OpenAI’s latest experimental model has achieved remarkable success in mathematics, securing gold medal-level performance at the prestigious International Math Olympiad (IMO). Alexander Wei, an OpenAI technical staff member, announced this achievement on X, highlighting the model’s ability to solve five out of six problems under human testing conditions. The IMO, founded in 1959, is renowned for its difficulty, featuring two days of intensely challenging questions. Some notable past winners include Grigori Perelman and Terence Tao. Despite Tao’s earlier skepticism about AI’s abilities, this model demonstrates a new level of endurance and problem-solving capability, according to Wei and colleague Noam Brown. They emphasize it marks a significant advancement in general-purpose reinforcement learning, distinct from specialized models. While OpenAI’s CEO Sam Altman noted this achievement signals substantial progress in AI, critics like Gary Marcus urge caution, questioning training methods and the practical implications of the results.
Source link
Share
Read more