Mathematicians Test AI, Find It Nearing Genius Level

During a secret gathering in Berkeley, thirty leading mathematicians challenged a powerful AI chatbot, o4-mini, with extremely difficult, previously unsolved mathematical problems to test its reasoning abilities. To their surprise, this new type of large language model, trained on specialized data, demonstrated exceptional problem-solving skills, even quickly tackling complex, research-level questions that would take human experts much longer. Although the mathematicians ultimately found some questions the bot couldn't answer, they were stunned by its rapid progress and sophisticated approach, which included understanding related literature and attempting simpler versions of problems. The meeting highlighted the AI's remarkable capabilities and sparked discussions among attendees about the potential future role of mathematicians, possibly shifting towards generating novel questions for AI assistance.