OpenAI’s experimental mannequin achieved gold on the Worldwide Math Olympiad

OpenAI has achieved “gold medal-level efficiency” on the Worldwide Math Olympiad, notching one other necessary milestone for AI’s fast-paced progress. Alexander Wei, a analysis scientist at OpenAI engaged on LLMs and reasoning, posted on X that an experimental analysis mannequin delivered on this “longstanding grand problem in AI.”

Based on Wei, an unreleased mannequin from OpenAI was in a position to resolve 5 out of six issues at one of many world’s longest-standing and prestigious math competitions, incomes 35 out of 42 factors complete. The Worldwide Math Olympiad (IMO) sees international locations ship as much as six college students to resolve extraordinarily troublesome algebra and pre-calculus issues. These workouts are seemingly easy however often require some creativity to attain the best marks on every downside. For this year’s competition, solely 67 of the 630 complete contestants acquired gold medals, or roughly 10 p.c.

AI is commonly tasked with tackling advanced datasets and repetitive actions, but it surely often falls brief in terms of fixing issues that require extra creativity or advanced decision-making. Nevertheless, with the newest IMO competitors, OpenAI says its mannequin was in a position to deal with difficult math issues with human-like reasoning.

“By doing so, we have obtained a mannequin that may craft intricate, watertight arguments on the degree of human mathematicians,” Wei wrote on X. Wei and Sam Altman, CEO of OpenAI, each added that the corporate does not anticipate to launch something with this degree of math functionality for a number of months. Meaning the upcoming GPT-5 will possible be an enchancment from its predecessor, but it surely will not function that very same spectacular functionality to compete within the IMO.