5：有引用就一定可信"
 
     引用也可能被编造，或者引用存在但无法支撑结论。
 
     真正要看的是：来源是否存在、来源是否权威、引用片段是否真的支持那句话。
 
 ## 最小实验：亲手观察一次幻觉
@@ -335,19 +335,19 @@
 ??? question "练习 1：识别幻觉类型"
 
     判断下面输出可能属于哪类幻觉：
 
     <div class="highlight"><pre><code>根据《2024 中国大模型安全白皮书》第 17 页，所有主流模型的幻觉率都低于 1%。</code></pre></div>
 
     ??? done "参考思路"
 
         先查这份白皮书是否存在，再查第 17 页有没有这句话。若来源不存在或页码不匹配，就是引用幻觉。若来源存在但结论不支持这句话，就是材料幻觉或事实幻觉。
 
 ??? question "练习 2：改写提示词"
 
     把下面提示词改得更不容易诱发幻觉：
 
     <div class="highlight"><pre><code>帮我总结一下这家公司过去三年的融资情况，要详细一点。</code></pre></div>
 
     ??? done "参考思路"
 
         可以改成：请只基于我提供的材料总结这家公司过去三年的融资情况。每一笔融资都要列出日期、金额、投资方和来源链接。材料里没有的信息请标注「未找到」，不要补全。

docs/basics/model-data-training.md

@@ -5,11 +5,11 @@ tags:
 
 # 模型、数据、训练与推理
 
-<div markdown="1" style="position:relative;overflow:hidden;border:1px solid rgba(255,112,67,0.28);border-radius:1rem;padding:1.25rem 1.35rem;margin:0.8rem 0 1.5rem;background:linear-gradient(135deg,rgba(255,112,67,0.14),rgba(63,81,181,0.08) 46%,rgba(0,188,212,0.10));box-shadow:0 0.65rem 1.8rem rgba(0,0,0,0.10);">
-<div style="position:absolute;right:-3.8rem;top:-4.2rem;width:12rem;height:12rem;border-radius:50%;background:radial-gradient(circle,rgba(255,112,67,0.26),rgba(255,112,67,0));"></div>
-<div style="position:absolute;left:-4rem;bottom:-5rem;width:14rem;height:14rem;border-radius:50%;background:radial-gradient(circle,rgba(33,150,243,0.18),rgba(33,150,243,0));"></div>
+<div markdown="1" style="position:relative;overflow:hidden;border:1px solid rgba(76,175,80,0.30);border-radius:1rem;padding:1.25rem 1.35rem;margin:0.8rem 0 1.5rem;background:linear-gradient(135deg,rgba(76,175,80,0.14),rgba(0,150,136,0.09) 46%,rgba(255,183,77,0.10));box-shadow:0 0.65rem 1.8rem rgba(0,0,0,0.10);">
+<div style="position:absolute;right:-3.8rem;top:-4.2rem;width:12rem;height:12rem;border-radius:50%;background:radial-gradient(circle,rgba(76,175,80,0.24),rgba(76,175,80,0));"></div>
+<div style="position:absolute;left:-4rem;bottom:-5rem;width:14rem;height:14rem;border-radius:50%;background:radial-gradient(circle,rgba(0,150,136,0.18),rgba(0,150,136,0));"></div>
 <div style="position:relative;z-index:1;">
-<span style="display:inline-block;padding:0.18rem 0.55rem;border-radius:999px;background:rgba(255,112,67,0.16);color:#e65100;font-size:0.78rem;font-weight:700;letter-spacing:0.02em;">AI 基础 · 第 6 站</span>
+<span style="display:inline-block;padding:0.18rem 0.55rem;border-radius:999px;background:rgba(76,175,80,0.16);color:#1b5e20;font-size:0.78rem;font-weight:700;letter-spacing:0.02em;">AI 基础 · 第 6 站</span>
 
 <strong>模型（Model）</strong>像一套会计算的规则，<strong>数据（Data）</strong>是它看过的材料，<strong>训练（Training）</strong>负责把规则调出来，<strong>推理（Inference）</strong>就是训练好的模型真正开始回答问题。
 

docs/basics/multimodal-ai.md

@@ -5,11 +5,11 @@ tags:
 
 # 多模态 AI：文字、图片、语音和视频
 
-<div markdown="1" style="position:relative;overflow:hidden;border:1px solid rgba(255,112,67,0.28);border-radius:1rem;padding:1.25rem 1.35rem;margin:0.8rem 0 1.5rem;background:linear-gradient(135deg,rgba(255,112,67,0.14),rgba(63,81,181,0.08) 46%,rgba(0,188,212,0.10));box-shadow:0 0.65rem 1.8rem rgba(0,0,0,0.10);">
-<div style="position:absolute;right:-3.8rem;top:-4.2rem;width:12rem;height:12rem;border-radius:50%;background:radial-gradient(circle,rgba(255,112,67,0.26),rgba(255,112,67,0));"></div>
-<div style="position:absolute;left:-4rem;bottom:-5rem;width:14rem;height:14rem;border-radius:50%;background:radial-gradient(circle,rgba(33,150,243,0.18),rgba(33,150,243,0));"></div>
+<div markdown="1" style="position:relative;overflow:hidden;border:1px solid rgba(0,188,212,0.30);border-radius:1rem;padding:1.25rem 1.35rem;margin:0.8rem 0 1.5rem;background:linear-gradient(135deg,rgba(0,188,212,0.14),rgba(63,81,181,0.08) 46%,rgba(233,30,99,0.10));box-shadow:0 0.65rem 1.8rem rgba(0,0,0,0.10);">
+<div style="position:absolute;right:-3.8rem;top:-4.2rem;width:12rem;height:12rem;border-radius:50%;background:radial-gradient(circle,rgba(0,188,212,0.26),rgba(0,188,212,0));"></div>
+<div style="position:absolute;left:-4rem;bottom:-5rem;width:14rem;height:14rem;border-radius:50%;background:radial-gradient(circle,rgba(233,30,99,0.16),rgba(233,30,99,0));"></div>
 <div style="position:relative;z-index:1;">
-<span style="display:inline-block;padding:0.18rem 0.55rem;border-radius:999px;background:rgba(255,112,67,0.16);color:#e65100;font-size:0.78rem;font-weight:700;letter-spacing:0.02em;">AI 基础 · 第 5 站</span>
+<span style="display:inline-block;padding:0.18rem 0.55rem;border-radius:999px;background:rgba(0,188,212,0.16);color:#006064;font-size:0.78rem;font-weight:700;letter-spacing:0.02em;">AI 基础 · 第 5 站</span>
 
 <strong>多模态 AI（Multimodal AI）</strong>就是让 AI 同时处理文字、图片、语音和视频，把不同形式的信息放到同一张理解地图里。
 

docs/basics/prompt-context-memory.md

@@ -5,11 +5,11 @@ tags:
 
 # Prompt、上下文和记忆
 
-<div markdown="1" style="position:relative;overflow:hidden;border:1px solid rgba(255,112,67,0.28);border-radius:1rem;padding:1.25rem 1.35rem;margin:0.8rem 0 1.5rem;background:linear-gradient(135deg,rgba(255,112,67,0.14),rgba(63,81,181,0.08) 46%,rgba(0,188,212,0.10));box-shadow:0 0.65rem 1.8rem rgba(0,0,0,0.10);">
-<div style="position:absolute;right:-3.8rem;top:-4.2rem;width:12rem;height:12rem;border-radius:50%;background:radial-gradient(circle,rgba(255,112,67,0.26),rgba(255,112,67,0));"></div>
-<div style="position:absolute;left:-4rem;bottom:-5rem;width:14rem;height:14rem;border-radius:50%;background:radial-gradient(circle,rgba(33,150,243,0.18),rgba(33,150,243,0));"></div>
+<div markdown="1" style="position:relative;overflow:hidden;border:1px solid rgba(63,81,181,0.30);border-radius:1rem;padding:1.25rem 1.35rem;margin:0.8rem 0 1.5rem;background:linear-gradient(135deg,rgba(63,81,181,0.14),rgba(126,87,194,0.09) 46%,rgba(38,166,154,0.10));box-shadow:0 0.65rem 1.8rem rgba(0,0,0,0.10);">
+<div style="position:absolute;right:-3.8rem;top:-4.2rem;width:12rem;height:12rem;border-radius:50%;background:radial-gradient(circle,rgba(63,81,181,0.24),rgba(63,81,181,0));"></div>
+<div style="position:absolute;left:-4rem;bottom:-5rem;width:14rem;height:14rem;border-radius:50%;background:radial-gradient(circle,rgba(126,87,194,0.16),rgba(126,87,194,0));"></div>
 <div style="position:relative;z-index:1;">
-<span style="display:inline-block;padding:0.18rem 0.55rem;border-radius:999px;background:rgba(255,112,67,0.16);color:#e65100;font-size:0.78rem;font-weight:700;letter-spacing:0.02em;">AI 基础 · 第 8 站</span>
+<span style="display:inline-block;padding:0.18rem 0.55rem;border-radius:999px;background:rgba(63,81,181,0.16);color:#283593;font-size:0.78rem;font-weight:700;letter-spacing:0.02em;">AI 基础 · 第 8 站</span>
 <strong>Prompt</strong>负责说明这次任务，<strong>上下文</strong>决定模型当前能看到什么，<strong>记忆</strong>负责把长期信息重新带回现场。
 
 <div markdown="1" style="display:grid;grid-template-columns:repeat(auto-fit,minmax(9rem,1fr));gap:0.75rem;margin-top:1rem;">

docs/basics/temperature-sampling.md

@@ -5,11 +5,11 @@ tags:
 
 # 温度与采样参数
 
-<div markdown="1" style="position:relative;overflow:hidden;border:1px solid rgba(255,112,67,0.28);border-radius:1rem;padding:1.25rem 1.35rem;margin:0.8rem 0 1.5rem;background:linear-gradient(135deg,rgba(255,112,67,0.14),rgba(63,81,181,0.08) 46%,rgba(0,188,212,0.10));box-shadow:0 0.65rem 1.8rem rgba(0,0,0,0.10);">
-<div style="position:absolute;right:-3.8rem;top:-4.2rem;width:12rem;height:12rem;border-radius:50%;background:radial-gradient(circle,rgba(255,112,67,0.26),rgba(255,112,67,0));"></div>
-<div style="position:absolute;left:-4rem;bottom:-5rem;width:14rem;height:14rem;border-radius:50%;background:radial-gradient(circle,rgba(33,150,243,0.18),rgba(33,150,243,0));"></div>
+<div markdown="1" style="position:relative;overflow:hidden;border:1px solid rgba(255,152,0,0.32);border-radius:1rem;padding:1.25rem 1.35rem;margin:0.8rem 0 1.5rem;background:linear-gradient(135deg,rgba(255,152,0,0.15),rgba(255,112,67,0.10) 46%,rgba(233,30,99,0.09));box-shadow:0 0.65rem 1.8rem rgba(0,0,0,0.10);">
+<div style="position:absolute;right:-3.8rem;top:-4.2rem;width:12rem;height:12rem;border-radius:50%;background:radial-gradient(circle,rgba(255,152,0,0.26),rgba(255,152,0,0));"></div>
+<div style="position:absolute;left:-4rem;bottom:-5rem;width:14rem;height:14rem;border-radius:50%;background:radial-gradient(circle,rgba(233,30,99,0.16),rgba(233,30,99,0));"></div>
 <div style="position:relative;z-index:1;">
-<span style="display:inline-block;padding:0.18rem 0.55rem;border-radius:999px;background:rgba(255,112,67,0.16);color:#e65100;font-size:0.78rem;font-weight:700;letter-spacing:0.02em;">AI 基础 · 第 9 站</span>
+<span style="display:inline-block;padding:0.18rem 0.55rem;border-radius:999px;background:rgba(255,152,0,0.18);color:#e65100;font-size:0.78rem;font-weight:700;letter-spacing:0.02em;">AI 基础 · 第 9 站</span>
 
 <strong>温度、Top-k、Top-p</strong>决定模型从候选 token 里怎么选下一个 token。它们不负责增加知识，负责控制输出时的随机性、稳定性和多样性。
 

docs/basics/token-embedding-context.md

@@ -5,11 +5,11 @@ tags:
 
 # Token、Embedding 与上下文窗口
 
-<div markdown="1" style="position:relative;overflow:hidden;border:1px solid rgba(255,112,67,0.28);border-radius:1rem;padding:1.25rem 1.35rem;margin:0.8rem 0 1.5rem;background:linear-gradient(135deg,rgba(255,112,67,0.14),rgba(63,81,181,0.08) 46%,rgba(0,188,212,0.10));box-shadow:0 0.65rem 1.8rem rgba(0,0,0,0.10);">
-<div style="position:absolute;right:-3.8rem;top:-4.2rem;width:12rem;height:12rem;border-radius:50%;background:radial-gradient(circle,rgba(255,112,67,0.26),rgba(255,112,67,0));"></div>
-<div style="position:absolute;left:-4rem;bottom:-5rem;width:14rem;height:14rem;border-radius:50%;background:radial-gradient(circle,rgba(33,150,243,0.18),rgba(33,150,243,0));"></div>
+<div markdown="1" style="position:relative;overflow:hidden;border:1px solid rgba(3,169,244,0.30);border-radius:1rem;padding:1.25rem 1.35rem;margin:0.8rem 0 1.5rem;background:linear-gradient(135deg,rgba(3,169,244,0.14),rgba(33,150,243,0.09) 46%,rgba(156,39,176,0.10));box-shadow:0 0.65rem 1.8rem rgba(0,0,0,0.10);">
+<div style="position:absolute;right:-3.8rem;top:-4.2rem;width:12rem;height:12rem;border-radius:50%;background:radial-gradient(circle,rgba(3,169,244,0.24),rgba(3,169,244,0));"></div>
+<div style="position:absolute;left:-4rem;bottom:-5rem;width:14rem;height:14rem;border-radius:50%;background:radial-gradient(circle,rgba(156,39,176,0.16),rgba(156,39,176,0));"></div>
 <div style="position:relative;z-index:1;">
-<span style="display:inline-block;padding:0.18rem 0.55rem;border-radius:999px;background:rgba(255,112,67,0.16);color:#e65100;font-size:0.78rem;font-weight:700;letter-spacing:0.02em;">AI 基础 · 第 7 站</span>
+<span style="display:inline-block;padding:0.18rem 0.55rem;border-radius:999px;background:rgba(3,169,244,0.16);color:#01579b;font-size:0.78rem;font-weight:700;letter-spacing:0.02em;">AI 基础 · 第 7 站</span>
 
 <strong>Token</strong>决定模型把文字切成什么颗粒，<strong>Embedding</strong>负责把这些颗粒变成向量，<strong>上下文窗口</strong>规定模型一次能看多少信息。
 

docs/basics/what-is-ai.md

@@ -1,9 +1,30 @@
+---
+tags:
+  - AI 基础
+---
+
 # 什么是 AI
 
-AI 的全称是 Artificial Intelligence，中文叫「人工智能」。
+<div markdown="1" style="position:relative;overflow:hidden;border:1px solid rgba(63,81,181,0.28);border-radius:1rem;padding:1.25rem 1.35rem;margin:0.8rem 0 1.5rem;background:linear-gradient(135deg,rgba(63,81,181,0.14),rgba(0,188,212,0.09) 48%,rgba(255,112,67,0.10));box-shadow:0 0.65rem 1.8rem rgba(0,0,0,0.10);">
+<div style="position:absolute;right:-3.8rem;top:-4.2rem;width:12rem;height:12rem;border-radius:50%;background:radial-gradient(circle,rgba(63,81,181,0.24),rgba(63,81,181,0));"></div>
+<div style="position:absolute;left:-4rem;bottom:-5rem;width:14rem;height:14rem;border-radius:50%;background:radial-gradient(circle,rgba(0,188,212,0.18),rgba(0,188,212,0));"></div>
+<div style="position:relative;z-index:1;">
+<span style="display:inline-block;padding:0.18rem 0.55rem;border-radius:999px;background:rgba(63,81,181,0.16);color:#283593;font-size:0.78rem;font-weight:700;letter-spacing:0.02em;">AI 基础 · 第 1 站</span>
+
+<strong>AI（Artificial Intelligence，人工智能）</strong>可以先理解成一组能力，让计算机在某些任务上表现出「类似智能」的行为。
 
-<div style="border-left:4px solid var(--md-accent-fg-color);padding:0.8rem 1rem;margin:1rem 0 1.4rem;background:var(--md-code-bg-color);border-radius:0.35rem;">
-  <strong>先记住一句话：</strong>AI 可以先理解成一组能力，让计算机在某些任务上表现出“类似智能”的行为。
+<div markdown="1" style="display:grid;grid-template-columns:repeat(auto-fit,minmax(9rem,1fr));gap:0.75rem;margin-top:1rem;">
+<div style="padding:0.8rem;border-radius:0.75rem;background:var(--md-default-bg-color);border:1px solid var(--md-default-fg-color--lightest);">
+<strong>目标</strong><br><span style="color:var(--md-default-fg-color--light);font-size:0.9rem;">让机器完成智能任务</span>
+</div>
+<div style="padding:0.8rem;border-radius:0.75rem;background:var(--md-default-bg-color);border:1px solid var(--md-default-fg-color--lightest);">
+<strong>能力</strong><br><span style="color:var(--md-default-fg-color--light);font-size:0.9rem;">识别、预测、生成、规划</span>
+</div>
+<div style="padding:0.8rem;border-radius:0.75rem;background:var(--md-default-bg-color);border:1px solid var(--md-default-fg-color--lightest);">
+<strong>定位</strong><br><span style="color:var(--md-default-fg-color--light);font-size:0.9rem;">机器学习和 LLM 都在其中</span>
+</div>
+</div>
+</div>
 </div>
 
 <figure markdown="1" style="margin:1.2rem 0;text-align:center;">

docs/basics/what-is-llm.md

@@ -5,18 +5,26 @@ tags:
 
 # 什么是 LLM
 
-<div markdown="1" style="position:relative;border:1px solid rgba(126,87,194,0.26);border-radius:1rem;padding:1.1rem 1.2rem;margin:0.8rem 0 1.5rem;background:linear-gradient(180deg,rgba(28,27,43,0.06),rgba(126,87,194,0.08));overflow:hidden;">
-<div style="position:absolute;right:1rem;top:1rem;width:0.55rem;height:0.55rem;border-radius:50%;background:#7e57c2;box-shadow:-1rem 0 #26c6da,-2rem 0 #ffb74d;"></div>
-<div markdown="1" style="font-family:var(--md-code-font-family);font-size:0.92rem;line-height:1.8;padding:0.75rem 0.85rem;border-radius:0.75rem;background:var(--md-code-bg-color);border:1px solid var(--md-default-fg-color--lightest);margin-bottom:0.9rem;">
-<span style="color:#7e57c2;font-weight:700;">user</span>  帮我把这段话改得像人话一点  
-<span style="color:#26a69a;font-weight:700;">model</span>  可以，我先保留原意，再把句子拆短一点……
-</div>
-
-LLM，全称 **Large Language Model**，中文叫「大语言模型」。
+<div markdown="1" style="position:relative;overflow:hidden;border:1px solid rgba(126,87,194,0.28);border-radius:1rem;padding:1.25rem 1.35rem;margin:0.8rem 0 1.5rem;background:linear-gradient(135deg,rgba(126,87,194,0.14),rgba(38,166,154,0.09) 48%,rgba(255,183,77,0.10));box-shadow:0 0.65rem 1.8rem rgba(0,0,0,0.10);">
+<div style="position:absolute;right:-3.8rem;top:-4.2rem;width:12rem;height:12rem;border-radius:50%;background:radial-gradient(circle,rgba(126,87,194,0.24),rgba(126,87,194,0));"></div>
+<div style="position:absolute;left:-4rem;bottom:-5rem;width:14rem;height:14rem;border-radius:50%;background:radial-gradient(circle,rgba(38,166,154,0.18),rgba(38,166,154,0));"></div>
+<div style="position:relative;z-index:1;">
+<span style="display:inline-block;padding:0.18rem 0.55rem;border-radius:999px;background:rgba(126,87,194,0.16);color:#5e35b1;font-size:0.78rem;font-weight:700;letter-spacing:0.02em;">AI 基础 · 第 4 站</span>
 
-先别急着把它想成一个懂你的大脑。更贴近它工作方式的说法是，它像一台巨大的语言续写机器。你给它一段文字，它根据上下文判断后面最可能接什么，然后一个 token 一个 token 地写下去。
+<strong>LLM（Large Language Model，大语言模型）</strong>可以先理解成一台巨大的语言续写机器。它根据上下文预测后面最可能接什么，再一个 token 一个 token 地写下去。
 
-神奇的地方在于，当这台续写机器读过足够多文本、代码、网页、论文和对话之后，续写这件小事，会长出很多看起来很像理解、总结、翻译、写代码和推理的能力。
+<div markdown="1" style="display:grid;grid-template-columns:repeat(auto-fit,minmax(9rem,1fr));gap:0.75rem;margin-top:1rem;">
+<div style="padding:0.8rem;border-radius:0.75rem;background:var(--md-default-bg-color);border:1px solid var(--md-default-fg-color--lightest);">
+<strong>输入</strong><br><span style="color:var(--md-default-fg-color--light);font-size:0.9rem;">提示词和上下文</span>
+</div>
+<div style="padding:0.8rem;border-radius:0.75rem;background:var(--md-default-bg-color);border:1px solid var(--md-default-fg-color--lightest);">
+<strong>核心动作</strong><br><span style="color:var(--md-default-fg-color--light);font-size:0.9rem;">预测下一个 token</span>
+</div>
+<div style="padding:0.8rem;border-radius:0.75rem;background:var(--md-default-bg-color);border:1px solid var(--md-default-fg-color--lightest);">
+<strong>输出</strong><br><span style="color:var(--md-default-fg-color--light);font-size:0.9rem;">回答、总结、改写、代码</span>
+</div>
+</div>
+</div>
 </div>
 
 ## 先盯住那个「发送」按钮

docs/preface/roadmap.md

@@ -3,14 +3,14 @@
 Hello-AI 建议按下面顺序学习：
 
 ```mermaid
-flowchart TD
+flowchart LR
     A[AI 基础] --> B[Prompt 基础]
     B --> C[工具使用]
     C --> D[RAG]
     D --> E[Agent]
     E --> F[Build]
     F --> G[Eval]
-    G --> H[Safety]
+    G --> H[Security]
     H --> I[Lab]
 ```