OpenAI is facing sharp criticism from across the artificial intelligence and academic communities after claims that its GPT-5 model had “solved” ten of Paul Erdős’s famous unsolved mathematical problems were exposed as inaccurate.

The controversy erupted after OpenAI executives publicly suggested that the AI system had made groundbreaking progress in pure mathematics — a claim later debunked as a misunderstanding of what constituted “unsolved” problems.

From “Breakthrough” to Blunder

The incident began when OpenAI Vice President Kevin Weil posted on X (formerly Twitter) that GPT-5 had “found solutions to ten unsolved Erdős problems and made progress on eleven more.” The announcement was echoed by OpenAI researcher Sébastien Bubeck, who framed the achievement as evidence that “AI-driven scientific acceleration has begun.”

The implication was clear: GPT-5 had autonomously generated new mathematical proofs, marking a potential leap forward in machine reasoning.

Mathematician Exposes the Error

However, mathematician Thomas Bloom, curator of the website ErdosProblems.com, quickly refuted the claims, describing them as “a dramatic misrepresentation.” Bloom explained that the “open” status of certain problems on his website simply meant he was unaware of published solutions — not that they were genuinely unsolved.

“GPT-5 did not discover new proofs,” Bloom said. “It merely retrieved existing academic papers that I hadn’t yet catalogued.”

Following Bloom’s clarification, OpenAI deleted the original posts. Bubeck later admitted that GPT-5 had “only found solutions in the literature,” defending the work as “still meaningful” for its advanced research retrieval capabilities.

Industry Leaders React

The backlash was swift. Google DeepMind CEO Demis Hassabis called the episode “embarrassing,” while Meta’s Chief AI Scientist Yann LeCun accused OpenAI of “believing its own hype.”

The controversy, experts said, underscored the dangers of exaggerated marketing claims in the fast-moving AI industry. “We’re seeing hype outpace reality,” one AI ethics researcher commented. “This incident should be a lesson in scientific humility.”

Real Contributions and Real Limits

Despite the debacle, some prominent mathematicians defended GPT-5’s underlying capabilities. Fields Medalist Terence Tao noted that the model demonstrated genuine value as a research assistant, particularly in locating obscure or scattered academic literature.

“GPT-5 may not be inventing new mathematics,” Tao said, “but it’s already helping human researchers work more efficiently — that’s not trivial.”

Broader Implications

The episode highlights the increasingly competitive — and at times contentious — atmosphere surrounding AI research, as companies like OpenAI and Google DeepMind race to claim milestones in artificial reasoning and scientific discovery.

While GPT-5 remains one of the most advanced language models available, the incident has reignited concerns about transparency and accountability in how AI progress is communicated to the public.

As one academic put it, “This isn’t about what GPT-5 can or can’t do. It’s about making sure the world hears the truth about what AI is actually achieving — and what’s still far beyond its reach.” (ILKHA)