How do I reduce hallucination when AI answers from internal knowledge?

The chapter's prompt enforces two constraints: (1) “if knowledge is missing, say so — do not fabricate”; (2) “list the knowledge sources used before answering”. Add a length cap — each answer under 80 chars. Forcing AI to cite sources before generating separates “I don't know” from “I guessed” at generation time, which is far safer than post-hoc audit.

Should I paste whole docs or use @ references when feeding internal knowledge to AI?

Use `@` to reference specific docs — do not paste entire sites. The chapter has AI summarize first, then enter generation with the summary in hand. This cuts token use and sharpens signal. Pasting everything dilutes attention and burns context — important details get drowned in noise.

What rules apply when handling internal data with PII?

Three rules: (1) redact PII before sending — request logs never store the raw text; (2) do not upload internal PDFs/spreadsheets to external models — prefer self-hosted/private models; (3) control log retention and label data purpose. Compliance is a prompt-layer boundary, not a post-hoc patch.

How do I orchestrate RAG when AI needs both DB and model calls?

Three steps: (1) have AI produce the query terms (keywords or SQL); (2) hit the DB or search API for results; (3) generate the final answer from those results. The key is making AI surface the query terms or SQL — without it, when something goes wrong, you can't tell whether retrieval missed or generation hallucinated.

For the “generate onboarding FAQ” exercise, how do I verify AI did not fabricate?

The exercise requires AI to cite paragraph numbers. After generation, manually cross-check each FAQ answer against the source wiki paragraphs to confirm the content actually exists there. If citations don't match, tighten the prompt — e.g. require AI to append `(source: para N)` after each answer.

Data & Knowledge Base Collaboration

学AI来匠人

📖

Vibe Coding

38 chapters

·Introduction

基础篇6

01
Vibe Coding 是什么
5分钟
02
安装 Cursor 与环境准备
10分钟
03
写出第一个 Prompt
10分钟
04
Vibe Coding 思维模型
8分钟
05
高质量 Prompt 模板
12分钟
06
理解与校验 AI 的回复
10分钟

进阶篇18

07
🔒用 AI 做 Debug
20分钟
08
🔒与 AI 结对编程
18分钟
09
🔒多文件协作与项目级修改
25分钟
10
🔒交付一个小功能的全流程
30分钟
11
🔒用 AI 做测试与质量保障
20分钟
12
性能与成本优化
18分钟
13
安全与隐私最佳实践
15分钟
14
🔒团队协作与 PR 流程
18分钟
15
工具与模型最新动态
12分钟
16
🔒AI Agents 与工具调用
25分钟
17
🔒前后端协同与 API 生成
25分钟
18
🔒数据与知识库协作
22分钟
19
持续改进与自动化
18分钟
20
🔒AI 辅助重构与架构演进
22分钟
21
AI 辅助 UX 文案与交互
15分钟
22
🔒可观测性与 Guardrails
18分钟
23
脚手架与自动化交付
15分钟
24
🔒实战：Todo App
45分钟

工具篇13

25
🔒MCP 服务器配置
20分钟
26
🔒MCP 配置同步
15分钟
27
Cursor Rules 编写指南
15分钟
28
Git Commit 规范与 Emoji
10分钟
29
五个为什么分析法
8分钟
30
Mermaid 图表生成
12分钟
31
任务实施方法论
10分钟
32
Claude Code Skills 进阶
15分钟
33
Claude Code + GLM 4.6 配置
15分钟
34
Claude Code 自定义命令
12分钟
35
Claude Code 子代理配置
15分钟
36
Claude Code 全面拆解
25分钟
37
Claude Skill Codebook 简明教程
12分钟

实战篇1

38
Vibe Coding 全流程 SOP
15分钟

Resources

🗺️ Learning Roadmap
📚 Wiki Docs
📋 Cheat Sheets

Home/Learn/Vibe Coding/Data & Knowledge Base Collaboration

🎯Learn More Efficiently

Keep your learning uninterrupted

✏️
Study Notes
Take notes while learning, synced to cloud
📈
Progress Tracking
Auto-save, resume anytime
🏅
Completion Certificate
Shareable learning achievement
💬
Community Q&A
Connect with 2000+ learners

Sign Up Free→Already have an account? Sign in

👥2,847 learners active

AI学习面试课程学习中心