Ask HN: How do you stay on top of new research?

9 points by just_human ↗ HN
I used to use Karpathy's gret arxiv-sanity.com but it now redirects to a paid product (undermind.ai). It's been a while since I've tried to read new research papers on a daily/weekly basis. How do others stay on top of new research papers?

7 comments

[ 2.9 ms ] story [ 25.9 ms ] thread
I created a custom feed of research links ingested from HN, Bluesky, and Reddit subreddits firehoses that meet url regex criteria (DOIs or paper repo TLDs) that funnel into karakeep.app for reading at my convenience. Firehose->folder.

(i would also ingest twitter/x firehose, but the access story is meh, alas)

You can't. 10k articles are published daily. It's doubtful you'll have time to even download them all.
Similar to a rule from zombie movies - Fight only when you absolutely must, run when you can. 'Fight' meaning reading a paper.
You can use undermind.ai on the free plan. I made two searches when they launched and still occasionally get updates with new relevant papers. Unfortunately, those papers are typically only new and relevant, but not novel. (Not a knock against the search quality. In an area where data acquisition is expensive, everyone working with the same handful of small datasets is probably unavoidable.) But the initial searches surfaces some old papers I had missed, so I'd say it's worth a try.
I try to keep a curated list of researchers on x.com.
The only way I know how is to see what research gets mentioned by 2 or more people I respect. Then I might check it out.

But there's a lot of bad ML research out there, my default is to ignore it.

google scholar and alerts for references to a curated author list