About
AI research is written for experts. Marginalia explains the work that matters in plain language, and lets you annotate it for the next reader.
The name is a nod to marginalia: the notes people scribble in the margins to make a difficult text their own. That's what we do, in public, for AI.
Each morning a pipeline pulls fresh papers from arXiv, drops anything we've covered, and drafts a plain-language explainer with Cloudflare's Workers AI. Then a person reviews every draft against the source before it publishes. The AI does the finding and drafting; humans own the quality bar.
We only ever publish original summaries, and we never republish a paper's abstract word for word. We credit and link the authors, the arXiv page, and the PDF for everything we cover. Wherever AI was involved, we say so. And nothing goes live until a person has checked the draft against the source, which is how the editorial queue catches anything the model gets wrong. The bar we hold ourselves to is simple: it should all hold up when a researcher clicks through to the paper.
Marginalia was founded by Alex Dong and runs as four programs, each led by a student director who does the work and can answer for it.
Owns the daily publication: the cadence, the review queue, the quality bar.
Owns the curriculum and the lessons, and runs the workshops.
Mentors the project teams and review authors, and holds the real-venues-only line.
Supports the chapter leads, runs the campaigns, owns the newsletter.
A title here means real involvement, never a name on a chart. Want to write explainers, build features, lead a program, or start a chapter? Come say hi.