AI答IMO难题坦承“不会”，OpenAI：这就是自我意识

Core Viewpoint - The article highlights a significant advancement in AI models, particularly OpenAI's model, which has demonstrated the ability to acknowledge its limitations rather than providing incorrect answers, marking a shift towards more reliable and self-aware AI systems [2][6][7]. Group 1: Model Performance and Characteristics - OpenAI's model received a zero score on the IMO's sixth problem but showcased "high IQ honesty" by admitting uncertainty when lacking evidence, which reduces hidden errors [2][3]. - The new generation of AI models is moving away from generating seemingly perfect but incorrect answers, learning instead to admit when they do not know the answer [6][7]. - This self-awareness allows the model to recognize its limitations in difficult problems, avoiding the generation of plausible yet incorrect solutions [17][16]. Group 2: Team Insights and Achievements - The core team behind the IMO gold medal consists of three researchers: Alex Wei, Sheryl Hsu, and Noam Brown, who shared insights during a conversation organized by Sequoia Capital [5][23]. - Alex Wei expressed that witnessing the model avoid hallucinations was a positive outcome, despite the disappointment of receiving a "I cannot answer" response after extensive computational effort [15]. - The team achieved their goal of winning the IMO gold medal in just two months, a remarkable feat considering initial skepticism about achieving this by 2025 [20][18]. Group 3: Team Backgrounds - Alex Wei holds degrees from Harvard University and a PhD from UC Berkeley, with prior experience at Google, Microsoft, and Meta before joining OpenAI in January 2024 [25]. - Sheryl Hsu graduated from Stanford University and was a researcher at the Stanford AI Lab before joining OpenAI in March 2025 [27]. - Noam Brown completed his undergraduate studies at Rutgers University and obtained his master's and PhD from Carnegie Mellon University, having previously worked at DeepMind and Meta before joining OpenAI in June 2023 [29].