Made a Website to Help Read Papers: Cool Papers
To Start With
I've long had the habit of browsing Arxiv daily, trying to stay as up to date as possible with the latest results in the field, telling myself "if you're not moving forward, you're falling behind." Quite a few readers have asked me how I go about browsing Arxiv, and what tools I use to help — but honestly, for a long time I just browsed the Arxiv website directly, without any algorithmic filtering, going through papers one by one myself. The process is tedious, but not unbearable, and the reason I didn't use an algorithm for pre-screening was mainly the fear of missing papers: since the whole point of "browsing" is to catch the newest work, any algorithmic miss means "missing the boat."
Ever since Kimi Chat launched, I'd been planning to build a companion website that combines it with the process of browsing papers, to speed things up. I've had a bit more free time these past few weeks, so with help from GPT-4 and Kimi, I put together an initial version of this site, and after several days of testing and tuning it has gradually stabilized. So now I'm officially inviting readers to give it a try.
Cool Papers: https://papers.cool
In the Middle
As the name "Cool Papers" suggests, the site aims to turn browsing papers into a genuinely "cool," immersive experience. Of course, the current implementation is still fairly rough — the "cool" part mainly comes from having Kimi answer a few FAQs about each paper, which lets us understand the paper's main content more accurately and efficiently (compared to just reading the title and abstract), and thus decide whether it's worth reading closely.
A few things worth pointing out specifically:
1. This is a site for skimming/browsing papers, not a site for "reading" papers. "Browsing" here means "filtering" — the goal is to identify which papers deserve a close read, not to replace close reading;
2. Currently only Arxiv is supported as a paper source, syncing the latest day's paper list from Arxiv. So browsing Cool Papers is basically equivalent to browsing Arxiv. In the future I may add other sources, such as OpenReview, depending on usage and feedback;
3. Since the site is positioned around "browsing," the philosophy is "consistency matters, and there's no catching up on what's past" — so for now only the latest day's papers are shown, with no historical lookback. This could of course change based on reader feedback;
4. FAQ is powered by Kimi Chat, so please be grateful and use it wisely. In fact, titles and abstracts alone can already filter out a lot of papers; the Kimi FAQ exists to give a more precise judgment on papers you're unsure about. So please don't click "[Kimi]" carelessly;
5. Clicking "[PDF]" lets you preview the paper's content (desktop browsers only — mobile browsers will trigger a download instead), but this relies on your own network connection to reach Arxiv, so if the PDF preview takes forever to load, it might be a network issue on your end;
6. Clicking "[Copy]" copies the paper's basic info (title, abstract, link, etc.) to your clipboard, so you can paste it elsewhere to share the paper;
7. The paper list keeps Arxiv's default publication order by default; adding "-sorted-by-stars" sorts by stars computed from all users' click activity;
8. Paper updates: paper updates are synced directly from the official Arxiv site, normally with a delay of no more than 10 minutes. Arxiv typically updates around 10am Beijing time on weekdays, though this can vary by a few hours, and there are no updates on weekends or during some U.S. holidays. If you notice the site's papers aren't updating, check the official Arxiv site to confirm the update status.
Further notes on the Kimi FAQ:
1. After clicking "[Kimi]" for a given paper, you'll enter a queue;
2. If there are a lot of people in the queue, you'll see a "Pending: xxx" result, where "xxx" is the queue length, which updates automatically;
3. Once the queue finishes, it will show "Loading: xxx%," followed by a streaming output of the FAQ content;
4. You don't need to keep the page open during queuing and generation — even if you close the page, queuing and generation will continue in the background, and reopening the page and clicking the same "[Kimi]" will resume the previous progress;
5. If you click "[Kimi]" and the FAQ content appears instantly, that means this paper has already been read by another reader and the FAQ has been cached;
6. Because of the queuing mechanism, clicking "[Kimi]" carelessly won't put too much strain on the site, but it will make other people's wait times longer, which isn't a very considerate thing to do.
To Close
Finally, I welcome everyone's opinions and suggestions, and I'd love to hear more feature requests. It's clear that Cool Papers is still quite rough at the moment, far from as "cool" as imagined, and using Kimi for the FAQ is only a very basic way of combining Kimi with papers. Kimi's extremely long context surely has much greater potential still to be explored. So I'm very much looking forward to everyone's creativity in finding an even better way to combine Kimi with the process of browsing papers.
Translated automatically with claude-sonnet-5; all equations are reproduced verbatim from the source. Copyright remains with the original author.