OpenAI jumps gun on International Math Olympiad gold medal announcement
1 min read
Summary
OpenAI has announced one of its AI models has matched the performance of the top 9% of contestants in the annual International Mathematical Olympiad (IMO) and therefore achieved ‘gold medal’ performance.
The model solved six proof-based problems under the same time constraints imposed on human contestants (4.5 hours), but without calculator or internet access.
However, questions are being asked about the legitimacy of the claims due to the fact that OpenAI self-graded its results, and the fact that IMO organisers had requested an embargo on results being released until 28 July.
The company has said it will publish the model’s proofs and grading rubrics for public review, and claims that its achievement shows that ‘scalable, general-purpose methods can now outperform hand-tuned systems in tasks long seen as out of reach’.
