OpenAI Faces Backlash Over False Erdős Math Breakthrough Claims

OpenAI faced significant criticism after claiming that its GPT-5 model solved several famous Erdős mathematical problems, which are longstanding conjectures believed to be true but not conclusively proven. Kevin Weil, OpenAI's VP, initially asserted that GPT-5 had solved ten such problems and made progress on eleven others, but mathematician Thomas Bloom, who curates the Erdős Problems website, clarified that these problems were labeled "open" simply because no known solution was documented on his site, not because they were unsolvable. Bloom emphasized that GPT-5 had only found references to existing solutions unknown to him, rather than discovering new proofs. OpenAI researcher Sébastien Bubeck acknowledged this distinction, highlighting the difficulty of literature search but admitting the model did not produce original mathematical breakthroughs. The episode drew sharp rebuke from AI leaders like Meta's Yann LeCun and Google DeepMind's Demis Hassabis, who described OpenAI's claims as misleading and embarrassing. This controversy raises important questions about AI's role in scientific discovery and the need for transparency and accuracy in presenting AI capabilities.