---
title: "AI 研究论文与预印本 - 信源与标准 - AI·外脑 | 思行致远"
canonical: "https://siyuanailab.com/knowledge/sources/ai-research-papers-directory/"
dateModified: "2026-07-07"
datePublished: "2026-07-07"
---

打开 AI·外脑目录  

 

AI·外脑

 [返回专栏首页](https://siyuanailab.com/knowledge/)

 ×

 

站内搜索 

⌕

   

主要栏目

 

- [AI 核心概念](https://siyuanailab.com/knowledge/ai-core/)

- [工作流](https://siyuanailab.com/knowledge/workflows/)

- [工具配置与选型](https://siyuanailab.com/knowledge/toolkits/)

- [信源与标准](https://siyuanailab.com/knowledge/sources/)

- [场景案例 待开发](https://siyuanailab.com/knowledge/scenarios/)

- [职业与能力](https://siyuanailab.com/knowledge/career/)

 

专题

 

- [从这里开始](https://siyuanailab.com/knowledge/start/)

- [获取与安装](https://siyuanailab.com/knowledge/install/)

- [建设日志](https://siyuanailab.com/knowledge/ops/)

  

[信源与标准子目录](https://siyuanailab.com/knowledge/sources/) 

信源分级与框架（1）

 

- [AI 学习信源：如何分级与使用](https://siyuanailab.com/knowledge/sources/ai-source-principle/)

官方文档与产品更新（6）

 

- [Google AI 官方动态](https://siyuanailab.com/knowledge/sources/google-ai-official-updates/)

- [MCP 官方文档](https://siyuanailab.com/knowledge/sources/mcp-official-docs/)

- [Midjourney 官方文档](https://siyuanailab.com/knowledge/sources/midjourney-official-docs/)

- [OpenAI 官方文档](https://siyuanailab.com/knowledge/sources/openai-official-docs/)

- [Anthropic 官方文档](https://siyuanailab.com/knowledge/sources/anthropic-official-docs/)

- [AI 官方文档与产品更新](https://siyuanailab.com/knowledge/sources/ai-official-docs-directory/)

研究论文与预印本（6）

 

- [Anthropic Research](https://siyuanailab.com/knowledge/sources/anthropic-research/)

- [arXiv](https://siyuanailab.com/knowledge/sources/arxiv/)

- [Google Research Blog](https://siyuanailab.com/knowledge/sources/google-research-blog/)

- [Lilian Weng](https://siyuanailab.com/knowledge/sources/lilian-weng/)

- [OpenAI Research](https://siyuanailab.com/knowledge/sources/openai-research/)

- [AI 研究论文与预印本](https://siyuanailab.com/knowledge/sources/ai-research-papers-directory/)

权威媒体与机构洞察（7）

 

- [a16z AI](https://siyuanailab.com/knowledge/sources/a16z-ai/)

- [MIT Technology Review AI](https://siyuanailab.com/knowledge/sources/mit-technology-review-ai/)

- [Sequoia AI](https://siyuanailab.com/knowledge/sources/sequoia-ai/)

- [TechCrunch AI](https://siyuanailab.com/knowledge/sources/techcrunch-ai/)

- [The Verge AI](https://siyuanailab.com/knowledge/sources/the-verge-ai/)

- [Y Combinator Blog](https://siyuanailab.com/knowledge/sources/y-combinator-blog/)

- [AI 权威媒体与机构洞察](https://siyuanailab.com/knowledge/sources/ai-media-insights-directory/)

深度播客与视频（6）

 

- [AI & I](https://siyuanailab.com/knowledge/sources/ai-and-i/)

- [Andrej Karpathy（YouTube）](https://siyuanailab.com/knowledge/sources/andrej-karpathy-youtube/)

- [Google DeepMind Podcast](https://siyuanailab.com/knowledge/sources/google-deepmind-podcast/)

- [Latent Space](https://siyuanailab.com/knowledge/sources/latent-space/)

- [Lenny's Podcast](https://siyuanailab.com/knowledge/sources/lennys-podcast/)

- [AI 深度播客与视频](https://siyuanailab.com/knowledge/sources/ai-podcasts-videos-directory/)

Newsletter 与策展者（7）

 

- [AI Valley](https://siyuanailab.com/knowledge/sources/ai-valley/)

- [AINews by smol.ai](https://siyuanailab.com/knowledge/sources/ainews-by-smol-ai/)

- [Ben's Bites](https://siyuanailab.com/knowledge/sources/bens-bites/)

- [Every](https://siyuanailab.com/knowledge/sources/every/)

- [Peter Yang (Behind the Craft)](https://siyuanailab.com/knowledge/sources/peter-yang/)

- [Zara's AI learning library](https://siyuanailab.com/knowledge/sources/zaras-ai-learning-library/)

- [AI Newsletter 与策展者](https://siyuanailab.com/knowledge/sources/ai-newsletters-curators-directory/)

社交媒体账号（6）

 

- [Andrej Karpathy（X）](https://siyuanailab.com/knowledge/sources/andrej-karpathy-x/)

- [Greg Isenberg（X）](https://siyuanailab.com/knowledge/sources/greg-isenberg-x/)

- [Justine Moore（X）](https://siyuanailab.com/knowledge/sources/justine-moore-x/)

- [Lenny Rachitsky（X）](https://siyuanailab.com/knowledge/sources/lenny-rachitsky-x/)

- [Swyx（X）](https://siyuanailab.com/knowledge/sources/swyx/)

- [AI 社交媒体账号](https://siyuanailab.com/knowledge/sources/ai-social-accounts-directory/)

  

信源与标准 · 进阶

 

# AI 研究论文与预印本

 

杨思远 / 思行致远 · 2026/7/7 · 置信度：高

 

信源研究论文预印本AI学习独立站

 

论文和预印本是 AI 领域第 1.5 级信源：比官方文档更深入原理，但阅读成本更高，需要判断可信度与相关性。

 

# AI 研究论文与预印本

 

> 论文和预印本是 AI 领域第 1.5 级信源：比官方文档更深入原理，但阅读成本更高，需要判断可信度与相关性。

 

---

 

## 一、是什么

 

这张卡整理了 AI 领域主流的研究论文与预印本入口。它们属于第 1.5 级信源，适合用来理解模型原理、训练方法、安全对齐和产业趋势，但通常需要一定技术基础或耐心阅读。

 

## 二、为什么存在

 

AI 领域变化快，三类需求让论文/预印本成为必要入口：

 

- **追源头**：模型能力、训练技巧、评估方法最早出现在论文中。
 
- **理解边界**：官方博客不会写全的失败案例、限制条件、安全分析，往往在论文里。
 
- **建立判断框架**：长期阅读能帮你识别哪些创新是本质突破，哪些是工程包装。

 

## 三、核心机制

 

| 来源 | 形态 | 一句话定位 | 代表作品/入口 | 细节卡 |
| --- | --- | --- | --- | --- |
| arXiv | 预印本平台 | AI/ML 论文最集中的开放仓库 | [https://arxiv.org/](https://arxiv.org/) | [arXiv](https://siyuanailab.com/knowledge/sources/arxiv/) |
| Anthropic Research | 研究博客 | Anthropic 在可解释性、对齐、能力上的研究输出 | [https://www.anthropic.com/research](https://www.anthropic.com/research) | [Anthropic-Research](https://siyuanailab.com/knowledge/sources/anthropic-research/) |
| Lilian Weng | 个人技术博客 | 以清晰长文解释 RLHF、Agent、安全等复杂主题 | [https://lilianweng.github.io/](https://lilianweng.github.io/) | [Lilian-Weng](https://siyuanailab.com/knowledge/sources/lilian-weng/) |
| Google Research Blog | 研究博客 | Google 在 Gemini、DeepMind、系统与应用研究上的发布入口 | [https://research.google/blog/](https://research.google/blog/) | [Google-Research-Blog](https://siyuanailab.com/knowledge/sources/google-research-blog/) |
| OpenAI Research | 论文索引 | OpenAI 在生成模型、对齐、系统上的研究论文库 | [https://openai.com/research](https://openai.com/research) | [OpenAI-Research](https://siyuanailab.com/knowledge/sources/openai-research/) |

 

### 3.1 阅读优先级

 

- **查事实与参数** → 优先官方文档（sources-02）。
 
- **理解原理与机制** → 看论文/预印本。
 
- **找观点与行业解读** → 看权威媒体和可信第三方（sources-03 及以后）。

 

### 3.2 预印本 vs 已发表论文

 

| 维度 | 预印本（如 arXiv） | 已发表论文（如顶会/期刊） |
| --- | --- | --- |
| 速度 | 快，研究完成后即可发布 | 慢，经过审稿周期 |
| 可信度 | 未经过同行评审，需自行判断 | 经过评审，质量相对有保障 |
| 用途 | 追新、找思路、看原始方法 | 验证成熟结论、引用 |

 

## 四、常见误解

 

| 误解 | 真相 |
| --- | --- |
| 论文一定很难读 | 很多论文摘要、图表和结论清晰，非技术读者也能抓住核心判断 |
| arXiv 论文都经过审稿 | arXiv 是预印本平台，只审核基本分类，不保证正确性 |
| 企业研究博客都是营销 | Anthropic、Google、OpenAI 的研究博客常附带论文链接和技术细节 |
| 读了论文就能跟上 AI | 论文是输入之一，还需结合官方文档、产品更新和实践反馈 |

 

## 五、怎么用

 

### Level 0：收藏入口

 

把这张目录卡中的 5 个主站加入书签或 RSS 阅读器。

 

### Level 1：会读摘要

 

读论文时先看标题、摘要、结论和图表，决定是否深入方法细节。

 

### Level 2：会追引用

 

从一篇论文的 Related Work 和 References 出发，快速建立对一个子领域的认知地图。

 

### Level 3：建立个人论文流

 

根据自己的方向（如 Agent、对齐、多模态），在 arXiv 订阅对应分类（cs.AI、cs.CL、cs.LG），每月扫一次标题。

 

## 六、延伸阅读

 

- 原则卡：[AI学习信源-如何分级与使用](https://siyuanailab.com/knowledge/sources/ai-source-principle/)
 
- 目录卡：[AI官方文档与产品更新](https://siyuanailab.com/knowledge/sources/ai-official-docs-directory/)
 
- 关联概念：`[[大语言模型]]` / `[[提示词工程]]` / `[[AI 安全]]`

 

### 适用场景与行动

 

**场景：**需要建立 AI 研究论文与预印本 的信源目录和选择顺序时。

 

**适合谁：**AI 学习者、研究跟踪者、需要追溯一手材料的产品与内容工作者。

 

**下一步：**按卡片中的分级与链接挑选 2-3 个来源，放入自己的固定阅读清单。

 

### 相关卡片

 

- ai-source-principle

- arxiv

- anthropic-research

- lilian-weng

- google-research-blog

- openai-research

 

这张卡片由杨思远基于真实工作流和实测经验整理。

 

如发现信息过时或有误，可通过 [/contact](https://siyuanailab.com/contact) 反馈。

  

[← 上一张：AI 深度播客与视频](https://siyuanailab.com/knowledge/sources/ai-podcasts-videos-directory/) [下一张：AI 社交媒体账号 →](https://siyuanailab.com/knowledge/sources/ai-social-accounts-directory/)

 [返回 AI·外脑入口](https://siyuanailab.com/knowledge/)