Skip to content

Commit 5130947

Browse files
committed
feat: Expand content with new articles, introduce a PDF slide shortcode, and update build process and shortcode documentation.
1 parent 78b642e commit 5130947

4 files changed

Lines changed: 151 additions & 8 deletions

File tree

content/zh/google/google_notebooklm.md

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -239,7 +239,7 @@ You are a highly capable research assistant and tutor. Create a detailed study g
239239
在点击生成幻灯片(Slides)时,在自定义提示词(Custom Prompt)区域输入以下指令。为了获得最佳效果,建议使用英文 Prompt:
240240

241241
```markdown
242-
Make the character from the uploaded image '[角色文件名.png]' appear in each slide.
242+
Make the character from the uploaded image '角色文件名.png' appear in each slide.
243243
Have them serve as the navigator (guide) for the presentation.
244244
Be sure to accurately reflect the character's characteristics and adjust their pose/expression to match the context of each slide.
245245
```

content/zh/google/nested_learning_hope_arch.md

Lines changed: 3 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -5,7 +5,7 @@ draft: false
55
tags: ["Nested Learning", "Hope架构", "持续学习"]
66
categories: ["big_companies"]
77
description: "Google Nested Learning原理剖析与Hope架构:迈向持续学习的新范式"
8-
wordCount: 4979
8+
wordCount: 4933
99
readingTime: 13
1010
ai_cover: "/images/generated-covers/e4d9bfb7b2241da6ed6444aa1df2bdde.webp"
1111
cover:
@@ -109,7 +109,7 @@ y_t = MLP^(fk)(MLP^(fk-1)(...MLP^(f1)(x_t)...)
109109

110110
Hope架构的另一个核心创新是**自修改机制**,使模型能够在推理过程中根据接受到的数据动态调整自身的参数甚至更新规则。这一机制基于Titans架构扩展,使Hope成为一个自引用模型,能够执行无限层级的上下文学习。
111111

112-
自修改机制的关键在于将**优化器本身视为可学习的联想记忆模块**。传统优化器(如Adam、带动量的SGD)通常是外部设计的、固定的更新规则,而Hope将优化过程内部化,使模型能够“学习如何学习”(learn how to learn)。
112+
自修改机制的关键在于将**优化器本身视为可学习的联想记忆模块**。传统优化器(如Adam、带动量的SGD)通常是外部设计的、固定的更新规则,而Hope将优化过程内部化,使模型能够“学习如何学习”(**learn how to learn**)。
113113

114114
具体而言,Hope中的**深度优化器**将优化器状态(如动量、二阶矩)显式建模为可微网络,并通过元学习或多任务训练对这些“优化器网络”进行端到端训练。这种设计使优化器能够适应数据分布的变化,对噪声和异常值更具鲁棒性。
115115

@@ -175,7 +175,7 @@ Nested Learning范式及Hope架构的创新价值主要体现在以下几个方
175175

176176
**超参数调优挑战**:CMS系统中各MLP模块的更新频率、容量分配等超参数需要精心调试,系统复杂度较高,这可能增加模型部署的难度。
177177

178-
**大规模扩展性**:当前实验主要针对中等规模模型(最多1.3B参数),对于超大规模模型(数十亿甚至数百亿参数)的扩展性和有效性仍有待验证。
178+
**大规模扩展性**:当前实验主要针对中等规模模型(最多1.3B参数),**对于超大规模模型(数十亿甚至数百亿参数)的扩展性和有效性仍有待验证**
179179

180180
**理论基础的完善**:Nested Learning作为一个新兴范式,其数学基础和理论体系仍需进一步完善,特别是关于收敛性、泛化能力等方面的理论分析。
181181

@@ -199,4 +199,3 @@ Nested Learning范式代表了深度学习领域的一次重要范式转换,
199199
8. [Nested Learning: The Illusion of Deep Learning Architectures](https://abehrouz.github.io/files/NL.pdf)
200200
9. [Introducing Nested Learning: A new ML paradigm for continual learning](https://research.google/blog/introducing-nested-learning-a-new-ml-paradigm-for-continual-learning/)
201201

202-
*以上参考文献请根据实际发表的论文信息进行补充和调整,建议直接引用原始论文《Nested Learning: The Illusion of Deep Learning Architectures》以及NeurIPS 2025会议论文集。*

content/zh/projects/info_collect/TrendRadar.md

Lines changed: 22 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -6,8 +6,8 @@ draft: false
66
tags: ["TrendRadar", "热点聚合", "AI智能分析"]
77
categories: ["projects"]
88
description: "TrendRadar 是一个用于多平台热点聚合和AI智能分析的开源项目,主要功能包括热点新闻监控、智能筛选及推送,同时支持基于MCP协议的深度分析。"
9-
wordCount: 1728
10-
readingTime: 5
9+
wordCount: 2029
10+
readingTime: 6
1111
---
1212

1313

@@ -44,7 +44,26 @@ TrendRadar ([github](https://github.com/sansan0/TrendRadar), 38k stars (截止
4444

4545
## TrendRadar 源码分析
4646

47-
> 本文档使用 UML 标准从多角度描述 TrendRadar 的架构设计,包括系统概览、核心组件、数据流、类结构等。
47+
### **架构深度评价**
48+
49+
#### **1. 架构优势**
50+
51+
- **极致轻量与灵活部署**: 同时支持 GitHub Actions (无状态)、Docker (容器化) 和本地运行,通过 `StorageManager` 智能切换 SQLite/S3 存储,极大降低了使用门槛。
52+
- **模块化核心设计**: 核心模块职责边界清晰 (Crawler, Core, Storage, Notification),新增推送渠道符合开闭原则。
53+
- **前瞻性 AI 融合**: 原生支持 MCP 协议,将传统爬虫转化为 AI Agent 的感知工具,提供自然语言日期解析和趋势分析等高级语义能力。
54+
55+
#### **2. 可扩展性**
56+
- **数据源扩展**: 通用的 `DataFetcher` 接口和 RSS 模块设计证明了对异构数据源的良好兼容性。
57+
- **AI 能力扩展**: FastMCP 2.0 架构使得新增 NLP 工具(如情感分析、摘要生成)非常便捷。
58+
59+
#### **3. 改进方向**
60+
- **并发性能**: `DataFetcher` 目前采用串行抓取,面对大量数据源时 IO 阻塞将成为瓶颈,建议引入异步并发。
61+
- **职责解耦**: `NewsAnalyzer` 类承担了过多编排职责,建议未来拆分为更细粒度的 Pipeline 模式。
62+
- **配置管理**: `config.yaml` 日益庞大,建议引入强类型配置验证 (如 Pydantic) 以减少运行时错误。
63+
64+
---
65+
66+
本文档使用 UML 标准从多角度描述 TrendRadar 的架构设计,包括系统概览、核心组件、数据流、类结构等。
4867

4968
---
5069

Lines changed: 125 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,125 @@
1+
---
2+
title: "每周一个MCP:[MCP Server Name]"
3+
date: "2025-01-01T10:00:00+08:00"
4+
tags: ["mcp", "[Language/Tech]", "[Category]"]
5+
categories: ["projects"]
6+
draft: true
7+
description: "[One-sentence hook: What is this MCP server and what main problem does it solve?]"
8+
wordCount: 567
9+
readingTime: 2
10+
---
11+
12+
## 背景/简介 (Background)
13+
14+
*在这里用通俗易懂的语言介绍这个 MCP Server 是什么,以及为什么我们需要它。*
15+
16+
* **痛点**:在没有这个 MCP Server 之前,我们通常是怎么做的?有什么局限性?(例如:手动复制粘贴、无法访问特定数据、流程繁琐等)
17+
* **解决方案**:这个 MCP Server 提供了什么能力?它是如何改变工作流的?
18+
* **核心价值**:一句话总结它带来的最大好处(例如:“让 Claude 拥有了操作数据库的能力”)。
19+
20+
---
21+
22+
## 核心功能 (Key Features)
23+
24+
*列出该 MCP Server 的主要功能点,可以使用列表或表格形式。*
25+
26+
1. **功能点一**:详细描述。
27+
2. **功能点二**:详细描述。
28+
3. **功能点三**:详细描述。
29+
30+
| 特性 | 传统方式 | 使用本 MCP 后 |
31+
| :--- | :--- | :--- |
32+
| **效率** | 低,需要手动切换窗口 | **高,直接在对话中通过工具完成** |
33+
| **准确性** | 依赖人工复制粘贴,易出错 | **机器自动读取,精确无误** |
34+
35+
---
36+
37+
## 快速开始 (Quick Start)
38+
39+
### 1. 安装 (Installation)
40+
41+
*提供不同环境下的安装命令,例如使用 `uv``pip``npm` 或直接通过 `git`*
42+
43+
**通过 uvx 运行 (推荐)**
44+
45+
```bash
46+
uvx --from [package-name] [command]
47+
# 例如: uvx --from git+https://github.com/xxx/xxx mcp-server-name
48+
```
49+
50+
### 2. 配置 (Configuration)
51+
52+
*展示如何将其添加到 Claude Desktop 或 Cursor/VS Code 的配置文件中。*
53+
54+
**Claude Desktop (`claude_desktop_config.json`)**
55+
56+
```json
57+
{
58+
"mcpServers": {
59+
"mcp-server-name": {
60+
"command": "uvx",
61+
"args": [
62+
"--from",
63+
"package-name",
64+
"server-executable",
65+
"--arg1",
66+
"value1"
67+
],
68+
"env": {
69+
"API_KEY": "your-api-key-here"
70+
}
71+
}
72+
}
73+
}
74+
```
75+
76+
**Cursor / VS Code (MCP Settings)**
77+
78+
* Name: `mcp-server-name`
79+
* Type: `stdio`
80+
* Command: `uvx` (or path to python/node)
81+
* Args: `...`
82+
83+
---
84+
85+
## 使用示例 (Usage Examples)
86+
87+
*提供几个具体的 Prompt (提示词) 示例,展示如何触发和使用该 MCP 的功能。*
88+
89+
### 场景一:[场景名称]
90+
91+
* **用户指令**
92+
> "请使用 [Tool Name] 分析一下这个文件..."
93+
* **预期行为**
94+
> AI 将自动调用工具读取文件内容,并返回分析报告...
95+
96+
### 场景二:[场景名称]
97+
98+
* **用户指令**
99+
> "帮我查找关于 xxx 的信息..."
100+
101+
---
102+
103+
## 最佳实践 (Best Practices)
104+
105+
*分享一些使用技巧、注意事项或高级玩法。*
106+
107+
* **技巧 1**:如何结合其他 MCP 使用?
108+
* **技巧 2**:如何优化 Prompt 以获得更好结果?
109+
* **注意**:无论是权限控制还是 Token 消耗方面的提示。
110+
111+
---
112+
113+
## 总结 (Conclusion)
114+
115+
*简短总结该 MCP Server 的价值,并鼓励读者尝试。*
116+
117+
这个 MCP Server 极大地扩展了 AI 在 [特定领域] 的能力,使得 [特定任务] 变得前所未有的简单。强烈推荐给 [目标用户群体] 尝试。
118+
119+
---
120+
121+
## 参考 (References)
122+
123+
* [GitHub Repository](https://github.com/...)
124+
* [Official Documentation](https://...)
125+
* [Other Related Posts](...)

0 commit comments

Comments
 (0)