OpenAI’s “Embarrassing” Mathematics | Tech Crunch

“Holded by its own GPTard.”

Yann LeCun, chief AI scientist at Meta, described the fightback after OpenAI researchers took a victory lap over GPT-5’s supposed mathematical breakthrough.

Google DeepMind CEO Demis Hassabis added: “This is embarrassing.”

In a since-deleted tweet, OpenAI Vice President Kevin Weil declared, “GPT-5 has found solutions to 10 (!) previously unresolved Erdos problems and has made progress on 11 others,” The Decoder reported. (The “Erdos problem” is a famous conjecture posed by mathematician Paul Erdős.)

But mathematician Thomas Bloom, who maintains the Erdos problem website, said Weil’s post was a “dramatic misinformation”. Bloom’s website does list these problems as “unresolved,” but he said that only means “I personally am not aware of any papers that resolve them.”

In other words, it is not accurate to claim that GPT-5 can solve previously unresolved problems. Instead, Bloom wrote, “GPT-5 has found a reference to solving these problems that I personally was not aware of.”

OpenAI researcher Sebastien Bubeck, who was also touting GPT-5’s work, acknowledged that “we just found a solution in the literature,” but suggested that this was still a real work: “We know how difficult it is to search the literature.”