If you do anything adjacent to ML, you know the feeling: NeurIPS drops, then ICML, then ICLR, then CVPR, and suddenly there are thousands of papers you're "supposed to" have read. My old workflow was 20 open tabs, a bookmarks folder I never revisited, and a low-grade sense of guilt.
This week I tried Paper List (https://paperlist.ai/), an explorer for AI conference papers, and it fixed one specific problem for me: going deep on a single topic instead of trying to boil the ocean.
Here's the workflow that clicked:
- Start from a topic, not a conference. I searched "retrieval-augmented generation" and got 366 papers spanning 13 conferences over 5 years, in one view.
- Filter down. Narrow by year and by conference (ICLR / NeurIPS / ICML / CVPR) so you end up reading the 15 papers that matter for your question, not the 3,000 that don't.
- Follow the thread over time. Because it spans multiple years, you can actually see how an idea evolved instead of reading one paper in isolation.
Why this beats my old habits:
- Topic-first framing matches how I actually work. I start with a problem ("how are people evaluating RAG?"), not with a conference table of contents.
- Cross-conference and cross-year in one place means less tab-hoarding and fewer "I'll read it later" lies.
- It lowers the activation energy to explore a subfield you're only mildly curious about.
It won't read the papers for you, and it won't replace your judgment about what's worth your time. But as a front door into a topic, it's the least stressful tool I've used for this.
If your "to read" list has quietly become a graveyard, try this: pick one topic you actually care about and explore it end to end. It's a much better feeling than 20 open tabs.
Top comments (0)