How to Read a Paper

(When You're Building a Model)

What’s the goal?

• Extract what’s important
• Figure out how to replicate core results
• Build something that actually works

Step 1: Skim first

• Read the paper once, high-level only
• Focus on:
  • Abstract
  • Intro
  • Discussion

Skip hard parts — it's okay!

Step 2: Zoom in

Re-read the paper:

• Focus on the Results
• Find the core result(s) you’ll replicate

Reproducing a figure is often enough!

Step 3: Deep dive

Go sentence by sentence through Results + Methods
Highlight anything needed to reproduce results:

• Equations
• Algorithms
• Diagrams
• Code snippets
• Implementation notes

Understand everything

For each highlighted thing:

• Use AI (carefully!)
• Watch a video / read a blog
• Try running shared code
• Ask someone
• Take lots of notes

Step 4: Make a task list

• From your notes, write out a to-do list
• Start high-level → break into sub-tasks
• Track your progress (Google Doc, GitHub Projects, etc.)

Tip: Write it out by hand if it helps you think

Missing a detail?

Try this:

• Re-read carefully (look for footnotes!)
• Check supplemental materials
• Look at shared code
• Email the authors!

How to email an author

• Be clear and concise
• Ask all your questions up front
• Be respectful of their time
• Say thank you

Be kind to yourself

This is hard!

• Take breaks
• Talk it out (even to yourself!)
• Go for a walk
• Sleep on it
• Keep chipping away

Final tips

• Building from primary sources is real science
• Confusion is part of the process
• Ask questions, keep notes, keep going!

You got this