/coglinker

LLM-powered Q/A over arXiv preprints

Primary LanguagePythonMIT LicenseMIT

Coglinker

What if you could ask an arXiv preprint questions directly?

This is a (really) early set of experiments on how LLMs can be used to build tools that help people understand research papers faster (as opposed to just directly reading a PDF).

Demo

Coglinker Demo

Here's a Colab version of this you can have up and running in ~2 minutes: Colab

How it works

In its current implementation, coglinker:

  1. Directly downloads the source LaTeX for a requested arXiv paper.
  2. Parses the LaTeX, and processes it into chunks.
  3. Given a question, retrieves relevant "chunks" and adds it to an LLM's (current ChatGPT) context.
  4. Generates an answer to the question using the LLM.

Open Questions

  • How do we build tooling that helps people maximize their transfer of learning when they read (and engage with) research?