OpenAI just dropped 700 preprints of mathematical proofs and counterexamples — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI published roughly 700 mathematical preprints to a public GitHub repository named "math," organized in a "preprints" subdirectory with each entry housed in its own dated folder
- Filenames follow a consistent "A-[description]-[date]" naming convention, with commit dates concentrated in September and October 2026
- A large share of the listed preprints are explicit counterexamples to long-standing conjectures, including the Entropy Conjecture, Kaplansky's Direct Finiteness Conjecture (in both characteristic two and odd characteristic), Ryser's Covering Conjecture, Kurosh's Division Ring Problem, the Group Ring Determinant Conjecture, Hadwiger's Conjecture, and the Infinite Matroid Packing-Covering Conjecture
- The collection also contains constructive proofs and new results, such as a "Cyclic Polytabloid Proof of Saxl's Conjecture," a "Modulus Proof of Cannon's Conjecture," a "Fock-Space Inequality and the Laughlin Spectral Gap," and a "Gepner Stability Condition on Every Smooth Quintic Threefold"
- The preprints cover a wide disciplinary spread — combinatorics, algebraic geometry, differential geometry, topology, number theory, quantum query complexity, and mathematical physics — each as its own discrete, individually citable document
Why it matters: Dropping around 700 distinct mathematical results in a single public repository — hosted on GitHub where any work can be downloaded, audited, and cited directly — is a volume and breadth of mathematical output that meaningfully alters the surface area available to working mathematicians. The disproportionate share of conjecture-disproving counterexamples, visible right in the filenames, gives the collection a distinctive character: it reads less like a standard paper series and more like an aggressive conjecture-clearing exercise.
Ask SkimNews
