Ask HN: Bots for summarizing and searching complex scientific papers?
Are there any language models and bots available yet that can take a complex and multilayered piece of technical work, such as a scientific paper, and summarize its arguments? Including perhaps how the paper fits into the landscape of technical papers (e.g. the whole arxiv.org)?
Could it write a technical abstract or accompanying summary for a general audience based on the main body text, e.g. Methods, Results and the context or significance of the work?
12 comments
[ 5.1 ms ] story [ 34.2 ms ] threadAt least another alternative pops up on my twitter Ads
It's just 15 lines to do document summarization and you can ask it to summarize arguments by customizing the summarisation prompt.
You could bring in other papers context with other tech, but indexing it all would be the tricky bit.
There's also something called GROBID which extracts scientific PDFs much better than generic PDF readers. it's worth spinning up a docker instance of it and getting the text that way for LangChain if you have time.
https://www.explainpaper.com
https://elicit.org
https://scite.ai
And it wasn't already in my list, though https://typeset.io/ posted in this thread by Xeophon also looks neat
https://consensus.app/search/ Ask a question, get conclusions from research papers
https://www.semanticscholar.org AI-powered research tool for scientific literature
https://app.humata.ai PDF AI document assistant
https://summit-lablab.groundedai.company Search, summarise and visualise research topics
https://searchthearxiv.com arXiv link to find similar papers or use natural language
https://www.semion.io search arXiv preprints
I think it's more interesting to find bots that can search across different papers to summarize the state of art in some question
I think they also had a website up for a while.
https://www.nature.com/articles/d41586-020-03277-2