PaperDemo — interactive supplements that publish alongside your arXiv paper
A publishing tool that lets authors attach one or more LLM-generated interactive widgets to their arXiv or blog post, with version pinning so reviewers see the same demo you submitted.
academic researchers and graduate students publishing technical papers
- Submit a theorem statement and get a draft visualization plus counterexample explorer
- arXiv-compatible sidebar embed that does not require reader accounts
- Reviewer-locked preview links so the same demo is reproducible across revisions
- Citation-ready widgets that count as 'supplemental material' in CVs and grant reports
Tao explicitly said he plans to ship interactive visualizations as supplements for future papers — a hosted 'paper-demo' service tied to arXiv would standardize what Tao just invented by hand.
Terry Tao explicitly stated intent to attach LLM-built interactive visualizations as paper supplements, validating the use case at the top of academia; broader demand exists across the ~2.4M arXiv corpus but currently concentrated in ML/data-viz communities using ad-hoc Jupyter/Colab.Old and new apps, via modern coding agents | What's new ↗arXiv.org e-Print archive ↗
No product standardizes 'interactive supplements tied to arXiv submissions with version pinning'; adjacent platforms (Hugging Face Spaces, Papers With Code, Observable) serve overlapping but distinct needs (model hosting, reproducibility, notebooks) without the arXiv-anchored widget framing.Paper page - Computational reproducibility of Jupyter notebooks from ... ↗Pricing - Hugging Face ↗
Academic researchers rarely pay out-of-pocket; competing free tiers from HF ($0 public) and existing free notebook hosting make paid conversion hard. Likely needs institutional/grant-funded B2B or grant-piggyback model.Hugging Face Pricing 2026: Free, PRO & Team Plans from $9/mo ↗
Interactive/reproducible-paper trend is durable and well-established (Distill.pub ~2017, Papers With Code ~2018, Jupyter reproducibility studies); LLM-generated widgets are a natural next step rather than a fad.Understanding and improving the quality and reproducibility of Jupyter ... ↗
Buildable on existing widget runtimes (Streamlit, Gradio, Observable) with arXiv ID + git hash pinning; main work is the authoring UX and LLM-code-sandboxing safety layer, not the hosting infra.Old and new apps, via modern coding agents | What's new ↗