awesome-diffusion-papers.vercel.app — time-sorted, searchable browser for awesome-diffusion-categorized

TL;DR

A small Claude-built companion website that ingests the Awesome Diffusion Categorized README and re-presents the same papers with chronological sorting and search. Built by Tucker as a personal tool to work around the upstream repo’s limitation of being organized by sub-area only (no dates, no time-sort, no full-text search across sections). Not a research artifact in itself — listed here so the wiki records the tool and the broader pattern of “ask Claude to wrap an awesome-list as a sortable site.”

Key claims

The site’s stated purpose, from Tucker’s Slack note, is to make awesome-diffusion-categorized “sorted by time” and searchable, because reading the upstream repo by git commits is painful [Slack source].
It was generated by asking Claude to “quickly create a website to sort and search through” the underlying awesome-list — i.e. a one-shot LLM-built tool rather than a maintained project [Slack source].
The hosted page (Vercel) serves a client-rendered SPA — at filing time fetch_url returned only the “Loading papers…” shell, with the actual paper list populated by JavaScript after page load [fetch attempt 2026-05-25].
Source of truth for the underlying data is still Awesome Diffusion Categorized; the site is a re-projection of that README, not an independent catalog [Slack source].

Method

Not described publicly — the working assumption from Tucker’s note is: scrape the awesome-diffusion-categorized README, parse each entry’s title / venue badge / arXiv link, extract date from arXiv ID (e.g. 2406.10210 → 2024-06), and render a sortable + searchable client-side table. No backend, no API, no per-paper synthesis. The chronological signal is derived from arXiv IDs, not from upstream commit dates.

Results

For day-to-day use the site is more navigable than the upstream README when the question is “what’s the most recent paper in sub-area X” or “show me everything in 2025-H2”. It does not add commentary, abstracts, or filtering beyond date and keyword, so for deep dives it still hands off to the underlying arxiv pages — same as the upstream awesome-list.

Why it’s interesting

The interesting thing isn’t the site itself, it’s the meta-pattern: community awesome-lists are a heavily-used research surface, but their organization is whatever the maintainer chose, which rarely matches a reader’s current need. Re-projecting one with a one-shot LLM-built UI is cheap and (from Tucker’s note) actually useful. This is adjacent to but distinct from Awesome World Models — the upstream list itself — and to ArxivQA: Training Retrieval Agents for arXiv Search, which does the same “make arXiv navigable” job at scale via retrieval agents instead of via a thin UI wrapper. Worth keeping the pattern in mind: if a curated index lands in #research-external and reading order matters, a 30-minute Claude wrap can be the right ergonomic answer.