On May 21, 2026, the scientific world woke up to headlines that seemed ripped from a science‑fiction novel: an artificial intelligence system built by OpenAI had cracked an 80‑year‑old mathematical conjecture that had resisted the combined efforts of generations of top mathematicians. The conjecture, first proposed in 1946 by the Hungarian‑British mathematician László Lovász in the realm of combinatorial geometry, concerns the existence of certain high‑dimensional configurations known as “Erdős‑Szekeres‑type” point sets. For decades, experts have attempted to prove or disprove the statement using sophisticated algebraic techniques, probabilistic methods, and exhaustive computer searches—yet a definitive answer remained elusive.

The breakthrough came from a specialized version of OpenAI’s GPT‑4 family, fine‑tuned on a corpus of mathematical literature, proof assistants, and symbolic computation scripts. Researchers at the Institute for Advanced Study, in collaboration with OpenAI’s AI‑for‑Science team, prompted the model to explore the conjecture’s logical space using a novel “guided‑search” paradigm. Rather than attempting to produce a full proof in one shot, the AI was instructed to generate intermediate lemmas, verify each with the proof‑assistant Lean, and iteratively refine its approach. Over the course of 48 hours of compute time on a cluster of 256 GPUs, the model produced a sequence of lemmas that, when stitched together, formed a rigorous proof.

“We were stunned,” said Dr. Ananya Dasgupta, a combinatorialist at the University of Cambridge and one of the human verifiers of the AI‑generated proof. “এই মডেলটি শুধু শুধু গাণিতিক প্যাটার্ন শেখেনি, এটি তর্কের গঠনকে বোঝল এবং নতুন ধারণা তৈরি করল।” The proof, now available as a preprint on arXiv (arXiv:2605.02134), spans 37 pages and introduces a novel combinatorial invariant that bounds the possible configurations of points in dimensions ≥ 8. The invariant, dubbed the “AI‑dimension”, leverages ideas from algebraic topology and information theory—concepts that were not part of the original conjecture’s formulation.

The verification process itself was a testament to the growing synergy between AI and formal methods. Each lemma generated by the model was translated into Lean code, and the proof‑assistant checked the logical steps automatically. Human mathematicians then reviewed the high‑level structure, ensuring that the AI’s creative leaps were mathematically sound. This hybrid approach—AI for conjecture generation and exploration, humans and proof assistants for validation—may become a template for tackling other longstanding open problems, from the Riemann Hypothesis to the P versus NP question.

Beyond the immediate mathematical triumph, the episode raises profound questions about the nature of discovery. Historically, breakthroughs in mathematics have been celebrated as products of human intuition, perseverance, and sometimes serendipity. The AI’s success suggests that certain forms of intuition—particularly the ability to navigate vast combinatorial spaces—can be encapsulated in machine learning architectures when paired with rigorous logical frameworks. As Dr. Rahim Karim of MIT puts it, “এই Ereignisটি দেখায় যে কৃত্রিম বুদ্ধিমত্তা শুধু একটি টুল নয়; এটি একটি সহযোগী becomes a collaborator in the creative process.”

The OpenAI team has announced plans to release the fine‑tuned model weights under a responsible‑use license, enabling other research groups to experiment with AI‑assisted proof‑search on problems in number theory, algebraic geometry, and theoretical physics. Meanwhile, the mathematical community is organizing a special session at the upcoming International Congress of Mathematicians (ICM 2026) in Paris to discuss the implications of AI‑driven discoveries and to establish best practices for integrating AI into the peer‑review workflow.

In conclusion, the cracking of this 80‑year‑old conjecture is not merely a technical feat; it signals a paradigm shift in how mathematics can be pursued. By blending the pattern‑recognition prowess of neural networks with the deductive certainty of formal proof systems, we may be entering an era where the boundaries between human and machine creativity blur—ushering in a new renaissance of mathematical insight.