OpenAI Math Results Repository Keeps Evolving
An update to OpenAI’s math results repository shows that this body of work is still undergoing active verification. OpenAI scientist Dan Roberts said the repository received six new Lean formalizations, 19 modifications, and three withdrawals; about 42% of the top-line results are now formalized.
What the Three Withdrawals Mean
The source context identifies the three withdrawn manuscripts as work on Weil classes, Kuga–Satake correspondences, and the rational Hodge conjecture for products of K3 surfaces. A withdrawal does not by itself disprove those research directions; it more narrowly means that the statements or proofs should no longer be presented as valid repository results in their current form.
The Value and Limits of Lean Formalization
Lean formalization translates mathematical definitions, assumptions, and deductions into structures that a proof assistant can check, making some errors easier to locate and revisions easier to track. It also creates a maintenance process resembling software release management: results can be added, changed, or withdrawn when problems are found.
However, roughly 42% coverage is not an overall correctness rate, nor does it show that the remaining results have been machine-checked. The key open questions are how the withdrawn manuscripts will be revised, which results may be formalized again, and whether continued repository updates will make the full mathematical arguments more auditable.