Introduction

How to Learn So Knowledge Actually Sticks

This course has no shortage of content — what's missing is your 30-second pause to reflect.

Let's start with a harsh truth

If you just scroll through from start to finish and think "got it, all makes sense," chances are you'll remember nothing three days later.
This isn't your fault — it's how the brain defaults: Input ≠ Understanding, Understanding ≠ Retention, Retention ≠ Application.

Watching vs. Learning: What's the difference?

Just watching Actually learning
See a case study "Oh, so that's how it works." "Could I apply this thinking to my situation?"
See a concept "I've memorized that term." "What fundamental problem does it solve?"
See a pitfall "Someone else fell into that. Good to know." "Does my project have a similar trap?"
Finish a lesson "Next one." "Wait — let me restate this in my own words."

The Three-Step Loop: Making Knowledge Truly Yours

See
Receive
→
Reflect
Pause & think
→
Transfer
Apply to context
→
Output
Teach someone
1

See: Read with a question, not on autopilot

Before opening each page, ask yourself: How does this topic relate to what I'm currently building? Even if you can't answer immediately, the question itself sharpens your focus.

Example: Before opening the "Context Window" page
Ask first: Our product's conversations often run very long — is that related to context window limits? When users say "you forgot what I just said," is the context being truncated?
2

Reflect: After each concept, pause for 30 seconds

Don't rush to the next page. Ask yourself three questions:

1. What fundamental problem does this concept solve?
2. What would I have done before learning this?
3. How will my approach change now that I know this?

Example: After finishing the "Hallucination" section
Reflect: So hallucination isn't a bug — it's an inevitable consequence of probabilistic sampling. That means my old approach of expecting the AI to produce accurate answers directly is flawed. I need to design a verification step; expecting the model to never be wrong is unrealistic.
3

Transfer: Map it to your own business context

This is the most critical step. Every case study and design decision in this course must be translated into your business language.

Different industries, product types, and user bases mean the same technical solution has vastly different applicability. This course teaches thinking frameworks, not copy-paste answers.

Example: After finishing the "RAG (Retrieval-Augmented Generation)" section
Transfer: Our customer service system has 2,000 knowledge base documents. What do RAG's recall-rate and precision problems look like in our context? When a user asks a question spanning multiple documents, how should I design the retrieval strategy? How does our document format (scanned PDFs vs. structured text) affect different stages?
4

Output: Teach someone, or write it down

The essence of the Feynman Technique: if you can't explain it simply to a non-expert, you don't truly understand it yourself.

No need to write an essay. Even dropping a message in a team chat saying "learned something today: xxx — I used to think yyy, but it's actually zzz" pushes knowledge from short-term into long-term memory.

Example: Trying to output
Tell a colleague: "Did you know? LLM Tokens aren't counted by character. A single Chinese character can be 1–3 Tokens. That Prompt template we have looks like 500 characters, but it might actually consume 1,500 Tokens — no wonder it hits the limit so often."

Page Takeaways