OpenAI is facing criticism after making claims about GPT-5’s mathematical abilities. The issue began when OpenAI Vice President Kevin Weil posted a celebratory tweet.
He claimed GPT-5 had “found solutions to 10 previously unsolved Erdős problems and made progress on 11 others.”
Erdős problems are well-known mathematical conjectures proposed by Hungarian mathematician Paul Erdős.
Many remain unresolved despite decades of work by experts. Solving even one is a significant accomplishment.
Weil’s statement spread rapidly online. It suggested GPT-5 had achieved something historic. However, the details told a different story.
Expert Opinion
Thomas Bloom, a mathematician who manages the Erdős Problems website, quickly responded. He described Weil’s tweet as “a dramatic misrepresentation.”
Bloom explained that when a problem is marked “open” on his website, it only means he has not found any published solution. It does not confirm that the problem remains unsolved globally.
In this case, GPT-5 had not solved new problems. Instead, it located references to existing solutions that Bloom had not previously listed.
He clarified, “GPT-5 found references, which solved these problems that I personally was unaware of.”
In short, GPT-5 identified solutions already present in the literature. It did not generate new mathematical insights.
Reactions
Meta’s Chief AI Scientist Yann LeCun used sharp language, saying the company had been “Hoisted by their own GPTards.”
His remark reflected frustration over what many saw as careless communication. Google DeepMind CEO Demis Hassabis also weighed in.
He called the situation “embarrassing.”
Clarification
As criticism mounted, OpenAI researcher Sebastien Bubeck stepped in. He acknowledged that GPT-5 had only found “solutions in the literature.”
However, Bubeck defended the model’s performance. He argued that searching through complex mathematical archives is difficult and time-consuming.
GPT-5’s ability to locate these references, he said, still demonstrated a valuable capability.
Communication and Accuracy
This incident highlights the gap between technological achievements and how they are communicated.
AI models are powerful, but their abilities must be described with precision. Overstated claims can mislead the public, fuel unrealistic expectations, and damage trust.
In this case, GPT-5 acted like a skilled research assistant. It did not solve unsolved problems but simply found existing solutions faster.
Therefore, the problem was not the model’s performance but how that performance was presented.
Implications
The backlash reflects broader concerns about transparency in the AI field. As AI tools become more advanced, the need for careful communication grows. Clear language matters. So does context.
Researchers and companies must ensure that their claims reflect reality. Exaggeration may create temporary excitement, but it can also lead to long-term credibility issues.
