# AI Dungeon Run | Andrew Ng's AI For Everyone – Chapters 3 & 4

https://ijuhsu.com/en/posts/ai-for-everyone-ch3-4 · 2025-08-24
Category: ai · Tags: ai, Andrew Ng, AI For Everyone, 讀書心得

---

# Ch. 3: Building AI In Your Company

A smart speaker, for instance, might involve several steps built by separate teams:

*   Trigger word detection
*   Speech recognition
*   Intent recognition
*   Execute

A typical AI team might include:

*   Software engineers (often 50%+)
*   ML engineers: train the A → B mapping from data
*   ML researchers: more academic, focused on pushing the state of the art
*   Data scientists: turn data into insights
*   Data engineers: organize and manage data infrastructure
*   AI Product Manager: figure out what to build, and what's actually valuable

Bottom line: **No project is too small.** A small but successful project is still a win.

## The AI Transformation Playbook

1.  **Execute pilot projects to build momentum.** The most important thing first is establishing successful proof-of-concepts — this gives everyone the motivation to keep going. Can be in-house or outsourced. Doesn't need to be the *most valuable* project, just a *successful* one. Expect 6–12 months.
2.  **Build an in-house AI team.** A dedicated AI team, separate from business units, is better positioned to find the right strategy.
3.  **Provide broad AI training** across the organization.
4.  **Develop an AI strategy.** Some companies put strategy first, but that often backfires — you can't make good strategy without knowing what AI can actually do. Build first, find users, gather data, then make better products. Be strategic about data collection.
5.  **Develop internal and external communications** — with investors, users, talent, and the broader internet.

* * *

# Ch. 4: AI and Society

We should stay balanced about AI — neither blindly optimistic that it'll solve all human problems, nor terrified it will destroy us. The grounded view: **AI is a powerful tool, but it has limits.**

Those limits include:

1.  **Performance limitations** — it can't do everything well
2.  **Explainability is hard** — AI can do things well without being able to explain why; without understanding what drives success or failure, improvement is hard
3.  **Bias from biased data** — garbage in, garbage out; biased training data produces biased results
4.  **Adversarial attacks** — bad actors can exploit AI to produce incorrect or harmful outputs

Yes, AI will eliminate many jobs. But it will also create many new ones. The point: build on your existing skills and expertise, then layer AI on top. Tearing everything down and starting over is rarely the right move.

* * *

# Wrap-Up

Andrew Ng is a genuinely great teacher — he made a lot of previously fuzzy concepts click for me. If you're new to AI like I was, this course is hard to beat.

After finishing it, I feel much more oriented about where I want to develop. I've already started Andrew Ng's Machine Learning specialization — it's fun but hard, and there's this excited feeling that a bunch of projects I couldn't crack before are suddenly starting to make sense.

I'll keep sharing as I go!

* * *

# **Thanks for reading :D**

If you enjoyed this post, **feel free to click the coffee button in the lower right to support us and give Lottery a can 🐾**