Somebody should make a tool that takes a mathematical paper and autogenerates a "lemma flowchart" that depicts which lemmas etc. are direct inputs to which other results.
Somebody should make a tool that takes a mathematical paper and autogenerates a "lemma flowchart" that depicts which lemmas etc. are direct inputs to which other results.
It is not a DAG
😱
I'd nominate @bengolub.bsky.social as the special someone.
This likely goes without saying, but I just tried asking free ChatGPT to do this with a paper and it failed dismally.
Do you have access to the TeX or only the PDF?
Great question! This seems like it should be much easier to do with TeX. I'd already be happy if I could do with with access to the TeX (to use for my own papers), but I'd be extra happy to do it with access only to the PDF (as a reading/refereeing aid).
Do readers of arXiv get access to the TeX? I know the author uploads. If the TeX is cleanly structured (cross references to lemmas and proposition has clear \end), I’d write a shell script to do this.
Yeah, you can download source code on arXiv.
here you go lemmatree.streamlit.app tried it on a random paper i found on arxiv econ.th ; seems to work fine
Holy shit, thank you! Excited to try this out.
vibe-coded with the new gemini model with the following initial prompt and some API tinkering afterwards [it was using an old clunky package and made a few mistakes in the program].
This is slick Apoorva
thx; quite pleased with how it turned out. the fact that one can write and deploy a reasonably functional program over morning coffee is kinda nuts.
Totally. Vibe coding can do a lot of fun stuff now.
Can you tell me a bit about how you deployed this?
streamlit cloud deployment (free tier) docs.streamlit.io/deploy/strea... as long as you have app.py and requirements.txt in a git repo, it can create an isolated env and deploy it. there are other options [railway, vercel etc] but this is the simplest for small toy applications
this could easily be extended to accepts PDFs (given how good gemini is at OCR); github.com/googleapis/p... exercise left for the reader / user :)
@ EL you should extend this to pdfs and host it with an API key [yale would pay for it right?] as a public good
I think this exists! I'll try to dig it up. I didn't find it very useful when I played with it but I can see how it might be.
I often make myself a lemma DAG as a refereeing aid when I’m refereeing a paper with intricate proofs. This thought came up when I made one for coauthors, we decided to put it in the paper, and I thought “It’d be cool if this were standard/easy to put in every theory paper.”