-
An arxivgen.com paper on Gradient Sharding.
Inspiration
Successfully generating new knowledge is the holy grail of AI research. This will not only allow AI systems to self improve - it will allow them to revolutionize industries like healthcare and biotechnology, financial services, and machine learning research itself.
What it does
Given a new research proposal, this model does a vast research survey and looks to write the methods that execute that new research proposal. It simulates the experiments run with those methods, and generates the rest of the research paper.
How we built it
The core of the research generator is a chain of prompts backed by a research datastore. We first generate paper scaffolding and then use it to write a NeurIPS caliber paper on the given subject. arxivgen.com is backed by a flask server with Jinja templating for laying out the papers.
Challenges we ran into
Figure generation is very challengin.
Accomplishments that we're proud of
Latex generation (first of its kind?)
What we learned
How to ship papers
What's next for ArxivGen
End-to-end generation.
Log in or sign up for Devpost to join the conversation.