Summary of Key Points
The latest big news in the tech community is that OpenAI used an undisclosed, internally developed large model, leveraging the collaboration of 10,000 AI agents, to solve the "Navy-Stokes Equation Existence and Smoothness Problem" within just 88 hours. This problem has plagued the global mathematics community for 90 years and is one of the seven most challenging mathematical problems of the millennium. The computation involved the equivalent amount of text as in tens of millions of ordinary books. Subsequently, OpenAI used a widely recognized mathematical verification tool to complete the validation process in another 17 hours.
However, the core of this article is not to boast about the power of AI in mathematics, but to delve into three issues that even ordinary people cannot ignore:
1. OpenAI’s actions were not aimed at advancing mathematics; rather, it was a display of strength against its main competitor, Anthropic, to see whose model is more capable.
2. The immediate concern that "mathematicians will lose their jobs" is overly simplistic. The AI-driven problem-solving approach could potentially disrupt the mathematical discipline that has been developed over thousands of years.
3. This incident serves as a warning: when AI’s capabilities surpass human understanding, we may no longer be able to comprehend the underlying logic of AI-created aircraft, designed nuclear power plants, or newly developed drugs. Have we truly considered the safety of relying on such technologies? This arms race among large model companies has quietly crossed a critical threshold for which no one was prepared.
Simplified Explanation
1. How outrageous was OpenAI’s achievement?
- 10,000 concurrent AI agents worked simultaneously, each with a specific role—some brainstorming ideas, some checking calculations, and some identifying flaws, without any rest.
- 2.7 million messages were exchanged during their discussions, covering several centuries of internal research at a top mathematics institute.
- The output of 130 billion tokens is equivalent to the text content of tens of millions of ordinary books, representing an exhaustive exploration of countless problem-solving approaches that humans could never have conceived.
Ironically, OpenAI’s goal was not to conduct mathematical research; it was more about showing off its technology to its competitor.
2. The notion that "mathematicians will lose their jobs" is misplaced. The real loss is the loss of the joy and satisfaction that comes from exploration.
Mathematicians spend centuries solving problems not just to obtain the answers. For example, the solution to Fermat’s Last Theorem led to the invention of new mathematical tools and the establishment of entire fields, with values far greater than the theorem itself. The pleasure of deriving conclusions logically and experiencing the beauty of mathematics is intrinsic to their work.
3. The most frightening consequence is a gap in mathematical knowledge that humans cannot bridge.
If AI continues to increase its computing power, it may solve problems in days that took humans centuries. This could create a gap in knowledge that prevents future generations from building on what has been achieved.
4. "Can you trust an AI-designed aircraft?" is not a science fiction question; it’s a real issue we face soon.
The current civil aviation aircraft are designed by humans, and we understand every line of code and the logic behind every component. However, if AI designs aircraft, its logic may contain hidden flaws that we cannot detect. How can we be sure it is safe? If AI performs safety tests, we won’t understand its testing methods, and it could potentially hide critical flaws.
5. The arms race among large models has crossed a critical threshold.
The industry’s focus on scaling models (increasing computing power and parameters) has unexpectedly reached the ceiling of human intellectual capacity. Competitors will continue to invest heavily in this area, potentially leading to technologies beyond our current understanding. The legal, ethical, and safety frameworks we have are inadequate for such advancements.
In summary, the impact of AI on mathematics and technology is profound, and we must act quickly to establish new rules and regulations to ensure the safety and reliability of these technologies.