Take a paper apart

A recent paper from your field. Find its biggest problem, then try to fix it.

It worked for me

  • I did this with a recent paper in my own field.
  • It found a real problem, then a new estimator: valid under weaker assumptions, and better than the one in the paper.
  • My only input was pushing.
  • I don’t know if it works in your field. Let’s find out.

Where your time goes

The usual way Reading the paper and the literature: weeks judging With the agent judging: the problem, the fix, the field it reads: 1 min · it tries a fix: 10 min

You know the field. The agent does the reading; you do the judging.

Get a paper

  • A recent paper in your field: one you know well, or one you want to read.
  • Make a folder for it, and start the agent there.
  • Open link (NBER, arXiv): the agent downloads it itself.
  • SSRN or a journal site: download it in your browser, then ask the agent to move it into your folder.

Some things to ask

What’s the main claim here?

What’s the biggest methodological problem?

Can you fix it?

Then push

  • “Show me where in the paper.”
  • “Do they address it anywhere, including the appendix?”
  • “Is that really the biggest problem?”
  • “Show that the fix works.”

A fix takes 5–10 minutes: it writes and runs code. If it asks to install Python, say yes.

Round-up

  • Best problem found?
  • Best fix?
  • Worst false alarm?
  • Keep the folder. After lunch, everything starts from it.

After lunch: you choose