🔥 arXivSub is officially online! It's been almost three months from the end of July to now. Thank you to everyone who participated in the testing and gave me valuable feedback. The original intention was to see that many friends needed daily delivery of arXiv papers, wanting to be the first to discover papers related to their field. Thinking that I also have the need to read papers to find ideas, I decided to build a website. In the end, even if no one uses it, I would consider it a way to improve my full-stack development skills. After clarifying the requirements, I opened Notion and started designing the core features: paper crawling, AI summarization, sending emails, etc., from user interaction to specific implementation for a general plan. Once I had a clear idea, I started building. With the help of Cursor, it took about a week to go from the birth of the idea, UI design, front-end, back-end, user system, database, server, domain name, email system, payment system, and various third-party services, and I implemented them all. There were some pitfalls along the way, such as the payment issue which was stuck for a long time before being resolved, and optimizing search using B-tree indexes and generalized inverted indexes. Overall, it was a complete product development process from 0 to 1. There is still a long way to go from 1 to 100, and there is still a lot to learn. Now, arXivSub has at least 80,000, if not 100,000 papers (including 7 conferences like CVPR), and on average, more than 500 new papers are added every day, all including AI-generated Chinese/English summaries: main content, innovations, technical methods, datasets, comparative analysis, and limitations. Nowadays, my workflow has become "when in doubt, ask AI first." For example, if I need to find a specific detail in an article, I just throw it to the AI, and it only takes a few seconds, saving me from having to go through it myself. Moreover, many articles are not worth spending precious hours of your life reading completely. You only need to read the AI summary to get a general impression. If you find it helpful, you can then give it to the AI for a detailed interpretation or open the PDF to read a specific paragraph. This is the positioning of arXivSub. It mainly solves the first step of getting an "impression." You can directly use keywords to search for papers you are interested in from major AI conferences over the past three years, for example, Graph Neural Network, which yields a total of 105 papers, with special notations for award-winners like "oral." Then you can filter the papers you are interested in based on the Chinese/English AI summaries. After finding one, you move on to the second step, "detailed reading." You can directly click the ChatGPT button in the upper right corner to have the AI give you a detailed interpretation and ask more specific questions, or open the PDF to look at the paper's structure, experimental results, and other details. You can also click the Google Scholar button to see how many citations this paper already has. If you are only interested in new arXiv paper delivery, you can also go to the subscription section to set keywords, and you will receive daily emails about arXiv papers related to that set of keywords; this is included in the free features. You can also turn off the emails at any time if you no longer want to receive them. If this tool can help >0 people, then for me, going through the entire development process is much more valuable than just a technical exercise.

arXivSub

AI-assisted research paper searching tool

AI-assisted research paper searching tool