Put on your math hats for a moment and let's take a look at what the buzz was all about in mid-October. This is a perfect example of what's wrong with AI right now.
Bubek was encouraged that GPT-5 seemed to have somehow solved a series of mysteries known as Erdős problems.
Paul Erdős, one of the most prolific mathematicians of the 20th century. centuries, leaving behind hundreds of mysteries. To keep track of which ones were solved, Thomas Bloom, a mathematician at the University of Manchester, UK, created erdosproblems.comwhich lists more than 1,100 problems and notes that about 430 of them have solutions.
When Bubek celebrated the GPT-5 breakthrough, Bloom rushed call him. “This is a serious misrepresentation,” he wrote on X. Bloom explained that a problem is not necessarily unsolved if the website does not list a solution. It just means Bloom didn't know about it. There are millions of articles on mathematics, and no one has read them all. But GPT-5 probably does.
It turned out that instead of finding new solutions to 10 unsolved problems, GPT-5 scoured the Internet for 10 existing solutions that Bloom had not seen before. Oops!
There are two conclusions here. First, don't make sweeping statements about big breakthroughs via social media: less knee-jerk and more intuition.
Second, GPT-5's ability to find references to previous work that Bloom was unaware of is also remarkable. The hype overshadowed what should have been pretty cool in itself.
Mathematicians are very interested in using LLM to explore the vast amount of existing results, François Charton, a research scientist studying the use of LLM in mathematics at the artificial intelligence startup Axiom Math, told me when I talked to him about this Erdős mistake.
