Epoch AI: AI use in math papers passes 50% in three subfields
- Source
- x.com
- Date
Epoch AI counted how often established mathematicians acknowledge AI assistance in their arXiv papers, broken out across 18 subfields. Three of those subfields now have AI acknowledged in more than half of papers. The sharpest move is in differential geometry, which went from roughly 8% in July to 57% in September. The measure is narrow by design: it counts disclosure by authors with a track record, so it reflects routine professional use rather than experiments by newcomers. It also depends on authors choosing to disclose, which means the real share could be higher. The speed of the change is the signal. A field moving from rare to majority use within two months suggests that once a tool clears a usefulness threshold for a given kind of work, adoption follows quickly rather than gradually.
In 3 of the 18 math subfields we track, more than half of arXiv papers by established authors now acknowledge using AI. Rates have increased rapidly: in differential geometry, the share climbed from ~8% of papers in July to ~57% in September.
Context
Epoch AI reports a sharp increase in disclosed AI use within a selected set of mathematics papers. In three of the 18 subfields it tracks, more than half of arXiv papers by established authors acknowledged using AI. Its October 9 post puts differential geometry at roughly 8% in July and 57% in September.
The unit being counted is a paper containing an acknowledgment. That establishes how often authors disclose some use of AI in this tracked population, without specifying how much of the research AI performed. The figures count disclosure by authors, not AI's share of the mathematics.
Checking sign-in…
Loading comments…