Returning to Toronto from a summit on the future of mathematics at Open AI, Daniel Litt reflects on a workshop where Sebastian Bubeck and Jacob Tsimerman advised that the premise was AI will become robustly superhuman at mathematics, yet progress may stall. He notes the goal was not to predict but to imagine a future where existing trends continue.

The number of combinatorics papers on arXiv has surged since late 2021, and a similar rise appears across other fields, suggesting a massive explosion of output. This trend has continued through 2026, while MathOverflow questions and answers have been in slow decline, especially after 2025, likely driven by AI, with fewer questions and answers overall, and little increase in answers to older problems.

In practice this means that a huge amount of duplicative labor, both in flesh and in silico, is being devoted to work whose marginal value to mathematics is, essentially, the cost of the tokens and perhaps a few bits of information indicating that the problem can be solved by existing models. 2027 Of course this work might have value to the people announcing it (credit, PR, etc.).

Bright spots include cheap autoformalization and error repair, but skepticism remains about model capability, and the profession must adapt or risk unsustainable practices. The long‑term health of mathematics will depend on rethinking reward structures and preserving deep human engagement.