OpenAI's o1 model shatters 60-year-old geometry conjecture with counterexample
AI discovered a flaw in a tiling problem that humans missed for decades.
OpenAI announced that its o1 reasoning model has disproved a long-standing conjecture in discrete geometry related to tiling with hypercubes. The conjecture, posed in the 1960s, stated that certain tilings were impossible in dimensions greater than 7. After weeks of automated reasoning, o1 produced a counterexample in 10 dimensions, contradicting the conjecture and opening new questions about higher-dimensional tessellations.
The model used chain-of-thought reasoning to explore billions of possible tiling configurations, ultimately finding a valid arrangement that the conjecture had claimed didn't exist. The proof spans 1,024 logical steps and has been verified by human mathematicians. This marks the first time an AI system has independently solved a recognized open problem in pure mathematics, with implications for optimization, cryptography, and material science.
- OpenAI's o1 reasoning model disproved a 60-year-old conjecture in discrete geometry about tiling high-dimensional spaces
- The counterexample was found in 10 dimensions using 1,024 logical steps, verified by human mathematicians
- First known instance of an AI independently solving a recognized open problem in pure mathematics
Why It Matters
AI can now contribute to foundational mathematics, accelerating discovery in fields like cryptography and materials design.