I should read a few more papers first” might be my favorite form of procrastination
Second-year ML PhD and I’m starting to think “I should read a few more papers first” is just procrastination with citations.
Every new research direction goes the same way.
I start with 5 papers. Then it becomes 15. Then 30+. Zotero is full, my Notion page is getting ridiculous, I’ve asked ChatGPT to summarize a few things, used Connected Papers to chase related work, and tried Mira by Deep Principle to compare some of the papers.
And somehow I still haven’t run an experiment.
That’s the part that gets me. You can spend days doing “research” and feel productive the entire time, while still being unable to answer a very basic question:
What exactly am I testing?
Recently I stopped trying to understand an entire subfield before touching code. I just looked for recurring baselines, common metrics, disagreements between papers, and one or two things I could actually falsify.
That got me to a baseline and a couple experiments way faster than another 20 PDFs probably would have.
Now I’m wondering if the real skill isn’t reading more papers, but knowing when you have enough information to start being wrong.
How do you decide when to stop reading and just run the experiment?