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
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:
- Directly downloads the source LaTeX for a requested arXiv paper.
- Parses the LaTeX, and processes it into chunks.
- Given a question, retrieves relevant "chunks" and adds it to an LLM's (current ChatGPT) context.
- 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?