Show HN: ArXiv Feed – An easy way to keep up with AI research | Hacker News Reader

Show HN: ArXiv Feed – An easy way to keep up with AI research

arxiv-feed.vercel.app

32 pointssubtech17 comments

Hey HN!

I've always found it hard to keep up with the latest AI research, so I built Arxiv Feed!

https://arxiv-feed.vercel.app/

It's basically a feed of AI research papers + a one-liner explaining what problem its solving, etc.

You can also click on any paper to get a TL;DR.

Right now, I've only indexed a few hundred large language model papers, but will expand to indexing AI papers in other topics.

Thinking of also adding a way for people to up-vote/down-vote papers.

Would love to hear any thoughts/feedback! :D

Thanks!

Open articleView on HN

Appears to be built with https://github.com/vercel/vercel

Edit: that is an OSS client for use on their hosting service.

Yup!

This is incredible! Just a couple of things.

The first time I opened it the app worked fine. But now it just says “loading”. I’d debug more if I wasn’t on mobile.

RSS feed. Please add one. In fact if you could email me the top papers every week it would make a great newsletter.

Upvote/downvote is less useful than comments imo.

Let me know if you’d like to add search. That is my specialty.

That is all.

Ah I think the loading error might be due to many people loading the site at the same time. The site should work now!

Newsletter coming, and will definitely add filtering by tags/topics + search!

Comments sounds like a good idea as well!

I honestly have the complete opposite problem. Trying to get a feed from arxiv that doesn't include artificial intelligence whitepapers is a pain. Very little of the content is even remotely interesting or practical, and it sneaks in to almost every single topic/category.

lool

Love it! Was thinking of making this myself, you saved me the trouble!

yooo thanks! Got some exciting updates shipping super soon (e.g., search, email newsletter, auto-freshing a new list of papers on every page refresh, etc.)

Did you build this for a school project? I go to a college where this was submitted as a senior capstone and was wondering if this is the same project.

Nah, I just built this as a fun personal project.

The TL;DRs are generated or hand written? Are they powered by an LLM?

GPT-3! Thank god I don't have to hand write them lol.

How are you getting past the token limit?

Likely only summarizing the abstracts.

Yup only abstracts for now, but I think it should be possible to parse the paper PDF to extract just the approach + experiment results w/ GPT-3 (despite 4k token limit). Will add this soon.

I would suggest sentence tokenization followed by clustering to pull the most representative sentence from each cluster. Unless you want to go abstractive, that is.

If you run into issues with the token limit, there are some nice architectures for handling longer inputs (e.g., Longformer, Reformer, etc.).

interesting ideas, thanks!

Reply on news.ycombinator.com