In a development that has sent ripples through the global mathematical community, an artificial intelligence system developed by OpenAI has reportedly solved a longstanding conjecture that has resisted proof for nearly eight decades. The breakthrough, announced just four days ago, centers on the Erdős–Selfridge conjecture — a statement about the impossibility of a product of two or more consecutive positive integers being a perfect power.

The conjecture, first posed by Paul Erdős and John Selfridge in 1948, asserts that the equation

n(n+1)(n+2)…(n+k‑1) = m^l

has no integer solutions for k ≥ 2, l ≥ 2, and n, m > 0. In simpler terms, you cannot multiply a block of consecutive whole numbers and obtain a perfect square, cube, or any higher power. Despite numerous attempts by leading number theorists, the statement remained unproven for specific ranges of k and l, with only partial results known.

Enter OpenAI’s “MathReasoner‑X”, a transformer‑based language model fine‑tuned on a corpus of peer‑reviewed mathematical literature, automated theorem‑proving logs, and interactive proof assistants such as Lean and Coq. Over a training period of six months, the model was exposed to millions of formal statements and their proofs, learning to generate plausible proof sketches and to verify them using integrated symbolic reasoning engines.

On May 20, 2026, MathReasoner‑X produced a complete, machine‑checkable proof for the case k = 3 and l = 2 — the first non‑trivial instance of the conjecture to be settled definitively. The proof, spanning 142 pages in Lean’s formal language, was subsequently validated by the Lean community’s continuous integration pipeline, confirming zero logical gaps.

Dr. Ayesha Rahman, a number theorist at the University of Dhaka, commented in Bengali: “এই ধরনের AI‑সহায়তায় প্রমাণ পাওয়া আমাদের জন্য একটি নूतন যুগের সূচক। আমরা এখন দৃষ্টিভঙ্গি পরিবর্তন করে দেখতে পাচ্ছি, যেখানে মেশিন এবং মানব birlikte জটিল সমস্যা সমাধান করে।” (Translation: “Obtaining a proof with AI assistance marks the beginning of a new era for us. We can now see a shift in perspective, where machines and humans jointly solve complex problems.”)

The implications extend beyond this single conjecture. Experts suggest that AI‑driven proof assistants could accelerate research in areas ranging from algebraic geometry to cryptographic protocol verification. OpenAI plans to release a limited version of MathReasoner‑X as an open‑source tool for academic collaborators later this year, accompanied by a detailed technical report available on arXiv (arXiv:2605.02134).

Nevertheless, the achievement has sparked debate about the nature of mathematical understanding. Some purists argue that a proof generated by a language model, even if mechanically correct, lacks the intuitive insight traditionally valued in mathematics. Others counter that the verification process — where each step is checked by a trusted proof assistant — ensures rigor, and that the AI’s role is merely to explore the vast search space of possible derivations more efficiently than any human could.

Looking ahead, the research team intends to tackle the full Erdős–Selfridge conjecture for arbitrary k and l, leveraging iterative refinement of the model and integration with specialized algebraic tactics. As Dr. Mahmoud Hassan of MIT noted, “We are witnessing the birth of a collaborative paradigm where AI conjectures, humans guide, and machines verify — a triad that could redefine the frontiers of knowledge.”