M2 · Lesson 2 — Reading Papers Effectively

The 3-Pass
Reading Method

Don't read papers from start to finish.
Invest only as much time as the paper deserves.

01
M2 · L2 — The Problem

Why most people read wrong

Linear reading is
slow and inefficient

"You try to understand everything at once, get lost in equations, and finish 2 hours later unsure what the paper even argued."

The 3-pass method gives you a decision framework:

  • Each pass answers one specific question
  • You stop when you have what you need
  • Most papers only deserve Pass 1 or Pass 2
02
M2 · L2 — Overview

The framework

Three passes,
three questions

Pass 01

The Skim

5–10 min

Should I even read this paper?

Pass 02

The Read

30–60 min

What are they claiming?

Pass 03

The Deep Read

1–2 hrs

Do I believe them?

03
M2 · L2 — Pass 1

Pass 1 · 5–10 minutes

The Skim

What you read

  • Title + Abstract
  • First and last paragraph of Introduction
  • Section headings only
  • Every figure caption
  • Conclusion

Output

A 3-sentence summary. If you can't write it after Pass 1 — the paper is poorly written or not worth your time.

💡 Figure captions are gold. Most papers tell their entire story through 2–3 figures. Read captions before reading equations.

04
M2 · L2 — Pass 2

Pass 2 · 30–60 minutes

The Read

What you read

  • Everything except proofs + appendices
  • Every equation — don't derive, just understand what it says
  • Experimental setup + results tables carefully
  • Related work quickly

Output

Filled paper notes template. You should be able to explain this paper to a colleague after Pass 2.

💡 Read experiments critically. Which dataset shows the weakest improvement? Which baseline is closest to theirs? These are the cracks.

05
M2 · L2 — Pass 3

Pass 3 · 1–2 hours

The Deep Read

What you read

  • Every proof and derivation
  • Appendix details
  • Implementation details
  • Every footnote

Reserve Pass 3 for: papers in your direct research area, papers you plan to reproduce, or papers you're reviewing.

Output

  • Are the assumptions valid?
  • Do the derivations hold?
  • Could I reproduce this?
  • What would I do differently?

Most papers don't need Pass 3. After Pass 2 you already have 90% of what you need.

06
M2 · L2 — Decision Points

The key skill

Knowing when
to stop

After Pass 1 — STOP if:

Not relevant to my research
I already know this approach well
Weak venue + suspicious claims

After Pass 2 — STOP if:

I understand the contribution

Continue to Pass 3 only if:

I need to reproduce this paper
I'm reviewing this for a venue
I want to propose an extension
This is directly in my research area
07
M2 · L2 — Worked Example

Applied to K-RagRec

What each pass
would look like

Pass 1 · 8 min

You already did this

LLM hallucination → noisy RAG → K-RagRec fixes this with KG subgraph retrieval. Relevant to me. Continue.

Pass 2 · 45 min

Read methodology + experiments

Decode Eq. 3 (hop-field indexing), Eq. 5 (retrieval). Notice LLaMA-3 improvements are smaller. Spot missing traditional RS baselines. Fill paper notes.

Pass 3 · if needed

Deep dive for extension work

Is the GNN indexing actually novel vs GraphSAGE? Is p=50% threshold justified? Would this work without a KG?

08
M2 · L2 — Key Takeaways

What to remember

Pass 1

5–10 min

Figure captions tell the whole story. If you can't write 3 sentences after — stop.

Pass 2

30–60 min

Fill your notes template. Read experiments critically. This is enough for 90% of papers.

Pass 3

1–2 hrs

Reserve for papers you'll reproduce, review, or extend. Don't default here.

Next: M2 · L3 — Critical Questions to Ask Every Paper

09
← → arrow keys to navigate
1 / 9