Math tests have been the benchmark method for testing the capabilities of AI models like ChatGPT and Claude. OpenAI says its models are now performing so well that existing math tests are becoming less useful. So researchers tried something harder.
They gave an unreleased model around 4,000 unsolved math problems. The results were surprising and concerning.
The AI produced 722 manuscripts, grouped into 372 families of related results. OpenAI says each result used about three hours of reasoning compute on average.
“Some pretty exciting days ahead for the mathematical community!” said Stefano Gogioso, a member of BeInCrypto’s Future Tech and AI Experts Council
Is AI Becoming Too Powerful Too Fast?
This is the bigger story. Frontier AI is beginning to move beyond answering known questions and into generating possible answers to unknown ones.
That could sharply increase how much intellectual work a researcher, engineer or analyst can attempt.
But output is not truth. Many of OpenAI’s papers have computer-checkable Lean proofs. Others do not. OpenAI warns that some unverified results “could have issues.”
That creates a new bottleneck: humans may struggle to check research as quickly as AI can produce it.
Could an AI Now Perform Predictive Analysis and Make Investment Decisions?
Possibly, but finance is harder in a different way.
A mathematical proof can eventually be shown right or wrong. Markets are noisy and constantly changing.
Recent finance benchmarks still show frontier AI struggling with complex investment research, while studies of market timing find limited predictive advantage.
The near-term opportunity is deeper analysis rather than perfect prediction.
An AI capable of hours of sustained reasoning could examine filings, earnings calls, macro data and competing scenarios simultaneously, then test far more hypotheses than one analyst could.
That may be the bigger signal from OpenAI’s experiment. AI is becoming capable of producing serious analytical work at extraordinary volume. The next problem is deciding which of it deserves to be trusted.
The post OpenAI Gave Its New Model 4,000 Unsolved Math Problems appeared first on BeInCrypto.
