Is there an LLM for condensing and surfacing recent research?

6 pointsby timlod3 yıl önce2 comments

I've been using ChatGPT for a while now, and find it (among other things) great to get a pointer at some seminal papers (usually older) which to seed my search with.

Something I yearn for is a way to quickly learn about recent advances in some more niche topics I'm not already too familiar with, mostly to quickly advance/curb some ideas. I feel like it should be possible to make an LLM aware of (via online fine-tuning or perhaps summarized context of articles which match via some other method) new papers on Arxiv or at least abstracts collected at journals. Essentially something that, given a prompt asking for recent advances in a certain field, spits out papers that may be useful to check, ideally with a quick summary relating to the prompt (could go beyond the abstract, pulling out an idea from the main body).

Right now my workflow for this problem is mostly finding some sort of seed paper, and juggling abstracts on something like connectedpapers.

Does anyone know what the state of the art for this is, and whether something useful already exists or is being worked on in the space?

2 comments

sargstuff3 yıl önce

https://cobusgreyling.medium.com/implement-custom-search-in-...

https://blog.rfox.eu/en/Programming/How_to_run_your_own_LLM_...

https://stackoverflow.com/questions/30904755/extract-abstrac...

https://generativeai.pub/what-the-heck-are-agents-10b2a99542...

https://simonwillison.net/2023/Aug/27/wordcamp-llms/

weallfeelit3 yıl önce

Chris Olah's Distill intended to do this: "a new open science journal and ecosystem supporting human understanding of machine learning. Distill is an independent organization, dedicated to fostering a new segment of the research community." The journal went to sleep in 2021, unfortunately. (https://blog.research.google/2017/03/distill-supporting-clar...)