OpenAI’s ‘embarrassing’ math | TechCrunch

“Hoisted by their own GPTards.”

That’s according to Meta’s lead AI scientist Yann LeCun described the backlash after OpenAI researchers took a victory lap over GPT-5’s supposed mathematical breakthroughs.

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

The decoder reports that OpenAI VP Kevin Weil stated in a since-deleted tweet that “GPT-5 found solutions to 10 (!) previously unsolved Erdős issues and made progress on 11 others.” (“Erdős problems”. famous conjectures stated by mathematician Paul Erdős.)

However, mathematician Thomas Bloom, who persists the Erdos Problems websitesaid Weil’s post was “a dramatic misrepresentation” – while these issues were indeed listed as “open” on Bloom’s website, he said that just means: “I’m personally not aware of any article that resolves this.”

In other words, it is not correct to claim that GPT-5 could solve previously unsolved problems. Instead, Bloom wrote, “GPT-5 found references that resolved these issues, which I personally was not aware of.”

Sebastien Bubeck, an OpenAI researcher who had also touted GPT-5’s performance, recognized that ‘solutions were only found in literature’, but he suggested that this remains a real achievement: ‘I know how difficult it is to search in literature.’