\n<\/p>\n
\u201cHoisted by their own GPTards.\u201d<\/p>\n
That\u2019s how Meta\u2019s Chief AI Scientist Yann LeCun described<\/a> the blowback after OpenAI researchers did a victory lap over GPT-5\u2019s supposed math breakthroughs.<\/p>\n Google DeepMind CEO Demis Hassabis added<\/a>, \u201cthis is embarrassing.\u201d<\/p>\n The Decoder reports<\/a> that in a since-deleted tweet, OpenAI VP Kevin Weil declared that \u201cGPT-5 found solutions to 10 (!) previously unsolved Erd\u0151s problems and made progress on 11 others.\u201d (\u201cErd\u0151s problems\u201d are famous conjectures<\/a> posed by mathematician Paul Erd\u0151s.)<\/p>\n However, mathematician Thomas Bloom, who maintains the Erdos Problems website<\/a>, said Weil\u2019s post was \u201ca dramatic misrepresentation\u201d<\/a> \u2014 while these problems were indeed listed as \u201copen\u201d on Bloom\u2019s website, he said that only means, \u201cI personally am unaware of a paper which solves it.\u201d<\/p>\n In other words, it\u2019s not accurate to claim GPT-5 was able to solve previously unsolved problems. Instead, Bloom wrote, \u201cGPT-5 found references, which solved these problems, that I personally was unaware of.\u201d<\/p>\n Sebastien Bubeck, an OpenAI researcher who\u2019d also been touting GPT-5\u2019s accomplishments, then acknowledged<\/a> that \u201conly solutions in the literature were found,\u201d but he suggested this remains a real accomplishment: \u201cI know how hard it is to search the literature.\u201d<\/p>\n Techcrunch event<\/p>\n