OpenAI Presents AI Model That Solves 4,000 Unresolved Math Problems

Math tests have long been a benchmark for evaluating AI model capabilities like ChatGPT and Claude. OpenAI now states that its models have become so proficient that traditional math exams are increasingly meaningless. Researchers therefore attempted something more challenging.
The team presented an undisclosed model with approximately 4,000 unresolved math problems. The results were surprising and concerning.
The AI generated 722 papers, which were grouped into 372 similar result sets. OpenAI revealed that each result underwent about three hours of reasoning computation on average.
Is AI becoming too fast and powerful? This is a more critical issue. Cutting-edge AI has begun to move beyond answering known questions toward potentially generating solutions for unknown problems.
This could significantly expand the scope of intellectual work researchers, engineers, and analysts can attempt.
However, not all results are valid. Many OpenAI papers include Lean proofs verifiable by computers, while others do not. OpenAI warns some unverified results may be problematic.
This creates a new bottleneck: humans may struggle to keep up with AI's research generation speed for verification purposes.
Can AI now make predictive analytics and investment decisions? Possibility exists, but finance is more complex in different ways.
Mathematical proofs can ultimately be proven right or wrong. Financial markets are complex and constantly changing.
Recent financial benchmarks still show cutting-edge AI struggling with sophisticated investment research. Limitations in prediction also emerge in market timing studies.
In the near term, deeper analysis may present greater opportunities than perfect predictions.
AIs capable of hours-long reasoning can review filings, earnings reports, macroeconomic data, and competitive scenarios simultaneously. They can test far more hypotheses than a single analyst.
This could be an important signal from OpenAI's experiment: AI is beginning to possess the capacity to produce vast amounts of serious analytical work. The next challenge is determining which results to trust.
Korean Source
This article is an English localization of a Korean-language crypto news report. Original headline: 오픈AI, 새 모델에 4000개 미해결 수학 문제