Skip to content

Latest commit

 

History

History
14 lines (10 loc) · 724 Bytes

plan.md

File metadata and controls

14 lines (10 loc) · 724 Bytes

So,

How this will work:

User queries academic-gpt:

  • I get GPT to turn the query into a search query to use on ARXIV
  • Get top results from ARXIV (search by relevance)
  • Rerank document titles with the query (top 20 ish docs)
  • Top 3 Documents, insert them into the prompt. (Process them concurrently, cut out references, include appendix)
  • Get the prompt to mark whenever it uses a quote from a document (like a reference [1], and somehow link it to a specific section of the paper)

Build a cool looking frontend. Get the thing deployed. Write up documentation on how I built it (blog post style)

instead of just arxiv, also get a collection of top ML/AI papers. Make it query other top stuff like conferences, etc.