Will at least 10 more currently unsolved Erdős problems be resolved by AI in 2026?
Delphi’s forecast
- Yesyour pick84.3%$0.84
- No15.8%$0.16
We could not pin down how this would settle, so there is no market to trade. The odds stand.
Forecast reasoning
OpenAI's October 2026 release documented 17 complete AI solutions to previously unsolved Erdős problems, with 12 verified via Lean proof, exceeding the 10-problem threshold by a significant margin. DeepMind's AlphaProof Nexus independently solved 9 problems, and earlier breakthroughs in May-June 2026 added multiple confirmed resolutions, establishing a clear pattern of AI capability on frontier mathematics.
Settlement
Resolution criteria: - "Yes": Wins if, strictly after 2026-01-01T00:00:00.000Z and at or before 2027-01-01T00:00:00.000Z, at least 10 distinct Erdős problems that were listed as unsolved/open at the start of that window are marked as solved on erdosproblems.com and the credited resolution is attributed to an AI system (i.e., an AI model or AI-assisted proof is identified as the primary driver of the solution in the problem's record or in the peer-reviewed or formally verified record supporting it). - "No": Wins if, as of 2027-01-01T00:00:00.000Z, fewer than 10 distinct Erdős problems meet the above conditions. Clarifications: - A qualifying resolution requires: (1) the problem was listed as open on erdosproblems.com as of 2026-01-01T00:00:00.000Z; (2) it is marked solved or closed on erdosproblems.com at or before 2027-01-01T00:00:00.000Z; and (3) the solution is credited to or driven by an AI system, as established by a peer-reviewed publication, formal verification, or an official announcement from a recognized research institution published within the window. - Each problem is counted at most once regardless of how many AI systems independently resolve it.
Jan 1, 2027, 12:00 AM UTC
