OpenAI’s Math Claims Spark Controversy and Correction

Article Summary

OpenAI’s recent promotion of GPT-5’s prowess in mathematics ignited a significant controversy after a misinterpretation of the model's accomplishments. Initial reports, spearheaded by OpenAI VP Kevin Weil, claimed GPT-5 had solved 10 previously unsolved Erdős problems and made progress on 11 others. However, mathematician Thomas Bloom, who maintains the Erdos Problems website, swiftly pointed out that the ‘open’ status of these problems simply meant he himself hadn’t yet found a solution. Bloom clarified that GPT-5 had merely identified existing literature containing the solutions, representing a sophisticated search capability rather than genuine problem-solving. Even OpenAI researcher Sebastien Bubeck acknowledged the core issue – the findings were based on the retrieval of existing research, rather than novel insights. This episode underscores the difficulty in assessing AI’s understanding versus its ability to effectively access and synthesize information. The incident also reveals a critical gap in public perception regarding the limitations of large language models.

Key Points

• GPT-5’s claimed solution of Erdős problems was based on identifying existing solutions documented in the literature, not on a novel mathematical breakthrough.
• The ‘open’ status of the problems, as indicated by Thomas Bloom, meant that he himself had not yet found a solution.
• The incident highlights a fundamental distinction between AI’s search capabilities and true understanding of complex mathematical concepts.