Mathematician Ernest Ryu on solving a 42-year-old problem in math with GPT-5 Pro (excitech.media)

Mathematician Ernest Ryu from UCLA recently used OpenAI's new GPT-5 Pro model to make significant advancements on a 42-year-old problem in optimization mathematics. After previously struggling with this issue, Ryu revisited it using ChatGPT, ultimately concluding that Nesterov’s Accelerated Gradient (NAG) method is indeed stable—a critical finding given NAG's wide-ranging applications in AI, image processing, and data analysis. The collaborative process, which Ryu likened to a maze search, allowed him to efficiently explore potential proofs and discard ineffective routes quickly, dramatically speeding up his research time. Ryu's experience underscores the transformative role AI technologies like ChatGPT can play in mathematical research, enhancing both speed and creativity in problem-solving. He noted that while approximately 80% of AI-generated suggestions were incorrect, some led him to new insights that might have otherwise gone unexplored. This event is a watershed moment in mathematics, echoing the transformative impact of AI in fields like chess. Despite these advancements, Ryu maintains that AI cannot yet independently generate groundbreaking mathematical discoveries, emphasizing that human intuition remains essential in defining new concepts. Nevertheless, he estimates that using AI tools has accelerated his research by up to tenfold, marking a promising shift in how mathematics is approached.