In a growing wave of criticism, OpenAI backlash is mounting after bold claims that its latest model GPT‑5 achieved major mathematical breakthroughs were swiftly challenged by experts across the AI and academic communities.
The controversy began when a now-deleted tweet from Kevin Weil, Vice President at OpenAI, asserted that GPT-5 had “found solutions to 10 (!) previously unsolved Paul Erdős problems and made progress on 11 others”.
That announcement raised eyebrows immediately. Mathematician Thomas Bloom, who curates the renowned Erdos Problems website, called the claim a “dramatic misrepresentation” — explaining that “open” simply meant he personally had no knowledge of a solution, not that the problems remained unsolved.
Industry voices were equally scathing. Yann LeCun, Chief AI Scientist at Meta, captured the mood bluntly: “Hoisted by their own GPTards.” Meanwhile, Demis Hassabis, CEO of DeepMind, described the situation as “this is embarrassing.”
What Really Happened
What followed closer scrutiny revealed that GPT-5 did not independently solve those famed mathematical conjectures. Instead, it surfaced existing proofs in academic literature — ones that had not been indexed on certain databases. Bubeck, an OpenAI researcher who initially echoed the claims, later clarified:
“Only solutions in the literature were found… I know how hard it is to search the literature.”
Why the Backlash Matters
This episode underscores the fine line between claiming innovation and delivering it. It also highlights how major AI labs are under intense public scrutiny — and how missteps in language, hype, or framing can trigger severe reputation risk.
For OpenAI, the backlash raises questions about transparency, scientific communication, and the pressure on AI firms to deliver breakthrough results. The company has long promoted GPT-5 as a leap toward general intelligence — now critics argue that the hype may have outpaced the facts.
Key Takeaway
The rising OpenAI backlash shows that even the most advanced AI labs can stumble when claims aren’t backed by rigorous verification. Navigating AI’s promises may be harder than building its capabilities.