ChatGPT solves decade-long dice problem in 13 minutes — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- ChatGPT produced a valid six-dice solution after roughly 13 minutes of 'thinking' and two clarifying prompts, solving a generalization of a problem that consumed mathematicians for more than a decade.
- Eric Harshbarger at Auburn University, one of the original dice researchers, verified the journalist's numbers were correct but called the solution inelegant — five 720-sided dice and one 20-sided die — and noted researchers already have a better six-dice design using all 360-sided dice.
- Paul Meyer, a Google software engineer collaborating on the dice project, questioned whether the AI truly generated the solution or located pieces online; a search found no trace, and the best explanation is that the model combined published work on the researchers' tracking site with a known 'induction' method for extending solutions to more dice.
- Sébastien Bubeck of OpenAI said he expects the same magnitude of AI mathematical progress over the next six months as the prior six, and predicted AI will eventually create new mathematical concepts rather than only solving or counter-exampling existing ones.
- Bubeck acknowledged the disruption directly, saying human mathematicians 'may be forgiven for a period of grieving' as the joy of manual discovery loses its special status and questions of credit for AI-derived results move to the center of the field.
- The journalist described feeling ambivalent — relief at the result but unease that a self-described layperson with 'dusty mathematics' could chip away at problems once reserved for sustained human research.
Why it matters: A non-mathematician produced a research-verified solution to a problem that consumed a decade of expert effort, and OpenAI's Sébastien Bubeck says comparable progress will arrive in the next six months — meaning the bar for contributing to certain kinds of mathematical problems is collapsing for non-experts while the field's norms around credit and authorship are about to be tested.
Ask SkimNews
