Auto-sync: 2026-04-18 05:18
This commit is contained in:
45
wiki/concepts/source-grounding.md
Normal file
45
wiki/concepts/source-grounding.md
Normal file
@@ -0,0 +1,45 @@
|
||||
---
|
||||
id: source-grounding
|
||||
title: "Source-grounding"
|
||||
type: concept
|
||||
tags: [AI, accuracy, citation]
|
||||
sources: []
|
||||
last_updated: 2025-04-18
|
||||
---
|
||||
|
||||
## Definition
|
||||
NotebookLM 的核心机制——仅使用用户上传的文档作为知识库,确保 AI 输出准确且可验证。
|
||||
|
||||
## How It Works
|
||||
1. 用户上传文档(PDF、网页、YouTube 视频等)
|
||||
2. NotebookLM 仅在这些文档范围内回答问题
|
||||
3. 每个答案都附带精确的文档引用
|
||||
4. 点击引用可一键跳转到原文位置
|
||||
|
||||
## Advantages
|
||||
- **准确性**:避免 AI 幻觉,答案仅来自可信来源
|
||||
- **可验证性**:每个答案都有引用,可人工核实
|
||||
- **可控性**:知识库完全由用户掌控
|
||||
- **隐私性**:文档不进入通用模型训练
|
||||
|
||||
## Comparison with General AI
|
||||
| 特性 | Source-grounding | General AI |
|
||||
|------|----------------|-----------|
|
||||
| 知识范围 | 用户上传的文档 | 互联网公开数据 |
|
||||
| 准确性 | 高(可验证) | 中(可能幻觉) |
|
||||
| 引用 | 精确到文档位置 | 可能无引用 |
|
||||
| 可控性 | 完全可控 | 不可控 |
|
||||
|
||||
## Use Cases
|
||||
- 研究文献综述
|
||||
- 法律文档审查
|
||||
- 技术文档学习
|
||||
- 跨版本对比
|
||||
|
||||
## Related Concepts
|
||||
- [[Source Citation]]
|
||||
- [[RAG]]
|
||||
|
||||
## Aliases
|
||||
- Grounding
|
||||
- Source Citation
|
||||
Reference in New Issue
Block a user