'Embarrassing': OpenAI criticized for lying about solving math problem
What's the story
OpenAI's recent claims about its latest artificial intelligence model, GPT-5, have drawn criticism from the AI community. The controversy was sparked by a now-deleted tweet from Kevin Weil, OpenAI's VP. He claimed that "GPT-5 found solutions to 10 (!) previously unsolved Erdos problems and made progress on 11 others." These are famous conjectures posed by mathematician Paul Erdos. Google DeepMind CEO Demis Hassabis described OpenAI's actions as "this is embarrassing."
Clarification
'Dramatic misrepresentation'
The claim was contested by mathematician Thomas Bloom, who runs the Erdos Problems website. He said Weil's post was "a dramatic misrepresentation." Bloom clarified that while these problems were listed as "open" on his site, it doesn't mean they are unsolved. It only indicates that he is unaware of a paper that provides a solution to them. Meta's Chief AI scientist Yann LeCun described the furore around OpenAI's claims as "hoisted by their own GPTards."
Acknowledgment
How did OpenAI respond?
Sebastien Bubeck, an OpenAI researcher who had been promoting GPT-5's achievements, admitted that "only solutions in the literature were found." However, he maintained this is still a significant achievement. He said, "I know how hard it is to search the literature," implying that finding these references was no small feat.