Resolution criteria: - "Yes": Wins if, strictly after 2026-10-08T12:49:05.783Z and at or before 2027-01-01T00:00:00.000Z, a cumulative total of at least 10 qualifying Erdős problems have been resolved by AI models, where each qualifying resolution is recorded in a publicly available mathematical source (including arXiv preprints, formally verified proofs in the Google DeepMind formal-conjectures repository, or peer-reviewed journals) as newly solved, and the solution is credited — in whole or in substantial part — to an AI model's autonomous generation of the key mathematical insight or counterexample (not merely formalization, verification, or literature retrieval of a human-discovered argument). - "No": Wins if, strictly after 2026-10-08T12:49:05.783Z and at or before 2027-01-01T00:00:00.000Z, the cumulative total of qualifying AI-resolved Erdős problems within the window remains fewer than 10. Clarifications: - "Previously unsolved Erdős problem": a problem listed in the Google DeepMind formal-conjectures repository (problems.yaml) or on erdosproblems.com (using its status as captured on or around 2026-10-08T12:49:05.783Z, e.g. via an Internet Archive snapshot) that held an open/unsolved status as of that date. - "Resolved by AI": the solution must be a genuinely novel mathematical result — not merely the retrieval of a pre-existing solution from the literature or the formalization of a human-discovered proof — and must be attributed, in whole or in substantial part, to an AI model's autonomous generation of the key mathematical idea, construction, or counterexample. Solutions that AI found exclusively via literature search do not qualify. - The 10-problem threshold is cumulative across all AI systems and teams; individual problems resolved multiple times count only once.

• https://github.com/google-deepmind/formal-conjectures
• https://arxiv.org/list/math/recent
• https://www.quantamagazine.org
• https://terrytao.wordpress.com
• https://erdosproblems.com

Jan 1, 2027, 12:00 AM UTC