OpenAI’s ‘embarrassing’ math

“Hoisted by their very own GPTards.”

That’s how Meta’s Chief AI Scientist Yann LeCun described the blowback after OpenAI researchers did a victory lap over GPT-5’s supposed math breakthroughs.

Google DeepMind CEO Demis Hassabis added, “that is embarrassing.”

The Decoder studies that in a since-deleted tweet, OpenAI VP Kevin Weil declared that “GPT-5 discovered options to 10 (!) beforehand unsolved Erdős issues and made progress on 11 others.” (“Erdős issues” are well-known conjectures posed by mathematician Paul Erdős.)

Nonetheless, mathematician Thomas Bloom, who maintains the Erdos Issues web site, stated Weil’s put up was “a dramatic misrepresentation” — whereas these issues have been certainly listed as “open” on Bloom’s web site, he stated that solely means, “I personally am unaware of a paper which solves it.”

In different phrases, it’s not correct to assert GPT-5 was in a position to clear up beforehand unsolved issues. As an alternative, Bloom wrote, “GPT-5 discovered references, which solved these issues, that I personally was unaware of.”

Sebastien Bubeck, an OpenAI researcher who’d additionally been touting GPT-5’s accomplishments, then acknowledged that “solely options within the literature have been discovered,” however he steered this stays an actual accomplishment: “I understand how laborious it’s to go looking the literature.”