How to (actively) read a paper

originally from: https://www.eugenevinitsky.com/posts/paper-games/

TLDR: the external experience and the internal experience of reading a paper are extremely different from everyone. You can't know how different unless we share how we do it so here are some ideas and games I use to engage more with a paper.

A template

The goal of this template is to help you avoid the situation where you've been reading a paper for an hour or two but come away realizing you've in fact learned nothing. The problem is that active engagement is the only way you actually learn something (seriously, passive consumption teaches you almost nothing, read about it. It's horrifying and once you realize it you stop wanting to ever passively consume educational material.) but many people don't have a model of what active engagement looks like for paper reading.

Why am I reading this paper?

Fill in this section to convince yourself that it's worthwhile to invest the time into reading this paper. If you can't fill it in, maybe you shouldn't actually read the paper? Some options:

• The abstract indicates that it might have the solution to a problem I'm working on

• The abstract indicates that it's highly related to the topic that I'm working on and I want to build out my breadth of knowledge in this field.

• It's blowing up on twitter (this isn't a great one)

• Having skimmed the conclusion, the result seems important

• It contains a new idea in my field that I wasn't aware of

• A friend suggested it because it is related to my subfield and they have good taste

• I'm very curious about the question posed in this paper

• It's written by one of my "must-read" authors who consistently do good work

Summary

• 1 sentence summary

• Paragraph summary

• Complete summary (I do this for vanishingly few papers)

Unresolved question

What bit of the paper didn't you understand? You can link this to a main sheet so that maybe you can come back to it later and see that you get it now. Obsidian is great for this because you can just tag anything and then create a unified page that aggregates everything.

Once you've completed this section, you might consider stopping if the "Why am I reading this paper" section was not super compelling. Everything below requires a lot more work and should not be done for most papers.

New math

Did you learn some new, useful theorem? You can summarize it and link it to a main sheet. Maybe convert it into some Anki cards!

Paper games

Here are some "games" you can play that are useful for making sure you're not blankly scanning the paper without real comprehension.

• Can I explain this section to someone in

  • < 1 minute who has a ton of technical background

  • < 3 minutes to someone with less background in this area

• Can I explain this section to someone in enough depth to last 3 minutes

• Imagine there is a panel grilling you on the paper, what question might they ask you that you wouldn't be able to answer?

• Try to complete the proof before reading it. This can be hard! You can also just try to fill in the next step in the proof

• What experiment would I have run to make this paper more convincing?

• Why did the author write this paper? What larger agenda might it fit into?

• If I were to do a new project based on what's in this paper, what would it be?