“Awkward” mathematics for OpenAI | TechCrunch

“Uploaded by their GPTards.”

That’s how Yann LeCun, a senior AI scientist at Meta, described the negative reaction after OpenAI researchers scored a victory over supposed GPT-5 math hacks.

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

The Decoder reports that in a since-deleted tweet, OpenAI Vice President Kevin Weil announced that “GPT-5 has found solutions to ten (!) previously unsolved Erdős problems and has made progress on 11 more.” (“Erdos problems” are famous conjectures made by mathematician Paul Erdös).

However, mathematician Thomas Bloom, who runs the Erdos Problems website, said Weil’s post was a “huge misrepresentation” — while these problems were indeed listed as “open” on Bloom’s website, he said that only meant, “I personally am not aware of a paper that solves them.”

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

Sebastian Bobeck, an OpenAI researcher who was also touting GPT-5’s achievements, admitted that “solutions have only been found in the literature,” but noted that this was still a real achievement: “I know how difficult it is to search in the literature.”