AI chatbot solves decade-long math problem in 13 minutes — SkimNews
Get the Health newsletter
Daily health & science — research, biotech, public health, the studies worth knowing. Free.
- ChatGPT solved a six-player non-transitive dice problem in roughly 13 minutes of "thinking," producing five 720-sided dice and one 20-sided die after the journalist described the puzzle in plain English and nudged twice.
- Eric Harshbarger at Auburn University verified the AI's numbers as correct, though he noted the solution was less elegant than a six-dice set using all 360-sided faces that he and colleagues had already derived.
- Paul Meyer, a Google software engineer who collaborates on the dice research, initially suspected the model had found the answer online, but searches found no trace — the best explanation is that ChatGPT pulled from the researchers' public tracking website and combined the published five-dice solution with a known inductive method.
- Sébastien Bubeck at OpenAI predicted the dramatic AI math progress of the past six months will repeat in the next six months, saying researchers can "raise our ambition" in mathematics and science generally.
- Bubeck acknowledged current AI models can find counterexamples and solutions to existing problems but cannot yet create new mathematical concepts or fields — a capability he considers "clearly going to happen" though the timeline is uncertain.
- Harshbarger said he would be "thrilled" by any solution regardless of how it was found, while the journalist confessed to embarrassment at solving the problem with no mathematical skill and unease about diminishing human input in the field.
- Meyer said he "almost constantly" feels impressed by recent AI capability gains and personally believes "there's some intelligence there," though he declined to overstate the model's originality given that the building blocks were publicly available.
Why it matters: A non-mathematician solving a niche combinatorial problem in 13 minutes via plain English suggests AI is collapsing the skill barrier for specialized mathematical puzzles, with OpenAI's Bubeck forecasting that comparable progress will repeat every six months — forcing mathematicians to confront credit and attribution as their decades-long labors are increasingly matched or surpassed by chatbots in minutes.
Ask SkimNews
