We operate a Twitter bot, @ArXivGPT, which tweets summaries of academic papers to facilitate quick access to trending papers on Twitter.

What it does

Our bot can answer questions posted in the thread under its tweets. Users can ask questions related to the academic papers, and the bot provides concise, intelligible answers.

How we built it

We implemented a daily cronjob that gathers all comments from the previous 24 hours. The bot then converts the associated paper to markdown format using OCR. The markdown file, along with the user's question, is fed into the new 100k context window Anthropic model.

Challenges we ran into

We needed to invest in the $100 Twitter API to read tweets, which was an unexpected cost🤯.

Accomplishments that we're proud of

We're extremely proud of the fully functioning system we've developed.

What we learned

We've learned the advantage of using Python classes over functions for such applications.

What's next for arXivGPT

Our future plan is to improve our bot to display only trending papers. We plan to do this by counting the frequency of a paper's appearance on Twitter, among other metrics.

Try it!

  1. Go to
  2. Go to a paper where you want to ask something
  3. Write your question but put before you start writing the arxiv link from the tweet like: " What is the Hasofer-Lind reliability index problem"

Built With

Share this project: