OpenAI just won gold at the world's most prestigious math competition. Here's why that's a big deal.

In a groundbreaking achievement, OpenAI's latest experimental reasoning model has secured gold medal-level performance at the International Math Olympiad (IMO), one of the world's most revered mathematics competitions. Alexander Wei, a technical staff member at OpenAI, proudly announced this milestone on social media, highlighting that this accomplishment addresses a long-standing challenge in artificial intelligence. The IMO, which began in 1959 in Romania, is renowned for its difficulty. The competition spans two days, with participants tackling a rigorous four-and-a-half-hour exam that poses three complex problems. Notable past winners include Grigori Perelman, known for his contributions to geometry, and Terence Tao, a Fields Medal laureate considered one of the foremost mathematicians alive today. In a recent podcast, Tao expressed skepticism about AI's prospects in the IMO, suggesting that researchers might want to set their sights lower than this elite competition. Despite these doubts, OpenAI's model successfully solved five out of the six problems presented, operating under the same conditions as human competitors. Noam Brown, a colleague of Wei, commented on the model's extraordinary endurance during the exam, noting that IMO problems require a level of sustained creative thought that surpasses previous benchmarks. He stated, "This model thinks for a long time," emphasizing its advanced capabilities. Wei characterized the model as a significant upgrade in general intelligence, claiming it is "breaking new ground in general-purpose reinforcement learning." In contrast to DeepMind's AlphaGeometry, which is specifically tailored for mathematical tasks, OpenAI's model represents a broader pursuit of general intelligence. OpenAI's CEO, Sam Altman, also reflected on the model's accomplishment, stating that when the organization was founded, such a feat seemed like a distant dream. However, he acknowledged that a model showcasing this level of capability will not be accessible to the public for several months. This achievement illustrates the rapid advancements in AI technology. Just a year ago, AI labs were assessing models based on elementary school math. Tech entrepreneur Peter Thiel previously predicted that it would take at least three more years for AI to tackle problems from the US Math Olympiad. Despite the excitement, some experts remain cautious. Gary Marcus, a prominent AI critic, labeled the model's performance as "genuinely impressive" but raised questions about the training methods, the true extent of its general intelligence, practical applications for the broader public, and the costs associated with each problem. He also noted that the IMO has yet to independently verify these results.