OpenAI and Google Achieve Gold in International Math Olympiad
The International Math Olympiad (IMO), a prestigious global competition for high school students, has witnessed an impressive performance from AI models developed by OpenAI and Google. Both organizations achieved gold medals by providing correct answers to a challenging set of mathematical problems.
The IMO is renowned for its rigorous standards, and this year, the competition featured six complex questions that needed to be solved over two days, with each participant allocated 4.5 hours of solving time. Both OpenAI and Google's DeepMind participated under the same conditions as human competitors, marking a significant milestone in the capabilities of artificial intelligence.
Last year, Google's AI models, including AlphaProof and AlphaGeometry 2, secured a silver medal but required human intervention to translate the problems into a format understandable by machines. In contrast, this year, both AI firms performed without such preprocessing, directly interpreting the mathematical challenges presented in natural language.
According to statements from the companies, both OpenAI and Google successfully answered five out of the six questions, a notable achievement considering the high difficulty level of the problems. The performance of these AI systems surpassed that of many human participants, demonstrating their advanced problem-solving capabilities.
Google's entry, a specialized version of its Gemini model equipped with Deep Think, achieved a score of 35 out of a possible 42 points. The model utilized a multifaceted approach, considering multiple solutions rather than following a single linear thought process. This innovative strategy allowed it to tackle the problems effectively. Additionally, Gemini had access to a curated database of mathematical problems and solutions, enhancing its performance.
OpenAI also reported a score of 35 points and confirmed that their approach adhered to the same standards applied to human competitors. This performance led to a noteworthy discussion within the AI community, particularly following a tweet from Alexander Wei, who shared a celebratory image featuring a strawberry--a nod to OpenAI's ongoing development of a product named Strawberry, which is designed to excel in mathematical applications.
However, the announcement of OpenAI's results sparked some controversy, as Demis Hassabis, CEO of DeepMind, expressed concern over the timing of OpenAI's disclosure before the official results for human participants had been released. This highlights the competitive nature of AI advancements and the scrutiny that accompanies such significant achievements in the field.
As AI continues to evolve, the implications of these results extend beyond academic competitions, influencing the ongoing development of artificial general intelligence (AGI) and its potential applications.