How ChatGPT-5 Helped Solve a 1992 Math Problem from Paul Erdős
Paul Erdős, one of the most prolific mathematicians of the 20th century, left behind a list of unsolved problems in 1992 that have challenged experts for decades. One of them just got cracked with a boost from artificial intelligence. A new report highlights how OpenAI’s ChatGPT-5, the latest large language model, assisted researchers in solving this long-standing puzzle, showing AI’s growing power in math.
The Erdős Problem and Its Challenge
Erdős posed several questions in 1992 about patterns in numbers and graphs, including conjectures that seemed out of reach. These problems sit at the heart of combinatorics, a field Erdős helped shape. For years, mathematicians have poked at them without full success.
In the recent breakthrough, ChatGPT-5 worked alongside human experts to tackle one of these. The AI didn’t solve it alone—it needed guidance—but it sped up the process by testing ideas and crunching calculations. As detailed in a Scientific American article covering a paper from OpenAI and collaborating scientists, this collaboration led to a genuine solution verified by mathematicians.
AI’s Step-by-Step Role in the Solution
The paper describes how ChatGPT-5 handled complex math steps that would take humans days. In this case, it navigated the problem’s logic, proposed paths, and even spotted connections in old literature. For instance, the AI once pulled a solution from a 1980s paper for another unsolved math question, overcoming language barriers like translating a 1960s German text.
- It produced results on wave behavior around black holes that matched known proofs.
- In nuclear fusion research, it built a model that cut coding time from days to minutes.
- For immune cell studies, it explained data in ways that aligned with lab results, suggesting new testable ideas.
These examples from the same OpenAI-led paper show ChatGPT-5 acting as a tireless assistant. Ryan Foley, an astrophysicist at UC Santa Cruz not involved in the work, called the math achievements impressive, noting that AI responds to human creativity but could accelerate discoveries.
What This Means for Math and Beyond
While the Erdős solution is modest, it points to bigger shifts. ChatGPT-5 also helped find rules on computer decision limits and patterns in branching diagrams. Prithviraj Ammanabrolu, a computer scientist at UC San Diego, told Scientific American that AI now mixes prior results into fresh insights faster than ever.
Still, humans stay essential. The AI sometimes gets facts wrong or invents references, so experts must check everything. Floor Broekgaarden, an astronomer at UC San Diego, emphasized that AI handles grunt work like data collation, freeing researchers for deeper thinking.
As AI models evolve, they could reshape how we approach old math challenges like Erdős’s conjectures. Who knows what a future version might unlock?
