Kaili Cao 曹开力

AI in Science · MapHub

MapHub

Daily relevance scores of new arXiv preprints for a research community, open and checkable.

In brief

MapHub logo: a three-by-three grid of squares in shades of blue, with a yellow square at the center, like a small heatmap.

MapHub, the MIDAS arXiv Preprint Hub, is the first deliverable of my Schmidt AI in Science Fellowship at MIDAS: a carpentry of the fellowship, which I lead with teammates Kyle Heyblom, Eva Albalghiti and Dimitrios Simatos. I built its scorer and viewer with AI assistance (Claude). Each day since October 2026, MapHub scores the new arXiv preprints in astrophysics and AI for their relevance to each participant, so that participants see which papers matter most to them and who is most likely to answer questions about a paper. A large language model assigns the scores against a fixed rubric, and everything else is open: the rubric, the code and the daily tables, with every score labeled as AI-generated. The model's reasoning may be opaque, but what MapHub delivers can be checked. Anyone can run a MapHub of their own for other arXiv listings, following the instructions.

What comes next

MapHub is still under development. The same scoring is working back through earlier listings, with arXiv's first listing, in 1991, as the goal, so that participants can go through a back catalog of the papers that matter to them. In MapHub's tables so far (listings from June 30 to October 16, 2026), about 7% of a day's new astro-ph papers score 30 or more out of 100 for each participant, on average: a handful worth reading out of some ninety. More participants are welcome, and a department could use its heatmap for a daily discussion of new papers across subfields: papers that touch several participants make natural topics. The live heatmaps are in the viewer, and the code, rubric and tables are on GitHub.

Terms on this page (3)
arXiv
A free online archive where physicists and astronomers post papers ("preprints"), usually before or while they are reviewed by a journal. Astronomy papers are in the section astro-ph.
MIDAS
The Michigan Institute for Data and AI in Society, the University of Michigan's institute for data science and AI, which hosts the Schmidt AI in Science postdoctoral fellows there.
Carpentry
In the Schmidt AI in Science fellowship, a small project that uses AI. MapHub is one.

Full glossary