Disclosure: I contribute to the OpenNomos ecosystem, which Paper List is part of. Opinions and reading habits below are my own.
I'm not a researcher. I work in brand and content strategy, and a growing part of my job is understanding where AI is actually heading — not where the loudest thread on my timeline says it's heading.
For a long time my "system" for following AI research was embarrassing: bookmark an arxiv link someone shared, feel productive, never open it again. My bookmark folder wasn't a reading list. It was a graveyard with good intentions.
Here's what finally worked for me, as someone who reads about papers more than she reads papers.
The real problem isn't volume, it's ordering
Everyone says the problem with AI research is volume — hundreds of papers a day, impossible to keep up. True, but incomplete.
The deeper problem is that most of us encounter papers in engagement order: whatever got quote-tweeted the hardest this week. Engagement order is great for discovery and terrible for understanding. You see the outliers, never the baseline. You know the three papers everyone joked about, but you can't answer a basic question like: is this topic growing, or is it three people being loud?
For a non-researcher, that distinction matters more than any individual paper. My job doesn't require me to reproduce results. It requires me to know which directions are compounding and which are noise.
Switching from "feed" to "shelf"
The mental shift that helped: stop treating research as a feed to keep up with, and start treating it as a shelf to walk past occasionally.
A feed punishes you for being away. A shelf just sits there, organized, waiting until you have a real question.
Concretely, my questions usually look like:
- Is retrieval-augmented generation still an active area, or has it been absorbed into something else?
- When did diffusion models jump from vision venues to everywhere?
- What did last year's NeurIPS actually emphasize, versus what Twitter said it emphasized?
None of these are answered by reading one paper. All of them are answered by looking at topic × venue × year — which is a browsing structure, not a feed.
How I actually do it now
The tool I use for this is Paper List. It organizes papers from top venues by topic and year, which sounds almost boringly simple — and that's the point.
My routine, maybe twice a month:
- Pick one question, not a topic. "Is agent evaluation maturing?" beats "let me check on agents."
- Scan the topic across 2–3 years. Count roughly, don't read. Ten minutes tells you whether a direction is expanding, plateauing, or migrating between venues.
- Pick at most two papers whose abstracts directly touch my question. Read the abstract, intro, and conclusion. Skip the math without guilt.
- Write three sentences in my notes about what changed since I last looked. If I can't write them, I didn't actually learn anything — scan again.
That last step is the whole system, honestly. The tool gives you structure; the three sentences force synthesis. One without the other doesn't stick.
What this replaced
Before: forty open arxiv tabs, a bookmark folder I feared, and a vague anxiety that everyone else understood "the state of AI" better than I did.
After: a shelf I visit with a question, and a notes file that slowly became my own timeline of the field. When someone asks "is X still a thing," I can answer with a shape — growing since 2023, mostly at ICLR, spreading into ACL — instead of a shrug or a hot take.
The anxiety didn't fully disappear. But it turns out most of it was never about missing papers. It was about having no structure to put them in.
If you're also a non-researcher
You don't need a daily reading habit. You need:
- a recurring question, tied to your actual work
- a structured place to check it against (topic × venue × year)
- a tiny writing ritual to force synthesis
That's sustainable at two hours a month. Keeping up with a feed is not sustainable at any number of hours — I've tried.
If your bookmark folder is also a graveyard, maybe don't organize the graveyard. Change where the papers live instead.
Top comments (0)